v0.7.3 — a game that asks before it ends, and a results screen worth reading

Two issues off the tracker, and they are halves of one thing: the end of a game.
Neither ships on the 0.4.9 line — Jesse's call, that line may be complete and
these are not fixes people mid-playtest need.

EXTENDED PLAY (#11). The official result is settled at the original game length
and never changes: in a five-Day game extended to eight, the winner is whoever
led at the end of Day 5. Extending grants exactly one Day and the question is put
again at the end of it — solitaire the player decides alone, multiplayer it is
unanimous and one refusal ends it there. Only days-based endings offer it; a §3.4
collision breach is final, during an extended Day exactly as during the scheduled
game.

It could not be a client-side change. `check` refused every intent once `status`
left `active`; the server never loads a `finished` game back into memory; and a
save is `{ seed, config, history }` replayed through the engine, so a "continue"
the history does not record did not happen. Hence a fourth status,
`awaitingExtension`, and a `game.extend` intent. `config.days` never moves —
`extraDays` counts the borrowed Days and `official` freezes the outcome, the
standings and the statistics at the first ending.

THE RESULTS SCREEN (#16). `GAME OVER — revenueFloor` was `outcome.reason`, an
internal enum interpolated into the page at the one moment the game has the
player's whole attention. Every reason now has a sentence with the game's own
numbers in it. Around it: the result and winner, standings, the rules the game
was dealt under, a per-player breakdown, and the railroad — trains through the
Division and how many worked en route, loads made up and broken, passengers, cars
switched, trains destroyed. It shares the Day-end dialog's blocks rather than
reimplementing them, and stays reopenable so continuing does not cost you the
results.

Statistics are folded, not recorded: `state.tally` counts what the event stream
says happened, hooked at `applyIntent` and `advance` because `reduce` never sees
the phase driver's events — and those are the interesting ones. Nothing in the
rules reads it, and it rides the Frame, so multiplayer gets the same numbers as
solitaire from one implementation.

THREE BUGS FOUND IN TESTING, all of which would have shipped:

  - a saved game containing a vote could not be resumed (NO_ACTOR). A history is
    a flat Intent[] with no seat recorded; the replay derives who acted from the
    turn order, which cannot work for an intent every seat may send in any order.
    `game.extend` carries its voter, checked against the authenticated seat.
  - an all-bot game hung on the question for ever. `driveBots` loops on
    `currentActor`, null the moment the game stops, so it cannot cast a vote, and
    the bot-vote driver returned early with no humans to follow.
  - the balance harness became unbounded — `test/sim.test.ts` went from under a
    second to never finishing. `randomBot` took another Day about half the time,
    so every seeded game ran to playGame's 50,000-turn cap. Fixed in the driver,
    not in a policy, so it holds for bots not yet written.

All three have regression tests. 832 tests pass, against 793 before this change.

NOT BUILT, and a correction. #16's own comment said `trainStoodStill` "is emitted
per Stage, so a run of them is exactly the sat-on-a-siding streak". It is not:
reading advance.ts, it fires once per game and only for a train whose profile
sets `stopEarnsPoint` — the X18 Circus — with `stopPointClaimed` preventing a
second. The streak was built, rendered "1 Stage at (0,0)", and was taken out
again. There is no per-Stage "this train did not move" signal in the engine, so
"longest an engine sat on a siding" needs one first; TODO.md #36 records what it
would take, and the Circus set-up is reported instead. Badges remain the second
pass #16 asks for (TODO.md #33), and because the statistics are derived rather
than recorded, that pass can add any of them retroactively to games already
played and saved.

Extended play has not yet been played at a real table (TODO.md #35): the
multiplayer vote has only been driven through `session.intent`, never through two
browsers.

Closes #11
Closes #16

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EAgJSmeV8zrMh55Mj85ESb

CHANGELOG.md

@@ -19,6 +19,277 @@ page as `v0.1.0 · <sha> · <date>`, so what is deployed can always be identifie
 
 ---
 
+## 0.7.3 — 2026-08-29
+
+Two issues off the tracker, and they are halves of one thing: the end of a game. Gitea#11 stops the
+game being over when the timetable runs out, and Gitea#16 replaces the four lines that were shown
+there with a results screen worth reading. Neither ships on the 0.4.9 line — Jesse's call
+(2026-08-29): that line may be complete, and these are not fixes people mid-playtest need.
+
+### The game asks before it ends (Gitea#11)
+
+"When game ends allow players to continue playing if they wish… don't force end."
+
+The rule, decided with Jesse: **the official result is settled at the original game length and never
+changes.** In a five-Day game extended to eight, the winner is whoever led at the end of Day 5.
+Extending grants exactly **one** Day and the question is put again at the end of it — solitaire the
+player decides alone, multiplayer it is unanimous and one refusal ends it there and then. Only
+days-based endings offer it: a §3.4 collision breach is final, during an extended Day exactly as
+during the scheduled game, because a railroad declared unsafe does not carry on regardless.
+
+**It could not be a client-side change**, for three reasons that each rule out the others' fixes.
+`check` refused every intent once `status` left `active`; `server/index.ts` never loads a `finished`
+game back into memory; and a save is `{ seed, config, history }` replayed through the engine, so a
+"continue" the history does not record did not happen — the extended game would evaporate on the next
+reload, Undo or restart. So there is a fourth status, `awaitingExtension`, and the vote is an intent.
+
+`config.days` never moves. `extraDays` counts the borrowed Days, and `official` — the outcome, the
+standings and the statistics, frozen at the first ending — is what the results screen reports. That
+freeze is done by a wrapper around `advance` rather than inside `checkVictory`, so it happens *after*
+the last Day's events have been counted rather than before them.
+
+**Three bugs found while testing it, all of which would have shipped:**
+
+- **A saved game with a vote in it could not be resumed** — `NO_ACTOR`. A history is a flat
+  `Intent[]` with no seat written down; the replay derives who acted from the turn order. That works
+  for every other intent, `mainline.clearance` included, because there is exactly one seat it could
+  have been. Not here: every seat may vote in any order. `game.extend` therefore carries its voter,
+  uniquely, and the server checks it against the seat it authenticated.
+- **An all-bot game hung on the question for ever.** `driveBots` loops on `currentActor`, which is
+  null the moment the game stops, so it cannot cast a vote; the bot-vote driver returned early when
+  there were no humans to follow, and nothing ever asked. With nobody to follow, the bots' own answer
+  stands — no — and a bot-only game ends on the timetable it was dealt.
+- **The balance harness became unbounded**, which is how the third one announced itself:
+  `test/sim.test.ts` went from under a second to never finishing. `randomBot` picks uniformly among
+  its legal options, so once the two votes were among them it took another Day about half the time —
+  and because a table may go on granting Days indefinitely, every seeded game ran to `playGame`'s
+  50,000-turn cap instead of a couple of hundred. Giving `developerBot` a policy was not enough: the
+  guarantee has to hold for every policy, so **`playGame` itself declines**. A simulated game plays
+  the timetable it was dealt, whatever the bot would have voted.
+
+All three have regression tests.
+
+Bots never lead. They agree only once every human has agreed, which is the rule Jesse set: "bots will
+not disagree with the human. Humans get to vote first."
+
+### The results screen (Gitea#16)
+
+`GAME OVER — revenueFloor` was not a message. It was `outcome.reason`, an internal enum, interpolated
+straight into the page at the one moment the game has the player's whole attention. Every reason now
+has a sentence with the game's own numbers in it, and that fix alone answers the issue's "why did the
+game end?".
+
+Around it: the result and the winner, the standings, the rules the game was actually dealt under, a
+per-player breakdown, and **the railroad** — trains through the Division and how many of them did any
+switching en route, loads made up and broken, passengers worked, cars switched, trains destroyed and
+what they took with them. It shares the Day-end dialog's standings, target and collision blocks rather
+than reimplementing them, because two screens reporting the same game must not be able to disagree.
+It opens itself once per ending and leaves a button to reopen it, which is what stops Gitea#11's
+extended play costing you the results.
+
+**"I don't know if we keep statistics on…"** — nothing was being kept, and now `state.tally` is,
+folded from the event stream. Hooked at `applyIntent` and at `advance`, because `reduce` never sees
+the phase driver's events and those are exactly the interesting ones: `trainCompleted`,
+`trainsDestroyed`, `trainStoodStill`. The test that matters asserts **exactly once** — it plays real
+games, counts the event stream independently, and checks the tally against that count, so a fold
+hooked twice or nowhere fails whatever the seed.
+
+Nothing in the rules reads the tally, so adding a counter is always safe. It rides the `Frame`, so
+multiplayer gets the same numbers as solitaire from one implementation — and `test/redaction.test.ts`
+gained a case proving a tally never carries a card id, since that is a property of what `tally.ts`
+chooses to count and not something the types enforce.
+
+**"Longest an engine sat on a siding" is not in this pass, and the issue comment was wrong about
+why it could be.** That comment said `trainStoodStill` "is emitted per Stage, so a run of them is
+exactly the 'sat on a siding' streak you describe". It is not. Reading `advance.ts` rather than
+trusting it: the event fires only for a train whose profile sets `stopEarnsPoint` — the X18 Circus,
+the one card in the deck that pays for standing still — and `stopPointClaimed` makes sure it can
+never fire twice for the same train. The streak was built, shipped nothing but "1 Stage" against a
+raw grid coordinate, and has been taken out again. The engine has **no per-Stage "this train did not
+move" signal at all**, so this needs one before it can be answered; `TODO.md` #36 records that. What
+is reported instead is the Circus set-up itself, which is a real thing that happened.
+
+**Badges are not here.** The issue asks for them in a second pass after "a whole conversation
+brainstorming session", and that is where they belong. The tally keeps the raw material — the
+switching join for "switching master", the standing runs for "longest engine sat on a siding" — and
+because the statistics are *derived* rather than recorded, a second pass can add any of them
+retroactively to games already played and saved.
+
+### Also
+
+- **A recorded replay plays to its end.** `save-replay.ts` stopped where `currentActor` went null,
+  which since Gitea#11 is the extension question — so a newly recorded file would have replayed to a
+  question nobody answered rather than to a finished game. It declines, like every other bot driver.
+  The three files already published in `public/replays` predate the vote and stop on the question
+  when replayed; `test/harness.test.ts` accepts that as the end of their history, since it is.
+- The status line stops lying past the last Day. It read `N of TARGET · D Days left` with both halves
+  false; in an extended game it now reads the Day and how far beyond the timetable play has got.
+- `objectiveOf` paces against the timetable actually being played rather than `config.days`, which
+  otherwise reported "the last Day is over" through every extended Day.
+
+---
+
+## 0.7.2 — 2026-08-26
+
+Five issues off the tracker. Two are engine bugs a player hit at the board, two are the design
+catching up with rulings from RAR that the code had got wrong or never had, and one is the Division
+map being redrawn. The first four ship as **0.4.9h** on the 0.4.9 line; the map does not — it is a
+multiplayer redesign, and pushing one of those into a build people are mid-playtest on invalidates
+the feedback.
+
+### A 45° leg is part of the row (Gitea#17)
+
+**REPORTED:** "Cars were West to East Caboose, Loaded boxcar, Loaded boxcar, Loaded boxcar. After
+backing into that square cars were attached to the train Loaded boxcar, Loaded boxcar, Loaded boxcar,
+Caboose, Engine." The caboose came back next to the engine instead of at the far end, which also
+leaves the train badly made up under §8.2.
+
+The square was a `sw` CURVE and the train backed in through its SOUTH leg. `standing` runs west to
+east, and both places that walk that row asked the PORT which end they were at: `exploreMoves`
+reversed the row for an `'e'` entry and nothing else, and `cutTowards` answered "you meet nothing"
+for a north or south exit. Neither is a property of the port. A 45° leg leaves through the MIDDLE of
+its edge, so its end of the run is whichever end the arc does not reach — a `sw` curve's south leg is
+the row's EAST end and an `se` curve's south leg is its WEST end.
+
+`rowEndAt` answers it from the card, and `cutTowards` is narrowed from `Port` to `'e' | 'w'` so the
+type checker forces every caller to resolve rather than leaving a fourth to be found later.
+
+**THE SECOND HALF WAS LIVE TOO, and is a report we had failed to reproduce.** With `cutTowards`
+returning nothing for a leg, a crew standing on a curve pulled out through it and DROVE AWAY LEAVING
+ITS OWN CUT STANDING — against §A.4's mandatory coupling. That is "cars left behind when backing up
+over them", carried as NOT REPRODUCED since 2026-08-22; the earlier sweep had tried that case only
+with an east or west exit. The report's second sentence — "I can later drive right through them" —
+is still unexplained and still open.
+
+### The deck is the sheet (Gitea#14)
+
+`docs/Deck cards5.xlsx` is in the repo, and every count in the catalogue is now its count. Track is
+halved: 16 straights, 8+8 curves, 8+8 turnouts, sharp curves at zero — which is where they already
+were, and where sheet 5 independently puts them.
+
+**BOTH LONG-STANDING MULTIPLIERS COME OUT WITH IT.** The Q12 office doubling (14 → 7) and the Gap 12
+industry tripling (27 → 9) were measured against a deck holding 96 track cards, and halving the track
+turns the tripling backwards — the deck keeps dealing industries while the district stays too small
+to reach them. Measured over 300 bot games on identical seeds, 96 track with the multipliers / 48
+with them / 48 without: reefer cars set out by a crew **49 / 0 / 39**, mean revenue **−0.20 / +0.22 /
++0.27**. The middle column wipes out the reefer chain completely. The sheet's own density is the best
+of the three on both counts.
+
+Q12's own failure was re-measured rather than assumed: 43 of 100 games now never upgrade off a
+Whistle Post, up from 25 — but they average −0.2 revenue against +1.4 for games that do, where the
+gap used to be −6.0 against −0.4, and collisions fell from 26 per 100 games to 6. Staying at a
+Whistle Post is now common and survivable rather than rare and fatal.
+
+Everything sheet 5 does not list is dealt ZERO copies rather than deleted, so the design stays
+visible and the rules stay implemented: the Telegraph/Telephone/Radio dispatching ladder, Facing
+Point Locks, Flying Switch, Section House and Vandalism — all confirmed by Jesse as deliberate
+removals — plus Poling and the sharp curves, which were already there. Card for card, **84 rows agree
+with the sheet**; the only ones that do not are the ten Safety, Event, Inspection and Space-use cards
+it adds that are not built, held out until they are.
+
+### A rail may stop dead against its neighbour (Gitea#15)
+
+Filed as "track placements must connect", against a right-hand curve laid with its north leg against
+an Ice House and the turnout below pointing at its portless south edge. **RAR reversed it on review:**
+the placement is fine, and a stub like that has a use — a siding to park cars on. What he asked to
+confirm instead is that no train can traverse the gap.
+
+It could not, and cannot: `exploreMoves` gates every hop on `joins`, which tests both ports AND that
+two 45° legs lie on the same diagonal. That was already true and simply unpinned; `track.test.ts`
+now holds it against the reported geometry, including a check that the curve IS reachable from the
+side that joins, so the negative test cannot pass on a card that is merely unreachable.
+
+A per-edge placement check was written and then taken out, along with a matching guard on
+`checkTurnoutUpgrade`. Both are documented in place as deliberately absent, because this is exactly
+the rule someone will "fix" again. A Modifier is scenery (Jesse): a rail pointing at a building is
+fine.
+
+### Crossing a Mainline card is regions, not miles per hour (Gitea#3)
+
+**REPORTED:** a train taking two Stages to clear Double Track, which prints 60. RAR, on review:
+"Ignore speed signs, they are just graphics. Regions shown on cards indicate how many stages it takes
+to cross."
+
+Two recorded rulings are superseded together — Q1, that the printed 60/30 are crossing time, and Q2,
+that a Slow train adds a Stage to every card. Q2 is what produced the report. Cards carry a region
+count now and **where a train STARTS is what varies**: Plains 1, Double Track 1, Trestle 1, Curves 2,
+Tunnel 2, Heavy Grade 3. Fast/Slow is read on Hilly and nowhere else — and Hilly no longer reads the
+consist, which had a fast freight crossing slower than a slow passenger train. The grade modifiers
+move the start rather than cutting the clock: Helpers start an uphill train a region on, Brakeman a
+downhill one, Airbrakes another again. The Uncontrolled Siding and the Interchange print a back
+region that is not road: a train through starts past it, one arriving to find the siding occupied
+takes it and runs a region behind, and an Extra beginning its run at an Interchange starts there.
+
+**THREE THINGS WERE MISSING RATHER THAN WRONG.** With Plains now one region, a following train is in
+the same place as the train ahead the moment it enters — and nothing tested that, because the
+catch-up check sits inside `stagesRemaining > 1`, which a one-Stage crossing never reaches. That is
+the case ABS describes ("the train moving onto the card is instead held back") and it needed
+building. ABS was also holding SILENTLY on the Office and Division Point paths, so the one card whose
+purpose is preventing a wreck did its job invisibly. And `collide` never released the A/D track,
+which did not matter while every collision happened out on the road — a train destroyed as it LEAVES
+is still standing at the Office.
+
+The Uncontrolled Siding was marked "trains may pass", which skipped the collision test altogether and
+made the siding do nothing at all. It is `false` now, with the siding entry doing the work.
+
+Measured: on the Mainline cards of a 3-player Division, a fast train pays ~5.6 Stages against ~5.4
+before and a slow one ~6.0 against ~9.4. **Fast traffic is unchanged; slow traffic is about a third
+quicker**, and the gap across a Division collapses from roughly four Stages to under one — so Q2's
+recorded consequence, that every Slow train is still on the road when the next Day begins holding its
+Crew Tray, no longer holds and `players + 3` is due a re-examination.
+
+Two effects are recorded in `TODO.md` rather than acted on: freight share fell 8% → 5% and completed
+freight loads got scarcer, which runs against the obvious expectation and nobody knows why yet; and
+the bot stopped playing Red Flags — offered 4,212 times in 600 games, played 4.
+
+### The Division is one row (Gitea#18)
+
+**REPORTED:** "track design should not be horseshoe / square, but a single row… Division map should
+not show any office area detail."
+
+It was laid out around a table — one row for a single seat, two facing rows for two, a horseshoe for
+three, a square for four — on the reasoning that players sit around one. Three reports came out of
+that, and the one that decided it is that **east stopped being to the right**: a player's east could
+be drawn south, west or north depending which lane their district landed in, on a map whose whole job
+is saying which way a train is going. `WDP · ML · Office · … · ML · EDP`, left to right, and the
+buffer stops simply face outward at the two ends.
+
+An Office no longer expands into its Running Track either, so the map stops carrying every straight,
+turnout and Limits sign of every district — that is the Office map's job, and it draws them properly,
+with the rails. It also stops the Division map growing sideways every time somebody lays a card.
+
+The trains stay, in two registers: those holding an A/D track ON the rail, and crews switching in the
+district UNDER it — the distinction drawn as position rather than colour, because that is where those
+trains are. Each chip carries its number, a direction arrow and a car count, with the consist and the
+train's printed card on the tooltip. Every district cell is four chips two-by-two regardless of tier,
+because a Whistle Post with one A/D track can still hold four trains when three of them are crews,
+and sizing by occupancy is what "The Roster Pass" fixed. 1,580px wide at four players against 842
+before, drawn at 1:1 so it scrolls rather than shrinking — Jesse's call, "zooming and scrolling
+worked fine".
+
+This supersedes `TODO.md` item 24 outright and makes 19, 20, 21, 22, 25 and 26 irrelevant; item 27 is
+fixed.
+
+### Six test fixtures that pinned a seed and meant "a game like this"
+
+Every one of them broke on a rules change and none was about the rule that changed: the deck's SIZE
+moves the RNG stream, so any count change re-deals every fixture that names a seed.
+`mainline-cards` now searches for a Division holding a single-track card, `multiplayer` for a game
+that reaches Day 3, `web` clicks every play verb rather than assuming the first goes on the board and
+searches for a seed that builds a board, and the `sim` commodity samples were re-measured — a tank is
+first set out at game 216 now, unload Revenue at game 46. A seventh, `mainline-cards`' `hand()`
+helper, threw when a card was not in the deck, so dealing Flying Switch zero copies took five passing
+tests of an unchanged rule with it; it mints one now, which is the point of keeping a row at zero.
+
+**And one real bug, in a test.** `the game conserves Rolling Stock` went red claiming the engine had
+conjured three cars. The engine was right: Gap 2c sends a wreck's cabooses to the Division Yard and
+everything else to Classification, destroying nothing. The test subtracted the wreck's consist from
+the expected census — a rule the engine does not have — and the branch had never executed, because
+none of its six seeds had ever collided. The census is held flat unconditionally now, which is both
+the true invariant and stricter than what it replaced.
+
+---
+
 ## 0.7.1 — 2026-08-25
 
 Four issues off the Gitea tracker, all of them things a player saw at the board. Two are engine or

README.md

@@ -132,6 +132,25 @@ is the thing this machinery exists to prevent.
   roughly a third of the event types are never reduced at all. Anything that needs to rebuild a game
   replays the intents.
 - **Never call `Math.random()`.** One ambient random call silently breaks replay.
+- **A game ends by PAUSING, and the first ending is the real one.** Running out of Days, or closing
+  short of the combined Revenue floor, puts the game in `awaitingExtension` rather than `finished`:
+  the table is asked whether to play one more Day, unanimously, and asked again at the end of every
+  Day it grants. `state.official` is written at the first ending and never rewritten, so the winner
+  is always the one decided at `config.days` however long play carries on — `config.days` itself
+  never moves, and `state.extraDays` counts the borrowed ones. A §3.4 collision breach is the
+  exception and finishes outright, during an extended Day exactly as during the scheduled game.
+  Because a save is a replay, the vote is an intent (`game.extend`), and it is the one intent that
+  **carries its own player**: every seat may vote in any order, so a replay cannot derive who did.
+- **Statistics are folded, not recorded.** `state.tally` counts what the event stream says happened —
+  trains through the Division and how many of them did any switching, loads made up and broken — and
+  is hooked
+  at the two boundaries every event crosses exactly once, `applyIntent` and `advance`. It is not
+  hooked in `reduce`, which never sees the phase driver's events at all. Nothing in the rules reads
+  it, so adding a counter is always safe; it rides the `Frame`, so a multiplayer client gets the same
+  numbers as solitaire from one implementation. **What it cannot count is anything the events do not
+  say.** `trainStoodStill` fires once per game for the X18 Circus alone, so "the longest an engine sat
+  on a siding" has no signal behind it — see `TODO.md` #36 rather than assuming an event means what
+  its name suggests.
 - **A game is one of four TYPES, and a type is a set of defaults rather than a ruleset.** Co-op,
   Competitive, Cutthroat and Solitaire (`src/web/presets.ts`) each name an opening hand, an Extra
   rule, three revenue rates and the victory conditions; picking one fills the form, and changing any

TODO.md

@@ -50,8 +50,10 @@ Queued 2026-08-22, from a v0.4.9d gameplay-testing report (six bugs, forwarded b
     it just boarded**~~ — done in v0.4.9e / the release below.
 11. ~~**The Grocer's Warehouse ships and the Refinery receives**~~ — done in v0.4.9e / the release
     below: both are one-way again.
-12. **NOT REPRODUCED: cars left behind when backing up over them** — see Rules Questions below. The
-    one report of the six that is still open, and it needs a board from whoever filed it.
+12. **PARTLY REPRODUCED: cars left behind when backing up over them** — see Rules Questions below.
+    Half of it turned out to be the second half of Gitea#17 and is fixed (2026-08-26): a train
+    pulling out through a 45° leg left its own cut standing. The "I can later drive right through
+    them" half is still unexplained and still needs a board from whoever filed it.
 
 Queued 2026-08-22, from the v0.4.9e gameplay-testing report filed as Gitea issues.
 
@@ -127,30 +129,31 @@ Queued 2026-08-22, from a session looking at the screen rather than the rules. A
     means something different and useful: hide it now, bring it back at the end of the phase.
 18. **Give every phase a visible beat.** The automatic phases are not too fast — they are never
     drawn at all, because `pump` runs them all before the page renders once.
-19. **Turn the track art vertical on a Division card laid vertically** — a side lane currently reads
-    as stacked left-right segments rather than one continuous run.
-20. **Run the inter-row connector round the OUTSIDE**, draw it as rail rather than a plain line, and
-    give the corners an angled piece.
-21. **The Division Point captions overflow the map**, and the buffer stops point the wrong way once
-    the route wraps.
-22. **Fill the dead centre of the Division map with the common board** — timetable, yards, the
-    Department and Salvage decks. Then the map is what everyone shares and the right column is yours.
+19. ~~**Turn the track art vertical on a Division card laid vertically**~~ — **SUPERSEDED by
+    Gitea#18** (Jesse, 2026-08-26). There are no vertical lanes any more: the Division draws as a
+    single row, west to east.
+20. ~~**Run the inter-row connector round the OUTSIDE**~~ — **SUPERSEDED by Gitea#18.** No second
+    row, so nothing to connect.
+21. ~~**The Division Point captions overflow the map**, and the buffer stops point the wrong way once
+    the route wraps~~ — **SUPERSEDED by Gitea#18.** Nothing wraps; both ends face outward.
+22. ~~**Fill the dead centre of the Division map with the common board**~~ — **SUPERSEDED by
+    Gitea#18.** A row has no centre to fill.
 23. **History: newest at the top?** Plus the timestamps question from item 14, which lands here.
-24. **Put the viewer's own district at the BOTTOM of the Division map** and wrap the table around
-    them, so the screen sits you at the table. Needs no new data; the cost is that "west to east"
-    stops starting where you start reading.
+24. ~~**Put the viewer's own district at the BOTTOM of the Division map** and wrap the table around
+    them~~ — **DEFINITIVELY SUPERSEDED by Gitea#18** (Jesse's word, 2026-08-26). A single row and a
+    table wrapped around the viewer cannot both hold, and the row wins: being able to rely on east
+    meaning right is worth more than being seated at the table.
 
-Queued 2026-08-23, from Jesse playing the v0.7.0 build on StartOS. **All three are the same drawing
-pass as 19-21 and 24 above, and he asked for them to be discussed together rather than picked off:**
+Queued 2026-08-23, from Jesse playing the v0.7.0 build on StartOS. **All three were the same drawing
+pass as 19-21 and 24 above, and all three are answered by Gitea#18 rather than fixed:**
 
-25. **The Division map does not draw track geometry at all** — a turnout laid on the Running Track
-    looks exactly like the straight it replaced, because the only thing that changes is a caption.
-    Reasoning and the measurement in Display below.
-26. **CONFIRMED IN PLAY: the buffer stop points the wrong way** at two players — see item 21, now
-    reported from a real game rather than read off the code.
-27. **CONFIRMED IN PLAY: east is not always to the right.** As a train crosses the Division the route
-    wraps through the lanes, so a player's east can be drawn south, west or north. See item 20; it is
-    the same wrap that puts the buffer stop in the wrong place.
+25. ~~**The Division map does not draw track geometry at all**~~ — **SUPERSEDED by Gitea#18.** The
+    Division map stops drawing office-area detail altogether, so there is no Running Track on it to
+    draw geometry for. The geometry belongs to the Office map, which already draws it.
+26. ~~**CONFIRMED IN PLAY: the buffer stop points the wrong way** at two players~~ — **SUPERSEDED by
+    Gitea#18**, with item 21.
+27. ~~**CONFIRMED IN PLAY: east is not always to the right.**~~ — **FIXED OUTRIGHT by Gitea#18**, and
+    the reason it wins over item 24. One row means east is always to the right.
 
 28. **INVESTIGATE: move the game's settings off the top line and into a card of their own** — and
     show ALL of them, not the four that fit. Reasoning in Display below.
@@ -161,25 +164,76 @@ pass as 19-21 and 24 above, and he asked for them to be discussed together rathe
 Queued 2026-08-25, from releasing 0.7.1 / 0.4.9g. **Both are blocked on the two commits being made
 and pushed** — they were still uncommitted when the session ended.
 
-30. **Close the Gitea issues by hand, each with a comment naming the commit that fixed it.** Raised
-    by Jesse 2026-08-25. `Closes #<n>` is in the main commit message and will auto-close #2, #8, #9
-    and #10 when it reaches the default branch, but auto-closing leaves the issue with no record of
-    WHICH commit or WHICH release answered it — and #9 supersedes #6, which nothing on #6 will ever
-    say. So each issue wants a comment of its own: the commit SHA on both lines, the version each
-    shipped in (0.7.1 and 0.4.9g), and for #2 and #9 the ruling that decided it, since neither was
-    fixed the way the report implies. #6 wants a comment too, pointing at #9. The token and the API
-    calls are in the workspace's `AGENTS.local.md`.
+30. ~~**Close the Gitea issues by hand, each with a comment naming the commit that fixed it.**~~ —
+    done, and done again for v0.7.2 / v0.4.9h (2026-08-26). #2, #6, #8, #9 and #10 carry their
+    comments from the v0.7.1 release, including the pointer on #6 saying #9 superseded it. #3, #14,
+    #15, #17 and #18 now carry theirs, each naming `2ab25e3` and `e9683cc` and the version each
+    shipped in — and, where the fix was not what the report implied, the ruling that decided it:
+    #15 was REVERSED on review (the placement is legal; what was confirmed is that no train crosses
+    the gap), and #14's comment lists the ten unbuilt cards that are held out, so they do not vanish
+    along with the issue.
 
-31. **Bump the StartOS wrapper to 0.7.1.** `station-master-startos` is clean, on `master`, pinned to
-    `v0.7.0` and versioned `0.7.0:3`. Once v0.7.1 is tagged: move the submodule pin to the tag, set
-    `startos/versions/current.ts` to **`0.7.1:0`** (a new upstream tag resets the downstream digit —
-    `UPDATING.md`), and write release notes in all five locales. **No new version FILE and no
-    migration** — `current.ts`'s `migrations.up` is empty, so nothing is carried from 0.7.0, and a
-    0.7.0 save replays under 0.7.1 unchanged (its intents cannot contain a Timetabled discard,
-    because that was illegal when they were recorded). `instructions.md` and `README.md` both name
-    0.7.0 and both need the three user-visible changes: the Day-end dialog, the Timetabled discard
-    and its New Game checkbox, and the reasons a blocked platform now gives. Remember the branch
-    trap: the game pushes to `main`, the wrapper to `master`.
+    **Keep doing this.** Auto-closing leaves an issue with no record of which commit or which release
+    answered it. The token and the API calls are in the workspace's `AGENTS.local.md`.
+
+31. ~~**Bump the StartOS wrapper to 0.7.2.**~~ — done 2026-08-26 (`1bfea8d`). Submodule pinned to
+    `v0.7.2`, `current.ts` at `0.7.2:0`, release notes rewritten in all five locales, `README.md` and
+    `instructions.md` updated. No new version file and no migration. Verified: `npm run check` clean
+    and `make x86` packs as `v0.7.2:0`.
+
+    **This is the first release that does NOT carry games in progress**, and the release notes lead
+    with it in every locale. A save is a seed plus the moves played; the deck going from 206 cards to
+    121 means a card id recorded under 0.7.1 refers to a different card or to none, so a save stops
+    replaying at its first `card.play`. There is nothing to migrate — those moves were made against a
+    deck that no longer exists — and it fails safe: `src/server/index.ts` refuses to resume a save the
+    rules reject, names the move it stopped at, and leaves the file untouched, so an operator can put
+    0.7.1 back on to finish a game that matters.
+
+32. **Tell the 0.4.9 playtesters their saves are dead, before they find out.** The same deck change
+    shipped as v0.4.9h, so every save filed before it — including the ones attached to Gitea#15 and
+    #17 — stops replaying at its first `card.play`, and any game a tester has in progress will be
+    declined on restart. They fail safe and the files are kept, but nobody has been told. Worth a
+    line wherever the tester build is announced, and worth knowing when the next bug report arrives
+    with a save that will not load.
+
+Queued 2026-08-29, from building Gitea#11 and #16 (both shipped in v0.7.3, main only):
+
+33. **The second pass on the results screen — badges, and the brainstorm Gitea#16 asks for.** The
+    first pass is in and reports everything the Frame and the event tally know. What it deliberately
+    does not have is the interesting half: "maybe create badges for anything interesting that
+    happened… there should be a whole conversation brainstorming session on what are the things that
+    might be interesting for people to be aware of at the end of the game." The raw material is
+    already being kept — `tally.trainsCompletedWithWork` is the switching-master join, `longestStand`
+    is the engine that sat on a siding — and because the statistics are DERIVED from the event stream
+    rather than recorded, a second pass can add any of them retroactively to games already played and
+    saved. Needs Jesse and a conversation, not code, to start.
+
+34. **`replay.ts` prints a raw outcome enum, exactly as the results screen used to.** Its summary
+    line is `` `${o.result} — ${o.reason}` ``, which renders "loss — revenueFloor" — the same defect
+    Gitea#16 was filed about, in the dev-side replay viewer rather than the playable page. The
+    sentences now exist (`panels.ts`'s `reasonSentence`), but they are written against a `Frame` and
+    the replay recorder has a `GameState`, so it is a small refactor rather than a one-line swap. Not
+    done in v0.7.3 because nothing about it is player-facing and the change earns its own look.
+
+35. **Extended play has never been played at a real table.** v0.7.3 is tested — engine, server,
+    replay, an all-bot regression — and compiled and exercised headlessly, but nobody has sat down,
+    run a game off the end of its timetable and voted. The multiplayer vote in particular has only
+    been driven through `session.intent`, never through two browsers: what a second player sees while
+    waiting on a first, and whether "waiting on Carol" is legible when Carol has closed her laptop,
+    are both unanswered. Worth being the first thing the next play session does.
+
+36. **There is no per-Stage "this train did not move" signal, so "longest an engine sat on a siding"
+    cannot be answered.** Gitea#16 asks for it and the comment on that issue said `trainStoodStill`
+    would supply it, "emitted per Stage, so a run of them is exactly the streak you describe". That
+    is wrong, and was found only by reading `advance.ts` while building the tally: the event fires
+    for a train whose profile sets `stopEarnsPoint` — the X18 Circus and nothing else — and
+    `tray.stopPointClaimed` guarantees it fires at most once per train per game. A streak folded from
+    it reads "1 Stage" for ever, which is what v0.7.3 built and then removed.
+    **What it would take:** either a new event emitted per Stage per stationary tray (cheap to emit,
+    but it is a lot of events for a statistic nothing scores), or sampling live state on the Stage
+    boundary the way `stats.ts`'s funnel probe does — which the tally cannot do today, because it
+    folds a batch of events AFTER `advance` has already mutated past the moment they describe. Worth
+    settling with the badge pass (#33) rather than on its own, since that is the only consumer.
 
 ---
 
@@ -315,6 +369,18 @@ The replay viewer, the save format, and how a game gets shared.
 
 ## Bot Performance
 
+- [ ] **THE BOT WILL NOT PLAY RED FLAGS, and the reachability canary now says so.** Measured over
+      600 games (2026-08-26): `maneuver.redFlags` is OFFERED **4,212 times**, first in game 5 — the
+      rule is live and constantly available — and the bot PLAYS it **4 times**, first in game 252.
+      It is exempted by name in `sim.test.ts`'s anomaly test alongside `flyingSwitch`, and deleting
+      that line is what proves this fixed.
+
+      Two changes compounded to make it rarer without breaking anything: Gitea#14 took Red Flags from
+      5 copies to the sheet's 3, and Gitea#3 shortened most crossings to a single Stage, so the
+      window in which a train is STANDING on a Mainline card — the only place the card may be played
+      — is now usually one Stage wide.
+
+
 What the developer bot can and cannot yet do, measured. Every revenue figure below measured before
 v0.4.7 is low by roughly half a point — see the stub-industry entry — and the rebalance pass should
 not read that drop as a deck problem.
@@ -428,6 +494,30 @@ not read that drop as a deck problem.
 
 ## Play Balance
 
+- [ ] **FREIGHT GOT SCARCER WHEN THE MAINLINE WENT ONTO REGIONS, and nobody knows why yet.**
+      Gitea#3, measured 2026-08-26 across the same 100 games: freight share of gross fell **8% → 5%**,
+      and completed freight loads went from something a 40-game sample caught reliably to needing
+      200 — on `sim.test.ts`'s seeds, 40 games now yield 0 loads, 80 yield 3, 120 yield 10, 200
+      yield 21.
+
+      **It runs against the obvious expectation.** The change SPEEDS crossings up, so more trains
+      should reach more districts, not fewer. Revenue is flat (−0.2 against 0.0) and collisions are
+      unchanged at 0.1 a game, so nothing is obviously eating the traffic. Candidates worth checking:
+      trains now clear a district before a crew can work them; the entry-time collision rule
+      (below) destroying trains at the Office; or simply that faster turnover means fewer trains
+      standing where freight can be loaded.
+
+- [ ] **THE CREW TRAY COUNT IS DUE A RE-EXAMINATION, and this is the change that triggers it.**
+      `players + 3` was set when a Slow train took roughly twice as long to cross as a Fast one, and
+      Q2's recorded consequence was that "every Slow train is still on the road when the next Day
+      begins, holding its Crew Tray". Gitea#3 removed the Slow penalty from every card but Hilly.
+      Measured on the Mainline cards alone, a 3-player Division now costs a fast train ~5.6 Stages
+      and a slow one ~6.0, against ~5.4 and ~9.4 before: **fast traffic is unchanged, slow traffic is
+      about a third quicker**, and the gap across a whole Division collapses from roughly four Stages
+      to less than one. RAR's own closing note on the issue: "been worried about the time it takes to
+      cross the division. More thunking on this is needed."
+
+
 Numbers chosen to fix a measured problem rather than taken from the design. Revisit once the victory
 target is settled and freight carries its intended share; read no balance conclusion from a revenue
 number until the rules stop moving.
@@ -1178,6 +1268,14 @@ Deferred while planning the server; decisions and reasoning are in `docs/archite
 
 ## Display
 
+- [ ] **THE BOARD STILL DOES NOT SAY WHICH WAY A HEAVY GRADE CLIMBS.** RAR, twice: "grade should
+      tell you which way is up." `gradeUp` is dealt at setup and drives which of Helpers or
+      Brakeman/Airbrakes can ever pay, and Gitea#3 made it matter more — the modifiers now move a
+      train's STARTING REGION, so playing the wrong one is three Stages of climb instead of two. The
+      tooltip says it (`mainlineDescription`); the map does not. Untouched by Gitea#3, which was
+      about the rules rather than the drawing.
+
+
 What is on the screen and where. Split out of Other 2026-08-22; the rules are elsewhere.
 
 - [ ] **The log's start marker only works while the whole log fits.** Added 2026-08-23: a multiplayer
@@ -1529,17 +1627,31 @@ What is on the screen and where. Split out of Other 2026-08-22; the rules are el
 
 ## Rules Questions
 
-- [ ] **NOT REPRODUCED: "when I back up to collect standing cars and, further down the tracks, the
+- [ ] **PARTLY REPRODUCED: "when I back up to collect standing cars and, further down the tracks, the
       caboose, I get the caboose but the cars remain. I can later drive right through them."**
       Reported against v0.4.9d by a playtester (not Jesse, who forwarded it and could not add detail;
       his guess was that the cars were spotted at an industry).
 
+      **HALF OF IT IS NOW REPRODUCED AND FIXED (2026-08-26), as the second half of Gitea#17.** The
+      one case the earlier sweep did not try is a train pulling out through a 45° LEG rather than an
+      east or west port. `cutTowards` answered "you meet nothing" for a north or south exit, so a
+      crew standing on a curve drove away and left the cut beside it standing — the reported symptom
+      exactly, and against §A.4's mandatory coupling. `rowEndAt` (`track.ts`) fixes it, and
+      `cut-ordering.test.ts` pins it.
+
+      **WHAT IS STILL UNEXPLAINED is the second sentence — "I can later drive right through them."**
+      Nothing found so far accounts for that. A card is swept by `carsOn` whenever a train enters it,
+      whichever port it enters by, so a later pass over those cars picks them up. Until that half has
+      a board behind it this stays open: the fix above may be the whole report, or only the part that
+      happened to be reachable from the code.
+
       **What was tried, all of which works.** Cars on plain track on the way to the caboose; cars
       SPOTTED AT AN INDUSTRY on the way; the train's own cut standing on the square it is pulling out
-      of; a stale `standingWest` on the intermediate card; the industry locked by MEN AT WORK (which
-      correctly blocks the whole route rather than letting the crew past). Every one couples the lot.
-      The first three are pinned in `apply.test.ts` — "backing up over a cut to something beyond it
-      takes both" — so if the case is found later it is somewhere none of them cover.
+      of *through an east or west port*; a stale `standingWest` on the intermediate card; the industry
+      locked by MEN AT WORK (which correctly blocks the whole route rather than letting the crew
+      past). Every one couples the lot. The first three are pinned in `apply.test.ts` — "backing up
+      over a cut to something beyond it takes both" — so if the remaining case is found later it is
+      somewhere none of them cover.
 
       **Why it is hard to make happen.** Coupling is mandatory (§A.4) and `exploreMoves` accumulates
       what it meets card by card, so a route that reaches the caboose has already met everything
@@ -1570,6 +1682,50 @@ What is on the screen and where. Split out of Other 2026-08-22; the rules are el
 
 Doesn't fit the above.
 
+- [ ] **THE DECK IS `docs/Deck cards5.xlsx` EXACTLY, BAR TEN CARDS THAT ARE NOT BUILT.** Gitea#14,
+      2026-08-26. Card for card, **84 rows agree with the sheet** and the only ones that do not are
+      the ten it adds that we have never implemented — Cargo Theft, Civic Improvement, Civilian
+      angel, Delayed Clearance, Flares 2, Robbery, Service Delays, Shipper complaints, Strike,
+      Union Hall 2: **12 copies**, held out on Jesse's instruction until they are built. Gitea#12
+      partly specifies the Inspections among them.
+
+      What landed: track halved; the Q12 office doubling and the Gap 12 industry tripling both
+      removed; Interlocking 2→1, Water column 2→1, ABS Signals 2→1, Red Flags 5→3. **Everything
+      sheet 5 does not list is dealt 0 copies rather than deleted** — the Telegraph/Telephone/Radio
+      dispatching ladder, Facing Point Locks (Enhancement and Mainline both), Flying Switch, Section
+      House, Vandalism, all confirmed by Jesse as deliberate removals from the design, plus Poling
+      and the sharp curves which were already there. The rows and their rules stay, so the design
+      stays visible and each mechanic works the moment it is dealt again. Deck 206 → **121** dealt.
+
+      Measured, 100 games, developer bot: revenue per player **−0.2 → +0.4**, trains scheduled
+      1.3 → 1.6, cards played 16.8 → 13.0. Freight share fell 9% → 4%, and part of that is Flying
+      Switch going to zero — it was a freight mechanic. Worth a look if freight is meant to carry
+      more.
+
+      **NOT A DISCREPANCY, though it looks like one in a card-by-card diff:** Second Section is on
+      neither sheet and is not a drawn card here either. It is a New Train phase intent
+      (`newTrain.secondSection`), and the `copies: 1` on the `SECOND_SECTION` constant is vestigial —
+      nothing deals it. Worth removing that field so the next diff does not flag it again.
+
+      **The deck reads 40% track against the sheet's 31%**, and the whole of that gap is the ten
+      held-out cards concentrating everything else. Building them moves the ratio to the sheet's on
+      its own, which is why the share is held to a loose band in `setup.test.ts` rather than pinned.
+
+- [ ] **FIVE TEST FIXTURES PINNED A SEED AND MEANT "A GAME LIKE THIS".** All five broke on Gitea#14
+      and none of them was about card counts — the deck's SIZE moves the RNG stream, so changing it
+      re-deals every fixture that names a seed. Fixed in place: `mainline-cards` now searches for a
+      Division holding a single-track card, `multiplayer` for a game that reaches Day 3, `web` clicks
+      every play verb rather than assuming the first one goes on the board, and the two `sim`
+      commodity samples were re-measured (tank is first set out at game **216** now, unload Revenue
+      at game **46**). The `web` fix is on BOTH lines — it broke on `main` at the next count change,
+      exactly as predicted. `multiplayer`'s seed search is still playtest-only; port it when
+      convenient.
+
+      A sixth turned up when the dropped cards went to zero: `mainline-cards`' `hand()` helper threw
+      if the card it wanted was not in the deck, so zeroing Flying Switch took five passing tests of
+      an UNCHANGED rule down with it. It mints a card that is no longer dealt now — which is the
+      point of keeping a row at zero, and the same will hold for the ladder if anyone tests it.
+
 - [ ] **THE 0.4.9 PLAYTEST LINE IS BEHIND ON A RULES RULING, and that was checked rather than
       assumed.** Recorded 2026-08-23, when Jesse asked whether any of v0.7.0 needed porting to the
       `playtest` branch. Almost none of it does — that line has no lobby and no server, and the
@@ -1577,12 +1733,29 @@ Doesn't fit the above.
       `configWith` only ever varies the house rules the save already restores.
 
       **But one v0.5.0 ruling is a CODE difference the testers do not have.** §A.4, the Local's
-      coach: `main`'s `apply.ts` reads `if (dropRules.coachStaysOnStationTrack && cut.some(coach) &&
-      !atOffice)`, and `playtest`'s is the same line **without `&& !atOffice`** — so on that build a
-      coach still may not be set out at the Office. The other two questions the 0.4.9 README calls
-      open are documentation-only there (Poling is already at 0 copies; Heavy Grade behaves
-      identically and even deals the Mainline deck without replacement, so the correction applies
-      word for word).
+      coach — so on that build a coach still may not be set out at the Office.
+
+      **IT IS THREE SITES, NOT ONE.** This entry named only the first until 2026-08-25, when a full
+      branch diff found the other two. Porting just the `apply.ts` line would leave the build in a
+      WORSE state than either line is in today: the coach could be set out at the Office and the next
+      arriving train would then collide with it.
+
+      1. `src/engine/apply.ts` — `main` reads `if (dropRules.coachStaysOnStationTrack &&
+         cut.some(coach) && !atOffice)`; `playtest`'s is the same line **without `&& !atOffice`**.
+         This is the one that refuses the drop.
+      2. `src/engine/track.ts` — `canDropCarsAt(area, coord, count, coachesOnly)` takes a fourth
+         `coachesOnly` parameter on `main` and returns the Office square as droppable when it is set.
+         `playtest`'s signature has no such parameter and returns `false` for the Office outright.
+      3. `src/engine/advance.ts` — the §8.3 "cars fouling the Running Track" check. `main` reads
+         `officeCard.standing.some((c) => c.type !== 'coach')`, so a coach parked at the Office is
+         not a hazard to the next arrival; `playtest` reads `officeCard.standing.length > 0`, which
+         collides with anything standing there. **This one is behavioural and easy to miss** — it is
+         in a different file from the drop rules and reads as a collision fix rather than a coach one.
+
+      The other two questions the 0.4.9 README calls open are documentation-only there (Poling is
+      already at 0 copies; Heavy Grade behaves identically — both lines run the same
+      `rng.nextInt(2)`, re-verified 2026-08-25 — and it even deals the Mainline deck without
+      replacement, so the correction applies word for word).
 
       **Jesse's call, 2026-08-23: do not port it now.** A settled rules change is not a playtest bug
       fix, and pushing one into the build people are mid-playtest on would invalidate the feedback
@@ -1690,6 +1863,32 @@ Doesn't fit the above.
 
 ## Done, kept for the reasoning
 
+- **Gitea#15 — a rail may stop dead against its neighbour, and the rule is on MOVEMENT.**
+  Filed 2026-08-25 as "track placements must connect": a right-hand curve had been laid with its
+  north leg against an Ice House and the turnout below pointing at its portless south edge, and the
+  report called that illegal. **RAR reversed it on review (2026-08-26)** — the placement is fine, and
+  a stub like that has a use, as a siding to park cars on. What he asked to confirm instead is that
+  no train can traverse the gap.
+
+  It could not, and cannot: `exploreMoves` gates every hop on `joins`, which tests both ports AND
+  that two 45° legs lie on the same diagonal — never a bare pair of `hasPort` calls. That was already
+  true; `track.test.ts` now pins it against the reported geometry, including the check that the curve
+  IS reachable from the side that joins, so the negative test cannot pass on a card that is merely
+  unreachable.
+
+  **A per-edge placement check was written and then taken out**, along with the matching guard on
+  `checkTurnoutUpgrade`. Both are documented in place as deliberately absent, because this is exactly
+  the rule someone will "fix" again. `canPlaceAt` keeps only the weaker requirement it always had:
+  the piece must touch the network somewhere, which is what stops orphaned track.
+
+  **A Modifier is scenery** (Jesse, 2026-08-26): a rail pointing at a building is fine, so nothing
+  guards Modifier placement either. Measured before the ruling: 24 of 283 Modifiers across 200 bot
+  games sit where a neighbour's rail points at them, and that is simply legal.
+
+  The attached save is dead — Gitea#14 took the deck from 206 cards to 121, so its card ids no longer
+  exist and the replay stops at the first `card.play`. **Every save filed before that deck change is
+  in the same position**, including Gitea#17's. Reproduce from the geometry, not the file.
+
 - [x] **Put rolling stock back into circulation.** The Classification Yard was write-only — seven
       writers, no readers — so 37% of all rolling stock left the game by Day 5. Returning it at the
       Day boundary is **+2.32 ± 0.52 (t = 8.79)**, the largest single change measured on this bot,

docs/rules/implications.md

@@ -40,8 +40,9 @@ All nine answers are implemented, **170 tests passing**:
 
 | Answer | Implemented as |
 | --- | --- |
-| Q1 crossing time | `crossingStages()` — a 60 card takes 1 Stage, a 30 takes 2. Mainline nodes now carry a **terrain type** dealt at setup, and trains count down Stages instead of stepping through regions. The `Region` model is gone. |
-| Q2 Fast/Slow | Slow adds one Stage to every card. Hilly reads the consist (any coach = passenger). |
+| ~~Q1 crossing time~~ | **SUPERSEDED 2026-08-26 — see Q1a below.** Was: a 60 card takes 1 Stage, a 30 takes 2. |
+| ~~Q2 Fast/Slow~~ | **SUPERSEDED 2026-08-26 — see Q1a below.** Was: Slow adds one Stage to every card; Hilly reads the consist. |
+| Q1a crossing time | `crossingStages()` — a card costs one Stage per **region printed on it**, and where a train STARTS is what varies. Plains 1, Double Track 1, Trestle 1, Curves 2, Tunnel 2, Heavy Grade 3. The printed mph are scenery. Fast/Slow is read on Hilly and nowhere else. |
 | Q3 Expedite | An expedited train departs the Stage it arrives — it gets a second `moveTrain` in the same Mainline Phase, still subject to §8.1 clearance. |
 | Q4 Lockouts | `isLockedOut()` rejects the placement with `FACILITY_LOCKED`. |
 | Q5 Run-around | Nothing to do — reachability is geometric, so a built bypass already works. |
@@ -76,6 +77,39 @@ Crossing time never falls below one Stage — a train cannot cross in no time.
 there is nothing to build. A test asserts it remains TBD, to stop anyone "fixing" it by inventing
 an effect; a silent no-op would be worse than a rejection.
 
+**Q1a, answered by RAR 2026-08-26 (Gitea#3), and it replaces Q1 and Q2 together.**
+
+> "Ignore speed signs. They are just graphics. Regions shown on cards indicate how many stages it
+> takes to cross. Plains is 1. Double track is 1, tunnel is 2, curves is 2, heavy grade is 3 unless
+> you have help… Some cards say fast / slow. This is an indication that if on the train card, the
+> train is listed as fast or slow, that's starting position / how many stages it takes to traverse
+> the card. Fast / Slow does not apply to every card — just those that say fast / slow on them.
+> Currently this is only hilly."
+
+What this changes, against what was recorded before:
+
+- **The printed 60/30 mean nothing.** Q1 read them as crossing time; they are ambiance.
+- **Fast/Slow is not a global penalty.** Q2 added a Stage to every card for a Slow train, which is
+  what made a Slow train take two Stages to clear Double Track — the report that opened the issue.
+  It now applies on Hilly alone, where a fast train starts in the second region.
+- **Hilly no longer reads the consist.** RAR: "I notice that you are basing stages in mainline cards
+  off coach/non-coach. Actually, all trains are rated as FAST and SLOW."
+- **Heavy Grade is three regions, not two**, and the modifiers move the START rather than cutting the
+  clock: Helpers start an uphill train a region on, Brakeman a downhill one, Airbrakes another again.
+- **The Uncontrolled Siding and the Interchange print a back region** that is not part of the road. A
+  train running through starts past it; a train arriving to find the siding occupied takes it and
+  runs a region behind, which is what keeps the two apart, and an Extra beginning its run at an
+  Interchange starts there too.
+- **ABS holds a train off the card** rather than letting it collide, on any Mainline card.
+
+Measured consequence, replacing the one recorded under Q2: on a **3-player Division the Mainline
+cards themselves now cost a fast train ~5.6 Stages and a slow train ~6.0**, against ~5.4 and ~9.4
+before. Fast traffic is unchanged; **slow traffic is about a third quicker**, and the Fast/Slow gap
+across a whole Division collapses from roughly four Stages to less than one. The Q2 note that "every
+Slow train is still on the road when the next Day begins, holding its Crew Tray" no longer holds, so
+the `players + 3` tray count is due a re-examination — RAR's own closing worry: "been worried about
+the time it takes to cross the division. More thunking on this is needed."
+
 **Q11, answered from the source.** The Heavy Grade card prints **"(Up)"** and **"Player sets
 orientation"**, so which way it climbs is a property of the placed card, not a fixed compass
 direction. `DivisionNode.gradeUp` records the direction a train travels when **climbing**; a train

package.json

@@ -1,6 +1,6 @@
 {
   "name": "station-master",
-  "version": "0.7.1",
+  "version": "0.7.3",
   "private": true,
   "type": "module",
   "description": "Station Master — a railroad operations game",

src/engine/advance.ts

@@ -23,23 +23,24 @@ import {
   enhancementRule,
   crossingStages,
   trainProfile,
+  startRegion,
   MOVES_PER_LOCAL_OPS,
   MOVES_PER_LOCAL_OPS_NIGHT,
   STAGES_PER_DAY,
   STAGES_PER_SHIFT,
   houseRules,
   officeProfile,
-  REGIONS_PER_MAINLINE_CARD,
   mainlineProfile,
 } from './content.ts';
-import type { Direction } from './content.ts';
+import type { Direction, MainlineEntry, MainlineKind } from './content.ts';
 import type { GameEvent } from './events.ts';
 // `trainNeedingCars` lives in apply.ts beside `check`'s copy of the same question, so the phase and
 // the legality test cannot disagree about which train is being assembled.
 import { areaAtSeat, areaOf, trainNeedingCars } from './apply.ts';
 import { legalActions } from './legal.ts';
-import type { CrewTray, DivisionNode, GameState, PlayerIndex, RollingStock, SeatIndex, TrayId } from './state.ts';
-import { coordKey, freshTurns, playerAtSeat, playerLeftOf, pooled, subdivisions, totalRevenue, turnOf } from './state.ts';
+import type { CrewTray, DivisionNode, GameState, Outcome, PlayerIndex, RollingStock, SeatIndex, TrayId } from './state.ts';
+import { cloneTally, coordKey, freshTurns, isExtendable, playerAtSeat, playerLeftOf, pooled, subdivisions, totalRevenue, turnOf } from './state.ts';
+import { tallyEvent } from './tally.ts';
 
 export type AdvanceResult = {
   events: GameEvent[];
@@ -71,10 +72,58 @@ const step = (d: Direction): number => (d === 'east' ? 1 : -1);
 // advance
 // ---------------------------------------------------------------------------
 
+/**
+ * The phase driver, plus the two things that have to happen around EVERY batch of events it
+ * produces. `advanceInner` below is the driver itself, unchanged.
+ *
+ * ORDER IS THE WHOLE POINT of this wrapper, and it is the one subtle thing in Gitea#16.
+ * `checkVictory` runs deep inside the driver, so if the official result froze a copy of the Tally
+ * from in there it would freeze it BEFORE this batch's events had been counted — and the batch that
+ * ends a game is exactly the one carrying the last Day's work. So the Tally is folded first and the
+ * result frozen second, both out here where the whole batch is in hand.
+ *
+ * Safe because both endings `return` the moment they fire: no scoring event is emitted after a game
+ * has ended within a single batch, so "everything in this batch" and "everything up to the ending"
+ * are the same set of events. `test/tally.test.ts` pins that.
+ */
 export function advance(s: GameState): AdvanceResult {
+  const r = advanceInner(s);
+  for (const e of r.events) tallyEvent(s, e);
+  freezeOfficial(s);
+  return r;
+}
+
+/**
+ * THE OFFICIAL RESULT, written once (Gitea#11).
+ *
+ * "The winner is based upon the original game length" — so the first ending is the real one and
+ * every later evaluation is informational. Idempotent by construction: it does nothing once
+ * `official` is set, which is what stops an extended Day, or a §3.4 breach during one, from
+ * rewriting a recorded win.
+ */
+function freezeOfficial(s: GameState): void {
+  if (s.official !== null || s.outcome === null) return;
+  s.official = {
+    day: s.config.days,
+    outcome: { ...s.outcome },
+    revenues: s.players.map((p) => p.revenue),
+    collisionsTotal: s.collisionsTotal,
+    tally: cloneTally(s.tally),
+  };
+}
+
+function advanceInner(s: GameState): AdvanceResult {
   const events: GameEvent[] = [];
   if (s.status === 'finished') return { events, needsInput: false };
 
+  /**
+   * §3.3 (Gitea#11) — the timetable has run out and the table is being asked whether to play one
+   * more Day. Nothing runs itself while that question is open, so this is `needsInput` rather than
+   * an ending: `pump` stops here, the server keeps the game in memory, and the only intent the
+   * rules will take is `game.extend`.
+   */
+  if (s.status === 'awaitingExtension') return { events, needsInput: true };
+
   // The Superintendent's clearance ruling interrupts the Mainline Phase (§8.1).
   if (s.clock.pendingDecision !== null) return { events, needsInput: true };
 
@@ -451,6 +500,104 @@ function badlyMadeUp(tray: CrewTray): string | null {
   return caboose === rear ? null : 'not made up — the caboose must be at the rear of the train';
 }
 
+/**
+ * WHICH REGION OF A MAINLINE CARD A TRAIN IS STANDING IN (Gitea#3).
+ *
+ * A card is `regions` boxes wide and a train advances one per Stage, so what it has LEFT to run says
+ * where it is: enter with `regions` still to go and you are at the beginning; enter with one to go
+ * and you are in the last box.
+ *
+ * This used to be derived from a single global `REGIONS_PER_MAINLINE_CARD = 2`, with an entry term
+ * that put a one-Stage train in region 1 of a two-region card — a fast train did not traverse a fast
+ * card, it appeared at the far half of it. Cards carry their own region count now, so the position
+ * is simply the count minus what is left.
+ */
+export function regionOfTransit(card: MainlineKind, stagesRemaining: number): number {
+  const regions = mainlineProfile(card).regions;
+  return Math.min(regions - 1, Math.max(0, regions - stagesRemaining));
+}
+
+/** The entry a train would make onto this card, before occupancy is taken into account. */
+function entryFor(
+  node: Extract<DivisionNode, { kind: 'mainline' }>,
+  tray: CrewTray,
+  startsAtBack = false,
+): MainlineEntry {
+  const profile = trainProfile(tray.trainNumber ?? 0, tray.trainIsExtra);
+  return {
+    trainSpeed: profile?.speed ?? 'slow',
+    direction: tray.direction,
+    gradeUp: node.gradeUp ?? 'east',
+    modifiers: node.modifiers ?? [],
+    ...(startsAtBack ? { startsAtBack: true } : {}),
+  };
+}
+
+/**
+ * THE UNCONTROLLED SIDING RULE (Gitea#3): "if a train already exists when you arrive, you go in the
+ * second stage back — you are in the siding and are one behind the other train. This prevents a
+ * collision, since you are not in same exact location."
+ *
+ * So arriving at an occupied siding is not a collision and not a hold; it is a different, slower
+ * entry. Anywhere else this returns false and the ordinary start applies.
+ */
+function takesTheSiding(node: Extract<DivisionNode, { kind: 'mainline' }>): boolean {
+  return node.card === 'uncontrolledSiding' && node.transits.length > 0;
+}
+
+/**
+ * IS MOVING ONTO THIS CARD A COLLISION? (Gitea#3)
+ *
+ * A card can be ONE region wide — Plains, Double Track and Trestle all are — so a following train
+ * granted clearance arrives in the same region as the train ahead the moment it enters. There was no
+ * test for that at all: the catch-up check lives inside `stagesRemaining > 1`, which a one-Stage
+ * crossing never reaches, so entering behind another train on a Plains was silently free.
+ *
+ * ABS is the card that answers it, in RAR's words: "This is played on a mainline card to prevent
+ * collisions. If a collision would normally occur, the train moving onto the card is instead held
+ * back." Held, not waved through — it tries again next Stage.
+ *
+ * The Uncontrolled Siding never conflicts on entry, because `takesTheSiding` has already moved this
+ * train a region back; that is the whole point of the card.
+ */
+function entryConflict(
+  s: GameState,
+  node: Extract<DivisionNode, { kind: 'mainline' }>,
+  id: TrayId,
+  tray: CrewTray,
+  events: GameEvent[],
+  startsAtBack = false,
+): 'collided' | 'held' | null {
+  if (mainlineProfile(node.card).trainsMayPass) return null;
+  const start = startRegion(node.card, entryFor(node, tray, startsAtBack || takesTheSiding(node)));
+  const ahead = node.transits.find(
+    (t) =>
+      t.tray !== id &&
+      t.direction === tray.direction &&
+      regionOfTransit(node.card, t.stagesRemaining) === start,
+  );
+  if (!ahead) return null;
+
+  /**
+   * A BACKSTOP, not the main path. `evaluateClearance` already refuses to clear a train onto a card
+   * carrying ABS, so in the ordinary run of things nothing reaches here with signals up. It stays
+   * because the two rules answer to different questions — clearance looks at the whole Subdivision,
+   * this looks at one region — and a card that promises no rear-enders should not depend on the
+   * wider check happening to fire first.
+   */
+  if (node.absSignals) {
+    events.push({
+      type: 'trainHeld',
+      trainNumber: tray.trainNumber ?? 0,
+      reason: 'ABS Signals — held short of the train ahead',
+    });
+    return 'held';
+  }
+  // §10 — the Superintendent cleared it into an occupied region, so it is the Superintendent's fault.
+  collide(s, s.clock.superintendent, [id, ahead.tray], events, 'ran into the train ahead', 'the Mainline');
+  return 'collided';
+}
+
 /** Puts a train onto a Mainline card with its crossing time already computed. */
 function enterMainline(
   s: GameState,
@@ -458,17 +605,9 @@ function enterMainline(
   id: TrayId,
   tray: CrewTray,
   index: number,
+  startsAtBack = false,
 ): void {
-  const profile = trainProfile(tray.trainNumber ?? 0, tray.trainIsExtra);
-  const carriesPassengers = tray.consist.some((c) => c.type === 'coach');
-  const stages = crossingStages(
-    node.card,
-    profile?.speed ?? 'slow',
-    carriesPassengers,
-    node.modifiers ?? [],
-    tray.direction,
-    node.gradeUp ?? 'east',
-  );
+  const stages = crossingStages(node.card, entryFor(node, tray, startsAtBack || takesTheSiding(node)));
   node.transits.push({ tray: id, stagesRemaining: stages, stagesTotal: stages, direction: tray.direction });
   tray.position = { at: 'mainline', index };
   // It is running now, so it is no longer being assembled (state.ts). A train at a Division Point
@@ -577,6 +716,10 @@ function moveTrain(s: GameState, id: TrayId, tray: CrewTray, events: GameEvent[]
     if (clearance === 'blocked') return 'held';
     if (clearance === 'ask') return 'needsClearance';
 
+    const conflict = entryConflict(s, node, id, tray, events);
+    if (conflict === 'held') return 'held';
+    if (conflict === 'collided') return 'moved';
+
     enterMainline(s, node, id, tray, target);
     const dp = s.division.nodes[dpIndex];
     if (dp?.kind === 'divisionPoint') dp.holding = dp.holding.filter((t) => t !== id);
@@ -636,6 +779,11 @@ function moveTrain(s: GameState, id: TrayId, tray: CrewTray, events: GameEvent[]
       if (clearance === 'blocked') return 'held';
       if (clearance === 'ask') return 'needsClearance';
 
+      const conflict = entryConflict(s, node, id, tray, events);
+      if (conflict === 'held') return 'held';
+      // The wreck's A/D track is released by `collide` itself, which is why it has to be.
+      if (conflict === 'collided') return 'moved';
+
       enterMainline(s, node, id, tray, target);
       area.adOccupancy = area.adOccupancy.filter((t) => t !== id);
       events.push({
@@ -696,8 +844,19 @@ function moveTrain(s: GameState, id: TrayId, tray: CrewTray, events: GameEvent[]
       }
       if (clearance === 'ask') return 'needsClearance';
 
+      /**
+       * AN EXTRA PULLING OUT OF THE INTERCHANGE STARTS IN THE BACK REGION (Gitea#3) — "interchange
+       * has new extras show up in second region (like uncontrolled siding)", and earlier, "Plains is
+       * 1 stage for ALL trains. So are interlockings, with a second stage for incoming extras to hold
+       * at." A train running THROUGH the Interchange starts past that region and crosses in one
+       * Stage; one that began its run here has the holding region to clear first.
+       */
+      const conflict = entryConflict(s, node, id, tray, events, true);
+      if (conflict === 'held') return 'held';
+      if (conflict === 'collided') return 'moved';
+
       node.holding = node.holding.filter((t) => t !== id);
-      enterMainline(s, node, id, tray, index);
+      enterMainline(s, node, id, tray, index, true);
       events.push({
         type: 'trainHighballed',
         trainNumber: tray.trainNumber ?? 0,
@@ -729,21 +888,22 @@ function moveTrain(s: GameState, id: TrayId, tray: CrewTray, events: GameEvent[]
        *
        * ABS Signals does what it says instead: the follower stops SHORT of the collision and holds.
        */
-      const regionOf = (t: { stagesTotal: number; stagesRemaining: number }): number => {
-        const entry = REGIONS_PER_MAINLINE_CARD - t.stagesTotal;
-        const elapsed = t.stagesTotal - t.stagesRemaining;
-        return Math.min(REGIONS_PER_MAINLINE_CARD - 1, Math.max(0, entry + elapsed));
-      };
-      // NOT on a card that prints "trains may pass". Double Track and Uncontrolled Siding hold two
-      // trains because they HAVE two roads, so a train catching another there goes past it — that
-      // is what the card is for. Without this the mechanic fired 0.41 times a game while the bot
-      // never once granted clearance, which is the tell: those were all passing cards.
+      // NOT on a card that prints "trains may pass" — Double Track holds two trains because it HAS
+      // two roads, so a train catching another there goes past it. That is what the card is for.
+      //
+      // The Uncontrolled Siding used to be in that set and no longer is: it keeps two trains apart
+      // by putting the second one in the siding a region back (`takesTheSiding`), not by letting
+      // them share a place. Marking it "may pass" skipped this test entirely and made the siding do
+      // nothing at all.
       const mayPass = mainlineProfile(node.card).trainsMayPass;
-      const next = regionOf({ stagesTotal: transit.stagesTotal, stagesRemaining: transit.stagesRemaining - 1 });
+      const next = regionOfTransit(node.card, transit.stagesRemaining - 1);
       const ahead = mayPass
         ? undefined
         : node.transits.find(
-            (t) => t.tray !== id && t.direction === transit.direction && regionOf(t) === next,
+            (t) =>
+              t.tray !== id &&
+              t.direction === transit.direction &&
+              regionOfTransit(node.card, t.stagesRemaining) === next,
           );
 
       if (ahead) {
@@ -882,10 +1042,27 @@ function evaluateClearance(
     // they are not, and the protection belongs where the train it protects actually is.
     if (onNode?.kind === 'mainline' && (onNode.redFlagged ?? []).includes(other)) return 'blocked';
 
-    // ABS Signals — "trains on this card will not rear-end each other; they stop short of a
-    // collision". With signals in place a following train simply holds, and the Superintendent has
-    // no judgment call to make. This is the amendment to Gap 2's unconditional collisions.
-    if (onNode?.kind === 'mainline' && onNode.absSignals) return 'blocked';
+    /**
+     * ABS Signals — "this is played on a mainline card to prevent collisions. If a collision would
+     * normally occur, the train moving onto the card is instead held back" (RAR, Gitea#3).
+     *
+     * With signals in place a following train simply holds and the Superintendent has no judgment
+     * call to make, which is the amendment to Gap 2's unconditional collisions. It is caught HERE
+     * rather than at the entry itself, so the train never gets as far as the card.
+     *
+     * IT USED TO HOLD SILENTLY. A blocked clearance emits nothing on the Office and Division Point
+     * paths, so the one card whose entire purpose is to stop a wreck did its job invisibly: the
+     * train simply did not move, Stage after Stage, with nothing on screen saying why. The card is
+     * unplayable to reason about without this line.
+     */
+    if (onNode?.kind === 'mainline' && onNode.absSignals) {
+      events.push({
+        type: 'trainHeld',
+        trainNumber: tray.trainNumber ?? 0,
+        reason: 'ABS Signals — held short of the train ahead',
+      });
+      return 'blocked';
+    }
 
     // Same direction — the Superintendent must rule (§8.1, fourth condition).
     s.clock.pendingDecision = { train: id, occupiedBy: other };
@@ -1041,6 +1218,18 @@ function collide(
       if (n.holding) n.holding = n.holding.filter((t) => t !== id);
       if (n.redFlagged) n.redFlagged = n.redFlagged.filter((t) => t !== id);
     }
+    /**
+     * AND OFF THE A/D TRACK, for exactly the same reason as the transit above.
+     *
+     * It never mattered while every collision happened to a train already out on the road. Gitea#3
+     * adds one that can happen as a train LEAVES — a following train entering an occupied region —
+     * and that train is still standing at the Office when it dies. Without this its A/D track stays
+     * marked forever: the Office reads as permanently full, and every later arrival collides against
+     * a train that no longer exists.
+     */
+    for (const [, area] of s.officeAreas) {
+      area.adOccupancy = area.adOccupancy.filter((t) => t !== id);
+    }
   }
 
   if (lost.length > 0) {
@@ -1173,6 +1362,14 @@ function shiftChange(s: GameState, events: GameEvent[]): AdvanceResult {
       s.config.maxCollisionsTotal > 0 && s.collisionsTotal >= s.config.maxCollisionsTotal;
     if (perDayBreach || totalBreach) {
       s.status = 'finished';
+      /**
+       * NOT EXTENDABLE, AND IT DOES NOT REWRITE A RECORDED RESULT (Gitea#11).
+       *
+       * A breach during an EXTENDED Day ends play at once, exactly as it would during the regular
+       * game — but by then the official result already exists, and a railroad declared unsafe on
+       * Day 9 does not retract who won on Day 5. `freezeOfficial` is what keeps that true: it
+       * writes only when `official` is still null, so assigning `outcome` here is safe.
+       */
       s.outcome = { result: 'loss', winner: null, reason: 'collisionFloor' };
       return { events, needsInput: false };
     }
@@ -1216,28 +1413,55 @@ function rotateSeats(s: GameState, events: GameEvent[]): void {
 
 function checkVictory(s: GameState, _events: GameEvent[]): boolean {
   const daysElapsed = s.clock.day - 1;
-  if (daysElapsed < s.config.days) return false;
+  /**
+   * `extraDays` is Gitea#11. `config.days` is never touched by an extension — it is what the
+   * OFFICIAL result is decided at — so the Day the timetable currently runs to is the sum of the
+   * two. On the first ending they are equal, which is why `freezeOfficial` can record `config.days`
+   * as the official Day without asking anything further.
+   */
+  if (daysElapsed < s.config.days + s.extraDays) return false;
 
-  s.status = 'finished';
+  s.outcome = decideOutcome(s);
+  /**
+   * §3.3, EXTENDED PLAY — an ending the table may play past PAUSES rather than finishing.
+   *
+   * `freezeOfficial` (the `advance` wrapper) records the first of these as the official result, so
+   * by the time a second one is reached the winner is already settled and everything here is
+   * informational. The votes are cleared each time because the question is asked again at the end
+   * of every extended Day: agreeing once does not agree to the rest of the game.
+   */
+  if (isExtendable(s.outcome.reason)) {
+    s.status = 'awaitingExtension';
+    s.extensionVotes = s.players.map(() => null);
+  } else {
+    s.status = 'finished';
+  }
+  return true;
+}
 
+/**
+ * WHO WON, on the evidence as it stands right now.
+ *
+ * Split out of `checkVictory` for Gitea#11: it is asked once per ending, and an extended game has
+ * more than one. Unchanged in substance — the revenue floor, then co-op's shared achievement, then
+ * the highest Revenue — it simply no longer writes to the state it is reasoning about.
+ */
+function decideOutcome(s: GameState): Outcome {
   const combined = totalRevenue(s);
   if (s.config.minCombinedRevenue > 0 && combined < s.config.minCombinedRevenue) {
-    s.outcome = { result: 'loss', winner: null, reason: 'revenueFloor' };
-    return true;
+    return { result: 'loss', winner: null, reason: 'revenueFloor' };
   }
 
   if (s.config.mode === 'coop') {
-    s.outcome = { result: 'win', winner: null, reason: 'daysElapsed' };
-    return true;
+    return { result: 'win', winner: null, reason: 'daysElapsed' };
   }
 
   const best = Math.max(...s.players.map((p) => p.revenue));
-  s.outcome = {
+  return {
     result: 'win',
     winner: s.players.findIndex((p) => p.revenue === best),
     reason: 'daysElapsed',
   };
-  return true;
 }
 
 // ---------------------------------------------------------------------------

src/engine/apply.ts

@@ -52,6 +52,7 @@ import type {
   TrayId,
   TurnoutOrientation,
 } from './state.ts';
+import { tallyEvent } from './tally.ts';
 import { createRng } from './rng.ts';
 import {
   carsOn,
@@ -77,6 +78,7 @@ import {
   facilityVariants,
   opposite,
   reachableDestinations,
+  rowEndAt,
   variantsFor,
   withinLimits,
 } from './track.ts';
@@ -331,6 +333,15 @@ function checkTurnoutUpgrade(existing: TrackCard, proto: TrackCard): RejectionCo
 
   if (existing.standing.length > 0) return 'UPGRADE_OCCUPIED';
   if (existing.enhancements.length > 0) return 'UPGRADE_ENHANCED';
+  /**
+   * NOTHING IS ASKED ABOUT THE NEIGHBOURS, deliberately (Gitea#15).
+   *
+   * A turnout adds a 45° leg the card underneath did not have, and that leg may well point into an
+   * occupied square with nothing to meet it. That is legal: RAR ruled (2026-08-26) that a rail may
+   * stop dead against its neighbour, and an upgrade is no different from laying the piece there in
+   * the first place. What must hold either way is that no train can cross the gap, which is
+   * `exploreMoves`' business and is tested in `track.test.ts`.
+   */
   return null;
 }
 
@@ -759,6 +770,23 @@ export function passengerRefusal(
 // ---------------------------------------------------------------------------
 
 export function check(s: GameState, player: PlayerIndex, i: Intent): RejectionCode | null {
+  /**
+   * §3.3, EXTENDED PLAY (Gitea#11) — asked ABOVE the status guard, because the whole point of the
+   * vote is that it is the one thing the rules will take from a game that has stopped.
+   *
+   * Out of turn like `mainline.clearance` below, and unlike it open to every seat at once: it is a
+   * table decision rather than a ruling, so there is no actor to be.
+   */
+  if (i.type === 'game.extend') {
+    if (s.status !== 'awaitingExtension') return 'NOT_AWAITING_EXTENSION';
+    // The intent NAMES its voter so that a save can be replayed (`intents.ts`), which makes it a
+    // claim until it is checked against the seat the caller authenticated. One seat may not vote
+    // for another.
+    if (i.player !== player) return 'NOT_YOUR_TURN';
+    if (s.extensionVotes[player] !== null) return 'ALREADY_VOTED';
+    return null;
+  }
+
   if (s.status !== 'active') return 'WRONG_PHASE';
 
   // The clearance ruling is the one intent that arrives out of turn order: it interrupts the
@@ -962,7 +990,7 @@ export function check(s: GameState, player: PlayerIndex, i: Intent): RejectionCo
       if (!node || node.kind !== 'mainline') return 'NO_PLACEMENT';
       const on = node.modifiers ?? [];
       if (on.includes(rule.key)) return 'OPTION_ALREADY_CHOSEN';
-      if (rule.gradeOnly && mainlineProfile(node.card).speed.kind !== 'grade') return 'NOT_A_GRADE';
+      if (rule.gradeOnly && node.card !== 'heavyGrade') return 'NOT_A_GRADE';
       if (rule.requiresOnCard && !on.includes(rule.requiresOnCard)) return 'NOT_CONNECTED';
       // "Not while a train is on it" — realigning under a moving train is exactly the situation the
       // restriction exists to prevent. A train standing in the Interchange's yard counts: it is on
@@ -1470,7 +1498,7 @@ export function ownCutFor(s: GameState, player: PlayerIndex, trayId: TrayId, rev
   const facing = facingPort(s, trayId);
   const exit: Port = reverse ? reversePort(s, player, here, facing) : facing;
   const card = areaOf(s, player).grid.get(coordKey(here)) ?? emptyCard();
-  return cutTowards(card, carsOn(card), exit);
+  return cutTowards(card, carsOn(card), rowEndAt(card, exit));
 }
 
 /**
@@ -1522,6 +1550,27 @@ export function movesFor(
 
 function execute(s: GameState, player: PlayerIndex, i: Intent): GameEvent[] {
   switch (i.type) {
+    /**
+     * §3.3, EXTENDED PLAY (Gitea#11) — the vote, and what it settles.
+     *
+     * Decided HERE rather than in `reduce` because the answer depends on the votes as they stand
+     * BEFORE this one lands, and `execute` is the half that still sees that. Three outcomes:
+     *
+     *   - a refusal ends it immediately. Unanimity means one "no" is decisive, so nobody is made to
+     *     wait on a player who has already said no (Jesse's call, 2026-08-28);
+     *   - the last outstanding "yes" grants the Day — ONE Day, and the question is put again at the
+     *     end of it;
+     *   - anything else is just a vote recorded, and the table waits.
+     */
+    case 'game.extend': {
+      const vote: GameEvent = { type: 'extensionVoted', player, agree: i.agree };
+      if (!i.agree) return [vote, { type: 'playConcluded', declinedBy: player }];
+      const after = s.extensionVotes.map((v, p) => (p === player ? true : v));
+      return after.every((v) => v === true)
+        ? [vote, { type: 'dayExtended', day: s.config.days + s.extraDays + 1 }]
+        : [vote];
+    }
+
     case 'localOps.choose':
       return [{ type: 'localOpsOptionChosen', player, option: i.option }];
 
@@ -1588,14 +1637,16 @@ function execute(s: GameState, player: PlayerIndex, i: Intent): GameEvent[] {
          */
         const grid = areaOf(s, player).grid;
         const startCard = grid.get(coordKey(from)) ?? emptyCard();
-        const startCut = cutTowards(startCard, carsOn(startCard), exitPort);
+        const startCut = cutTowards(startCard, carsOn(startCard), rowEndAt(startCard, exitPort));
         const lifted = [
           ...(startCut.length > 0 ? [from] : []),
           ...dest.path.map((step) => step.coord),
           i.to,
         ].filter((c) => carsOn(grid.get(coordKey(c)) ?? emptyCard()).length > 0);
         const sides = standingSides(startCard, carsOn(startCard));
-        const stayed = exitPort === 'e' ? sides.west : exitPort === 'w' ? sides.east : [];
+        // The OTHER end's cut, which stays behind — so it is the other end of the row, not the
+        // other port. A 45° leg is an end of the row too (`rowEndAt`, Gitea#17).
+        const stayed = rowEndAt(startCard, exitPort) === 'e' ? sides.west : sides.east;
         // §A.3 — "engines also have couplers on the front end, so a train can pick cars up onto
         // its nose". Running forward the engine meets cars head-on and takes them in front; backing
         // up, they couple behind. Which end they land on is the whole point of a run-around: it
@@ -1930,6 +1981,30 @@ function findTimetableSlot(s: GameState, from: number): number | null {
 
 export function reduce(s: GameState, e: GameEvent): void {
   switch (e.type) {
+    // -- §3.3, extended play (Gitea#11)
+    case 'extensionVoted':
+      s.extensionVotes[e.player] = e.agree;
+      break;
+
+    /**
+     * One more Day, and the votes are wiped: agreeing once does not agree to the rest of the game.
+     *
+     * `config.days` is deliberately untouched. It is what the OFFICIAL result was decided at
+     * (`state.ts`'s `FinalReport`), so leaving it alone is what makes "the winner is decided at the
+     * original game length" a fact about the code rather than a comment on it. `outcome` is left
+     * alone too — it is the last evaluation, and it is what the results screen shows while the extra
+     * Day is played.
+     */
+    case 'dayExtended':
+      s.extraDays += 1;
+      s.extensionVotes = s.players.map(() => null);
+      s.status = 'active';
+      break;
+
+    case 'playConcluded':
+      s.status = 'finished';
+      break;
+
     case 'localOpsOptionChosen':
       turnOf(s, e.player).option = e.option;
       break;
@@ -3040,6 +3115,9 @@ export function applyIntent(s: GameState, player: PlayerIndex, i: Intent): Apply
 
   const events = execute(s, player, i);
   for (const e of events) reduce(s, e);
+  // Gitea#16 — the intent half of the fold; `advance` does the phase driver's half. See `tally.ts`
+  // for why it cannot simply live inside `reduce`.
+  for (const e of events) tallyEvent(s, e);
   return { ok: true, events };
 }
 

src/engine/content.ts

@@ -72,6 +72,26 @@ export type TrackProfile = {
  * from `docs/Deck cards2.xlsx`, a fixed document, and stay here as the audit trail for the
  * transcription — they are not claims about what the game deals today.
  *
+ * THE COUNTS BELOW NOW COME FROM `docs/Deck cards5.xlsx` (Gitea#14), which HALVES every track row
+ * against sheet 2: straight 32 → 16, each curve 16 → 8, each turnout 16 → 8. Track is the only
+ * section of that sheet whose numbers moved — every station, industry, modifier and train row is
+ * character-for-character what sheet 2 said — so this is the whole of the deck change it asks for.
+ *
+ * Sheet 5 also deals the sharp curves ZERO, which is where they already were: Jesse took them out
+ * for the reason below, and RAR arrived at the same number independently. Nothing to do, but worth
+ * recording that the two agree rather than leaving it looking like a coincidence.
+ *
+ * IT LANDS ON RAR'S OWN TARGETS, which is the check that matters — the top right of sheet 5 states
+ * the draw rates he is designing to. Against his denominators (start cards counted for track, only
+ * the non-track deck counted for trains): he wants track 48/167 = **28.7%** and trains 22/107 =
+ * **20.6%**; this deck gives 48/170 = **28.2%** and 22/110 = **20.0%**.
+ *
+ * THAT MATCH IS PARTLY A CANCELLATION, and whoever retunes next should know it. The engine holds
+ * ~25 more office and industry cards than the sheet (doubled and tripled, below) and is missing the
+ * ~33 Safety, Event, Inspection and Space-use cards sheet 5 lists, which Gitea#14 defers. The two
+ * errors are opposite and nearly equal today. Build the deferred categories and they stop
+ * cancelling, so the ratios have to be re-measured then rather than assumed to have held.
+ *
  * It matters well beyond bookkeeping. Track competes for the draw with industry, trains and
  * enhancements, so building a district is paid for in cards you did not draw instead — and the
  * hand-of-three is the real constraint on how fast a railroad grows.
@@ -86,9 +106,9 @@ export type TrackProfile = {
  * published replay still plays. Reordering to look tidy would silently re-deal every saved game.
  */
 export const TRACK_CARDS: readonly TrackProfile[] = [
-  { geometry: 'straight', hand: 'none', name: 'Straight track', copiesInDeck: 32, isOperationalRail: true, moveCost: 1 },
-  { geometry: 'curved', hand: 'right', name: 'Curved track (right)', copiesInDeck: 16, isOperationalRail: true, moveCost: 1 },
-  { geometry: 'curved', hand: 'left', name: 'Curved track (left)', copiesInDeck: 16, isOperationalRail: true, moveCost: 1 },
+  { geometry: 'straight', hand: 'none', name: 'Straight track', copiesInDeck: 16, isOperationalRail: true, moveCost: 1 },
+  { geometry: 'curved', hand: 'right', name: 'Curved track (right)', copiesInDeck: 8, isOperationalRail: true, moveCost: 1 },
+  { geometry: 'curved', hand: 'left', name: 'Curved track (left)', copiesInDeck: 8, isOperationalRail: true, moveCost: 1 },
   /**
    * SHARP CURVES ARE DEALT ZERO COPIES — Jesse's call, and the same treatment as Poling.
    *
@@ -103,8 +123,8 @@ export const TRACK_CARDS: readonly TrackProfile[] = [
    */
   { geometry: 'sharpCurved', hand: 'right', name: 'Sharp Curved Track (right)', copiesInDeck: 0, isOperationalRail: true, moveCost: 2 },
   { geometry: 'sharpCurved', hand: 'left', name: 'Sharp Curved Track (left)', copiesInDeck: 0, isOperationalRail: true, moveCost: 2 },
-  { geometry: 'turnout', hand: 'right', name: 'Turnout (right)', copiesInDeck: 16, isOperationalRail: false, moveCost: 1 },
-  { geometry: 'turnout', hand: 'left', name: 'Turnout (left)', copiesInDeck: 16, isOperationalRail: false, moveCost: 1 },
+  { geometry: 'turnout', hand: 'right', name: 'Turnout (right)', copiesInDeck: 8, isOperationalRail: false, moveCost: 1 },
+  { geometry: 'turnout', hand: 'left', name: 'Turnout (left)', copiesInDeck: 8, isOperationalRail: false, moveCost: 1 },
 ];
 
 /** Summed from `copiesInDeck` above, never written down — it moves whenever the deck is retuned. */
@@ -143,29 +163,39 @@ export type OfficeProfile = {
  * passenger modifier cards (Waiting Area, Restaurant, Hotel).
  */
 /**
- * Office cards. Every tier's `copiesInDeck` was **doubled** against the recovered design — Q12.
+ * Office cards, at `docs/Deck cards5.xlsx`'s counts exactly: Depot 4, Station 2, Terminal 1.
  *
- * Players always start at a Whistle Post, which has ONE A/D track, so a second arrival is an
- * automatic collision (§8.3, Gap 2a). Measured at the original density, 25 of 100 games never drew
- * a Depot and never escaped: they averaged **−6.0** revenue against **−0.4** for games that
- * upgraded at least once, and 25 of 26 collisions happened at Whistle Post.
+ * THE Q12 DOUBLING IS GONE (Gitea#14). Every tier used to be dealt at twice the sheet, to remove a
+ * 25% chance of an unwinnable opening deal: players always start at a Whistle Post, which has ONE
+ * A/D track, so a second arrival is an automatic collision (§8.3, Gap 2a), and measured at the
+ * sheet's density 25 of 100 games never drew a Depot and never escaped — averaging **−6.0** revenue
+ * against **−0.4** for games that upgraded at least once, with 25 of 26 collisions at a Whistle
+ * Post.
+ *
+ * That measurement was taken against a deck with 96 track cards in it, and the failure it describes
+ * does not survive the halving of track. RE-MEASURED at the sheet's counts, 100 games: **43 of 100**
+ * never upgrade off a Whistle Post, up from 25 — but they average **−0.2** revenue against **+1.4**
+ * for games that do upgrade, where the gap used to be −6.0 against −0.4. Collisions fell from 26 per
+ * 100 games to **6**, and only 3 of those are in games that never upgraded, against 25 of 26 before.
+ *
+ * So staying at a Whistle Post is now common and survivable rather than rare and fatal, which is the
+ * opposite of the shape Q12 was answering: with fewer trains reaching an Office, a single A/D track
+ * is seldom contested. The doubling was the blunt instrument its own note called it, and at the
+ * sheet's deck size it costs more than it buys — see `TRACK_CARDS` for the whole comparison and
+ * `INDUSTRY_PROFILES` for the other half of the same decision.
  *
  * Upgrades are strictly sequential (Gap 3b, no skipping), so Station and Terminal are rarer than
- * their raw counts imply — Terminal needs all three cards in order. Station and Terminal were
- * doubled with Depot to keep that ladder in proportion rather than making Depot a special case.
+ * their raw counts imply — Terminal needs all three cards in order.
  *
- * PROVISIONAL — re-evaluate. This was chosen to remove a 25% chance of an unwinnable opening deal,
- * not from the recovered design, and it is a blunt instrument: it lifts the whole office ladder and
- * dilutes every other category slightly. Revisit once the victory target is settled and freight is
- * carrying its intended share; the right answer may instead be fewer Terminals, a cheaper first
- * upgrade, or more A/D capacity at Whistle Post. The counts themselves are in the rows below, which
- * is the only place they should be read from.
+ * IF THE OPENING BITES AGAIN, the fix is not to re-double this. The note it replaces already listed
+ * the better options: fewer Terminals, a cheaper first upgrade, or more A/D capacity at a Whistle
+ * Post. Any of those answers the collision without diluting every other category to do it.
  */
 export const OFFICE_PROFILES: readonly OfficeProfile[] = [
   { tier: 'whistlePost', name: 'Whistle Post', isControlPoint: false, isPassengerFacility: false, adTracks: 1, porters: 0, passengerOut: 0, passengerIn: 0, copiesInDeck: 0 },
-  { tier: 'depot', name: 'Depot', isControlPoint: true, isPassengerFacility: true, adTracks: 2, porters: 1, passengerOut: 1, passengerIn: 1, copiesInDeck: 8 },
-  { tier: 'station', name: 'Station', isControlPoint: true, isPassengerFacility: true, adTracks: 3, porters: 2, passengerOut: 2, passengerIn: 2, copiesInDeck: 4 },
-  { tier: 'terminal', name: 'Terminal', isControlPoint: true, isPassengerFacility: true, adTracks: 4, porters: 3, passengerOut: 3, passengerIn: 3, copiesInDeck: 2 },
+  { tier: 'depot', name: 'Depot', isControlPoint: true, isPassengerFacility: true, adTracks: 2, porters: 1, passengerOut: 1, passengerIn: 1, copiesInDeck: 4 },
+  { tier: 'station', name: 'Station', isControlPoint: true, isPassengerFacility: true, adTracks: 3, porters: 2, passengerOut: 2, passengerIn: 2, copiesInDeck: 2 },
+  { tier: 'terminal', name: 'Terminal', isControlPoint: true, isPassengerFacility: true, adTracks: 4, porters: 3, passengerOut: 3, passengerIn: 3, copiesInDeck: 1 },
 ];
 
 export const OFFICE_ORDER: readonly OfficeTier[] = ['whistlePost', 'depot', 'station', 'terminal'];
@@ -202,15 +232,24 @@ export type IndustryProfile = {
 };
 
 /**
- * Industry density (Gap 12). The recovered sheet lists 9 industries in a 115-card deck; the
- * prototype ran 10 in 52. At the sheet's density a game saw 1.6 Freight Facilities, freight was 10%
- * of gross revenue, and `carsCoupled` fired 4 times per 100 games — the freight loop, which is the
- * point of the game, effectively never ran.
+ * Industry density, at `docs/Deck cards5.xlsx`'s counts exactly (Gitea#14).
  *
- * Each industry's `copies` is TRIPLED against the sheet, which restores roughly the prototype's
- * ratio while preserving the sheet's proportions exactly: the outbound/inbound balance and the
- * lockout structure are unchanged, because every kind scales by the same factor. The multiplier is
- * the decision; the resulting totals are in the rows below and move with every retune.
+ * THE GAP-12 TRIPLING IS GONE. The recovered sheet listed 9 industries in a 115-card deck and the
+ * prototype ran 10 in 52; at that density a game saw 1.6 Freight Facilities, freight was 10% of
+ * gross revenue, and `carsCoupled` fired 4 times per 100 games, so the freight loop effectively
+ * never ran. Tripling every kind restored roughly the prototype's ratio.
+ *
+ * ALL OF THAT WAS MEASURED AGAINST A DECK WITH 96 TRACK CARDS. Sheet 5 halves the track, and the
+ * tripling then works backwards: the deck keeps dealing industries while the district stays too
+ * small to reach them. Measured over 300 bot games on identical seeds — 96 track with the multiplier
+ * / 48 track with it / 48 track without — reefer cars set out by a crew went **49 / 0 / 39** and
+ * mean revenue **−0.20 / +0.22 / +0.27**. The middle column is the tripling meeting the halved
+ * deck: it wipes out the reefer chain completely. The sheet's own density is the best of the three
+ * on both counts.
+ *
+ * The sheet's proportions were always preserved by the multiplier, since every kind scaled by the
+ * same factor — so removing it changes the density and nothing else. The outbound/inbound balance
+ * and the lockout structure below are the sheet's, as they always were.
  */
 /**
  * LOCKOUTS, from the sheet's "Lockouts" column verbatim:
@@ -233,8 +272,8 @@ export type IndustryProfile = {
  * enforced for every kind in `isLockedOut`, not repeated in each row here.
  */
 export const INDUSTRY_PROFILES: readonly IndustryProfile[] = [
-  { kind: 'freightHouse', name: 'Freight House', carTypes: ['boxcar'], flow: 'both', baseOut: 1, baseIn: 1, baseLoaders: 1, lockouts: ['grocersWarehouse'], copies: 6 },
-  { kind: 'mineTipple', name: 'Mine Tipple', carTypes: ['hopper'], flow: 'outbound', baseOut: 1, baseIn: 0, baseLoaders: 1, lockouts: ['powerPlant'], copies: 6 },
+  { kind: 'freightHouse', name: 'Freight House', carTypes: ['boxcar'], flow: 'both', baseOut: 1, baseIn: 1, baseLoaders: 1, lockouts: ['grocersWarehouse'], copies: 2 },
+  { kind: 'mineTipple', name: 'Mine Tipple', carTypes: ['hopper'], flow: 'outbound', baseOut: 1, baseIn: 0, baseLoaders: 1, lockouts: ['powerPlant'], copies: 2 },
   /**
    * OUTBOUND ONLY. A Refinery ships oil out and receives nothing; reported from playtesting and
    * confirmed by Jesse (v0.4.9e): "only ships out tanks, does not receive anything".
@@ -252,9 +291,9 @@ export const INDUSTRY_PROFILES: readonly IndustryProfile[] = [
    * the game with no way to raise the direction it is supposed to use half its capacity on.
    * `StationMaster-Home-Deck-v0.4.5.md` prints it "Outbound, 1 out / 0 in".
    */
-  { kind: 'refinery', name: 'Refinery', carTypes: ['tank'], flow: 'outbound', baseOut: 1, baseIn: 0, baseLoaders: 1, lockouts: ['powerPlant'], copies: 3 },
-  { kind: 'powerPlant', name: 'Power Plant', carTypes: ['hopper', 'tank'], flow: 'inbound', baseOut: 0, baseIn: 1, baseLoaders: 1, lockouts: ['mineTipple', 'refinery'], copies: 6 },
-  { kind: 'packingSheds', name: 'Packing Sheds', carTypes: ['reefer'], flow: 'outbound', baseOut: 1, baseIn: 0, baseLoaders: 1, lockouts: ['grocersWarehouse'], copies: 3 },
+  { kind: 'refinery', name: 'Refinery', carTypes: ['tank'], flow: 'outbound', baseOut: 1, baseIn: 0, baseLoaders: 1, lockouts: ['powerPlant'], copies: 1 },
+  { kind: 'powerPlant', name: 'Power Plant', carTypes: ['hopper', 'tank'], flow: 'inbound', baseOut: 0, baseIn: 1, baseLoaders: 1, lockouts: ['mineTipple', 'refinery'], copies: 2 },
+  { kind: 'packingSheds', name: 'Packing Sheds', carTypes: ['reefer'], flow: 'outbound', baseOut: 1, baseIn: 0, baseLoaders: 1, lockouts: ['grocersWarehouse'], copies: 1 },
   /**
    * INBOUND ONLY — the mirror of the Refinery above, and the same correction. Reported from
    * playtesting and confirmed by Jesse (v0.4.9e): "Grocer's Warehouse should be receive only, does
@@ -267,7 +306,7 @@ export const INDUSTRY_PROFILES: readonly IndustryProfile[] = [
    * other example. The Truck Dock (+1 inbound) and Local Small Groceries (+1 Laborer) are the two
    * that do work here.
    */
-  { kind: 'grocersWarehouse', name: "Grocer's Warehouse", carTypes: ['boxcar', 'reefer'], flow: 'inbound', baseOut: 0, baseIn: 1, baseLoaders: 1, lockouts: ['packingSheds', 'freightHouse'], copies: 3 },
+  { kind: 'grocersWarehouse', name: "Grocer's Warehouse", carTypes: ['boxcar', 'reefer'], flow: 'inbound', baseOut: 0, baseIn: 1, baseLoaders: 1, lockouts: ['packingSheds', 'freightHouse'], copies: 1 },
 ];
 
 /** Legacy alias; the engine still reads FREIGHT_PROFILES in places. */
@@ -504,23 +543,35 @@ export type MainlineKind =
   | 'uncontrolledSiding' | 'tunnel' | 'trestle' | 'interchange';
 
 /**
- * Speed as printed. `60` and `30` appear on the cards; Hilly prints P60/F30, and Heavy Grade prints
- * "G" — no number at all, plus "Player sets orientation", **which the game deliberately does not do**
- * (see `gradeReduction` below, and implications.md §10 Q11).
+ * THE PRINTED SPEEDS ARE GRAPHICS. RAR, 2026-08-26 (Gitea#3): "please ignore the speed signs I put
+ * on the cards — those are nothing but scene-setting graphics that mimic the speed you are
+ * travelling. It's just ambiance, nothing more."
  *
- * WHAT THESE NUMBERS MEAN IS NOT YET SETTLED — see implications.md §10 Q2. Transcribed as data so
- * the answer can be applied without re-reading the cards.
+ * They used to decide everything: a `MainlineSpeed` of 60 meant one Stage and a 30 meant two, plus
+ * one more for a Slow train. Both rules are gone. **What crosses a card is REGIONS** — the boxes
+ * printed on it — one per Stage, and where a train STARTS decides how many it has left to run.
  */
-export type MainlineSpeed =
-  | { kind: 'uniform'; value: number }
-  | { kind: 'byTrainType'; passenger: number; freight: number }
-  | { kind: 'grade' };
-
 export type MainlineProfile = {
   kind: MainlineKind;
   name: string;
-  speed: MainlineSpeed;
-  /** Double Track and Uncontrolled Siding: "Trains may pass". */
+  /** Regions printed on the card. A train advances one per Stage, so a full run costs `regions`. */
+  regions: number;
+  /**
+   * The region an ordinary train enters at. Zero on nearly everything — but the Uncontrolled Siding
+   * and the Interchange print a back region that is a siding or a holding spur rather than part of
+   * the road, so a train running straight through starts past it and crosses in one Stage.
+   */
+  defaultStart: number;
+  /**
+   * Cards that print a FAST and a SLOW start, and the region each begins at. "Some cards say fast /
+   * slow. This is an indication that if on the train card, the train is listed as fast or slow,
+   * that's starting position / how many stages it takes to traverse the card. Fast / Slow does not
+   * apply to every card — just those that say fast / slow on them. Currently this is only hilly."
+   *
+   * So the train's rating is read HERE and nowhere else. It used to add a Stage to every card.
+   */
+  speedStarts?: { fast: number; slow: number };
+  /** Double Track: "Trains may pass". */
   trainsMayPass: boolean;
   /** Interchange: "Sort cars in new order". */
   sortsCars: boolean;
@@ -529,30 +580,56 @@ export type MainlineProfile = {
 };
 
 export const MAINLINE_PROFILES: readonly MainlineProfile[] = [
-  { kind: 'plains', name: 'Plains', speed: { kind: 'uniform', value: 60 }, trainsMayPass: false, sortsCars: false, entryPoints: ['start'] },
-  { kind: 'curves', name: 'Curves', speed: { kind: 'uniform', value: 30 }, trainsMayPass: false, sortsCars: false, entryPoints: ['start'] },
-  { kind: 'hilly', name: 'Hilly', speed: { kind: 'byTrainType', passenger: 60, freight: 30 }, trainsMayPass: false, sortsCars: false, entryPoints: ['passenger', 'freight'] },
+  { kind: 'plains', name: 'Plains', regions: 1, defaultStart: 0, trainsMayPass: false, sortsCars: false, entryPoints: ['start'] },
+  { kind: 'curves', name: 'Curves', regions: 2, defaultStart: 0, trainsMayPass: false, sortsCars: false, entryPoints: ['start'] },
   /**
-   * `entryPoints` is TRANSCRIBED, NOT READ — nothing anywhere reads this field on any profile, and
-   * the printed start positions are not modelled: `crossingStages` counts Stages instead. Recorded
-   * here because implications.md §6 describes the card as having FIVE distinct starts (plain,
-   * brakemen, airbrakes, plain, helpers) against the four listed, and that discrepancy should be
-   * settled against `Mainline Cards.pdf` if the starts are ever implemented — not quietly "fixed"
-   * now, when nothing depends on it either way.
+   * THE ONLY CARD THAT READS A TRAIN'S FAST/SLOW RATING. A fast train starts in the second region
+   * and is across in one Stage; a slow one starts at the beginning and takes two.
+   *
+   * It used to read the CONSIST instead — any coach aboard made the train "passenger" for this card
+   * — off the printed P60/F30. RAR corrected that directly: "I notice that you are basing stages in
+   * mainline cards off coach/non-coach. Actually, all trains are rated as FAST and SLOW."
    */
-  { kind: 'heavyGrade', name: 'Heavy Grade', speed: { kind: 'grade' }, trainsMayPass: false, sortsCars: false, entryPoints: ['start', 'brakemen', 'airbrakes', 'helpers'] },
-  { kind: 'doubleTrack', name: 'Double Track', speed: { kind: 'uniform', value: 60 }, trainsMayPass: true, sortsCars: false, entryPoints: ['start'] },
-  { kind: 'uncontrolledSiding', name: 'Uncontrolled Siding', speed: { kind: 'uniform', value: 60 }, trainsMayPass: true, sortsCars: false, entryPoints: ['noPass', 'passingTrains'] },
-  { kind: 'tunnel', name: 'Tunnel', speed: { kind: 'uniform', value: 30 }, trainsMayPass: false, sortsCars: false, entryPoints: ['start'] },
-  { kind: 'trestle', name: 'Trestle', speed: { kind: 'uniform', value: 60 }, trainsMayPass: false, sortsCars: false, entryPoints: ['start'] },
+  { kind: 'hilly', name: 'Hilly', regions: 2, defaultStart: 0, speedStarts: { fast: 1, slow: 0 }, trainsMayPass: false, sortsCars: false, entryPoints: ['fast', 'slow'] },
   /**
-   * RENAMED FROM "Yard" after play. The card is unchanged — same 60, same "sort cars in new
-   * order", same entry points, same art — but "Yard" collided with the Division Yard, the
-   * Classification Yard, the Salvage Yard, the Yard Office and the Small Yard, none of which are
-   * this. Its own key is renamed with it, so the two never drift apart. Those OTHER yards are
-   * deliberately left alone: they are different things that merely shared a word.
+   * THREE REGIONS, AND THE MODIFIERS MOVE THE START RATHER THAN CUTTING THE TIME — which comes to
+   * the same number of Stages and is how the card is actually printed and played. "If you play the
+   * home deck card 'helpers' against the mainline card heavy grade, it remains there the rest of the
+   * game and helps all trains going up hill by starting 1 region easier — so 2 to traverse, not 3.
+   * Other cards help the other direction, similar idea. Airbrakes is an upgrade from brakemen (which
+   * must be played first)."
+   *
+   * So: Helpers moves an UPHILL train up one region; Brakeman moves a DOWNHILL train up one, and
+   * Airbrakes another on top of it. A fully-equipped grade is one Stage downhill and two up.
    */
-  { kind: 'interchange', name: 'Interchange', speed: { kind: 'uniform', value: 60 }, trainsMayPass: false, sortsCars: true, entryPoints: ['start', 'sortCars'] },
+  { kind: 'heavyGrade', name: 'Heavy Grade', regions: 3, defaultStart: 0, trainsMayPass: false, sortsCars: false, entryPoints: ['start', 'brakemen', 'airbrakes', 'helpers'] },
+  { kind: 'doubleTrack', name: 'Double Track', regions: 1, defaultStart: 0, trainsMayPass: true, sortsCars: false, entryPoints: ['start'] },
+  /**
+   * TWO REGIONS, AND THE BACK ONE IS THE SIDING. A train with the card to itself starts past it and
+   * crosses in one Stage. "Uncontrolled siding: if a train already exists when you arrive, you go in
+   * the second stage back (you are in the siding and are one behind the other train). This prevents
+   * a collision — since you are not in same exact location."
+   *
+   * `trainsMayPass` is FALSE here, and used to be true. Two trains fit, but not by passing: the
+   * second one takes the siding and sits a region behind, which is what keeps them apart. Leaving it
+   * true skipped the collision test altogether and made the siding do nothing at all.
+   */
+  { kind: 'uncontrolledSiding', name: 'Uncontrolled Siding', regions: 2, defaultStart: 1, trainsMayPass: false, sortsCars: false, entryPoints: ['through', 'siding'] },
+  { kind: 'tunnel', name: 'Tunnel', regions: 2, defaultStart: 0, trainsMayPass: false, sortsCars: false, entryPoints: ['start'] },
+  { kind: 'trestle', name: 'Trestle', regions: 1, defaultStart: 0, trainsMayPass: false, sortsCars: false, entryPoints: ['start'] },
+  /**
+   * RENAMED FROM "Yard" after play. The card is unchanged — same "sort cars in new order", same art
+   * — but "Yard" collided with the Division Yard, the Classification Yard, the Salvage Yard, the
+   * Yard Office and the Small Yard, none of which are this. Its own key is renamed with it, so the
+   * two never drift apart. Those OTHER yards are deliberately left alone: they are different things
+   * that merely shared a word.
+   *
+   * TWO REGIONS, the back one a holding spur, exactly as the Uncontrolled Siding: "interchange has
+   * new extras show up in second region (like uncontrolled siding)", and earlier, "Plains is 1 stage
+   * for ALL trains. So are interlockings, with a second stage for incoming extras to hold at." A
+   * train running through crosses in one Stage; an Extra beginning its run here starts at the back.
+   */
+  { kind: 'interchange', name: 'Interchange', regions: 2, defaultStart: 1, trainsMayPass: false, sortsCars: true, entryPoints: ['through', 'extraStart'] },
 ];
 
 /**
@@ -579,44 +656,63 @@ export const MAINLINE_DECK: readonly MainlineKind[] = [
 ];
 
 /**
- * How many Stages a train needs to cross a Mainline card.
+ * WHERE A TRAIN ENTERS A MAINLINE CARD, and therefore how long it takes to cross (Gitea#3).
  *
- * Q1 — the printed 60/30 are miles per hour expressed as crossing time: a 60 card takes one Stage,
- * a 30 card takes two. The cells drawn on the cards are decoration.
- * Q2 — a Slow train adds one Stage to every card.
+ * "Regions shown on cards indicate how many stages it takes to cross. Plains is 1. Double track is
+ * 1, tunnel is 2, curves is 2, heavy grade is 3 unless you have help." A train advances one region
+ * per Stage, so the whole of crossing time is `regions - startRegion`.
  *
- * Hilly prints P60/F30, so it reads the consist rather than the speed class: a train carrying any
- * coach is "passenger" for this purpose.
+ * FOUR THINGS MOVE THE START, and nothing else does:
+ *
+ *   1. the card's own `defaultStart` — the Uncontrolled Siding and the Interchange print a back
+ *      region that is not part of the road, so a train running through begins past it;
+ *   2. `speedStarts`, on a card that prints a Fast and a Slow start. Only Hilly does;
+ *   3. the permanent Heavy Grade modifiers, which move a train one region up the hill each;
+ *   4. `takesSiding` / `startsAtBack`, the two occupancy cases below.
+ *
+ * WHAT NO LONGER MOVES IT: the printed mph, which is now scenery, and a train's Fast/Slow rating on
+ * any card but Hilly. That rating used to add a Stage to EVERY card, which is what made a Slow train
+ * cross Double Track in two Stages and produced the report this issue opened with.
  */
-export function crossingStages(
-  kind: MainlineKind,
-  trainSpeed: TrainSpeed,
-  carriesPassengers: boolean,
-  modifiers: readonly string[] = [],
-  direction: Direction = 'east',
-  gradeUp: Direction = 'east',
-): number {
-  const profile = MAINLINE_PROFILES.find((m) => m.kind === kind);
-  if (!profile) throw new Error(`unknown mainline card: ${kind}`);
+export type MainlineEntry = {
+  trainSpeed: TrainSpeed;
+  direction: Direction;
+  gradeUp: Direction;
+  modifiers: readonly string[];
+  /**
+   * The Uncontrolled Siding with a train already on it: this one takes the siding and sits a region
+   * behind, which is what keeps them out of the same place. Also the Interchange, where an Extra
+   * beginning its run starts in the holding region rather than on the road.
+   */
+  startsAtBack?: boolean;
+};
 
-  let mph: number;
-  switch (profile.speed.kind) {
-    case 'uniform':
-      mph = profile.speed.value;
-      break;
-    case 'byTrainType':
-      mph = carriesPassengers ? profile.speed.passenger : profile.speed.freight;
-      break;
-    case 'grade':
-      // Heavy Grade has no printed number; the modifier cards are what improve it, so it is a 30
-      // until one is placed.
-      mph = 30;
-      break;
+export function startRegion(kind: MainlineKind, entry: MainlineEntry): number {
+  const profile = mainlineProfile(kind);
+  if (entry.startsAtBack) return 0;
+
+  const base = profile.speedStarts
+    ? profile.speedStarts[entry.trainSpeed]
+    : profile.defaultStart;
+
+  const climbing = entry.direction === entry.gradeUp;
+  let help = 0;
+  if (kind === 'heavyGrade') {
+    if (climbing) {
+      if (entry.modifiers.includes('helpers')) help++;
+    } else {
+      // Airbrakes is an upgrade on Brakeman and cannot be played without it, so this counts both.
+      if (entry.modifiers.includes('brakeman')) help++;
+      if (entry.modifiers.includes('airbrakes')) help++;
+    }
   }
+  // Never past the last region: a card always costs at least one Stage to cross.
+  return Math.min(profile.regions - 1, base + help);
+}
 
-  const base = mph >= 60 ? 1 : 2;
-  const stages = base + (trainSpeed === 'slow' ? 1 : 0);
-  return Math.max(1, stages - gradeReduction(profile, modifiers, direction, gradeUp));
+/** How many Stages a train needs to cross a Mainline card — the regions it has left to run. */
+export function crossingStages(kind: MainlineKind, entry: MainlineEntry): number {
+  return mainlineProfile(kind).regions - startRegion(kind, entry);
 }
 
 /**
@@ -634,35 +730,50 @@ export function crossingStages(
 export function mainlineDescription(kind: MainlineKind, gradeUp: Direction = 'east'): string {
   const p = mainlineProfile(kind);
   const stages = (n: number): string => `${n} Stage${n === 1 ? '' : 's'}`;
+  const run = (entry: Partial<MainlineEntry>): number =>
+    crossingStages(kind, { trainSpeed: 'fast', direction: 'east', gradeUp, modifiers: [], ...entry });
   const parts: string[] = [];
 
-  if (p.speed.kind === 'byTrainType') {
-    // Hilly. The split is by CONSIST, not by the train's speed class: anything with a coach on it
-    // takes the passenger figure.
+  parts.push(`${p.regions} region${p.regions === 1 ? '' : 's'} — one Stage each.`);
+
+  if (p.speedStarts) {
     parts.push(
-      `P${p.speed.passenger} / F${p.speed.freight} — a train carrying ANY coach crosses as a ` +
-        `${p.speed.passenger} (${stages(crossingStages(kind, 'fast', true))} for a fast train), and a ` +
-        `freight-only train as a ${p.speed.freight} (${stages(crossingStages(kind, 'fast', false))}). ` +
-        `A slow train adds one Stage either way.`,
+      `This card reads the train's FAST/SLOW rating: a fast train starts further along and crosses ` +
+        `in ${stages(run({ trainSpeed: 'fast' }))}, a slow one in ${stages(run({ trainSpeed: 'slow' }))}. ` +
+        `No other card cares which it is.`,
     );
-  } else if (p.speed.kind === 'grade') {
+  } else if (kind === 'heavyGrade') {
     parts.push(
-      `A grade, climbing ${gradeUp === 'east' ? 'eastward' : 'westward'}. It crosses as a 30 — ` +
-        `${stages(crossingStages(kind, 'fast', false, [], gradeUp, gradeUp))} for a fast train, and one ` +
-        `more for a slow one. Brakeman and Airbrakes each take a Stage off a train running DOWNHILL; ` +
-        `Helpers takes one off a train running UPHILL. Never below one Stage.`,
+      `A grade, climbing ${gradeUp === 'east' ? 'eastward' : 'westward'}. ` +
+        `${stages(run({ direction: gradeUp }))} to climb it and ${stages(run({ direction: gradeUp === 'east' ? 'west' : 'east' }))} to run down, ` +
+        `before help. Helpers start an UPHILL train a region further on; Brakeman does the same ` +
+        `DOWNHILL and Airbrakes another again, and Airbrakes cannot be played without Brakeman. ` +
+        `Never less than one Stage.`,
+    );
+  } else if (p.defaultStart > 0) {
+    parts.push(
+      `A train with the card to itself starts past the back region and is across in ` +
+        `${stages(run({}))}.`,
     );
   } else {
+    parts.push(`${stages(run({}))} for every train — the printed speed is scenery.`);
+  }
+
+  if (kind === 'uncontrolledSiding') {
     parts.push(
-      `${p.speed.value} — ${stages(crossingStages(kind, 'fast', false))} for a fast train, ` +
-        `${stages(crossingStages(kind, 'slow', false))} for a slow one.`,
+      'UNCONTROLLED SIDING — arrive to find a train already here and you take the siding, a region ' +
+        'behind it. You are not in the same place, so you do not run into it; it costs you the ' +
+        'extra Stage instead.',
     );
   }
+  if (kind === 'interchange') {
+    parts.push('An Extra beginning its run here starts in the back region and takes the extra Stage.');
+  }
 
   if (p.trainsMayPass) {
     parts.push(
       'TRAINS MAY PASS — two trains may stand on this card at once, so a following train is not held ' +
-        'behind a slower one. Only this and the Double Track allow it.',
+        'behind a slower one.',
     );
   } else {
     parts.push('One train at a time — anything following has to wait for it to clear.');
@@ -672,43 +783,7 @@ export function mainlineDescription(kind: MainlineKind, gradeUp: Direction = 'ea
   return parts.join(' · ');
 }
 
-/**
- * WHICH WAY THE GRADE CLIMBS, AND WHY NO PLAYER CHOOSES IT.
- *
- * Q11, answered from the card: Heavy Grade prints "(Up)" and "Player sets orientation", so the climb
- * is a property of the PLACED CARD rather than a compass constant. `gradeUp` is the direction a train
- * is travelling when it goes UPHILL; a train heading the other way is descending.
- *
- * **The second half of that print is deliberately overridden.** No player sets it — `setup.ts` rolls
- * it from the seed. Settled v0.5.0 and re-confirmed 2026-08-23 after the question was raised again:
- * a Heavy Grade always sits BETWEEN two districts (or beyond an end Division Point next to one),
- * never inside one player's own, so there is no player with a fair claim to the choice — and the
- * choice is not cosmetic, because it decides which of the three modifiers below can ever pay and
- * therefore which direction of traffic is favoured, permanently. Giving it to the Superintendent was
- * considered and rejected in that re-examination: the office rotates every three Stages, the
- * advantage does not. Full reasoning in implications.md §10 Q11.
- *
- * Each applicable card takes a Stage off, never below one: a train cannot cross in no time.
- * Airbrakes only counts when Brakeman is already there, which the placement rule enforces
- * (`MAINLINE_MODIFIER_RULES`, `requiresOnCard`).
- */
-function gradeReduction(
-  profile: MainlineProfile,
-  modifiers: readonly string[],
-  direction: Direction,
-  gradeUp: Direction,
-): number {
-  if (profile.speed.kind !== 'grade') return 0;
-  const downhill = direction !== gradeUp;
-  let n = 0;
-  if (downhill) {
-    if (modifiers.includes('brakeman')) n++;
-    if (modifiers.includes('airbrakes')) n++;
-  } else if (modifiers.includes('helpers')) {
-    n++;
-  }
-  return n;
-}
+
 
 export function mainlineProfile(kind: MainlineKind): MainlineProfile {
   const p = MAINLINE_PROFILES.find((m) => m.kind === kind);
@@ -769,7 +844,8 @@ export const SPACE_USE_CARDS: readonly SimpleCard[] = [
   { key: 'flopHouse', name: 'Flop house', copies: 1, placement: 'adjacent to any straight, curve, turnout', effect: 'Burns tablespace.' },
   { key: 'watertower', name: 'Watertower', copies: 1, placement: 'adjacent to any straight, turnout on Running Track', effect: 'Burns tablespace.' },
   { key: 'hoboJungle', name: 'Hobo Jungle', copies: 1, placement: 'adjacent to any straight, turnout, Limit on Running Track', effect: 'Burns tablespace. Vandalism can loot a boxcar passing it.' },
-  { key: 'sectionHouse', name: 'Section House', copies: 1, placement: 'adjacent to any straight, curve, turnout', effect: 'Burns tablespace.' },
+  // Not in sheet 5 — dealt 0 copies (Jesse, 2026-08-26), the same treatment as the ladder.
+  { key: 'sectionHouse', name: 'Section House', copies: 0, placement: 'adjacent to any straight, curve, turnout', effect: 'Burns tablespace.' },
   { key: 'cityBlocks', name: 'City blocks', copies: 4, placement: 'adjacent to any straight, curve, turnout, Limit', effect: 'Burns tablespace.' },
   { key: 'engineShops', name: 'Engine Shops', copies: 1, placement: 'adjacent to any straight, curve, turnout', effect: 'Burns tablespace.' },
   { key: 'tenderloin', name: 'Tenderloin District', copies: 1, placement: 'adjacent to any straight, curve, turnout, Limit', effect: 'Burns tablespace.' },
@@ -876,16 +952,32 @@ export function enhancementRule(key: string): EnhancementRule | null {
 }
 
 export const ENHANCEMENT_CARDS: readonly SimpleCard[] = [
-  { key: 'interlocking', name: 'Interlocking', copies: 2, placement: 'any Running Track Straight', effect: 'May stop an inbound train on the Limit Track.' },
-  { key: 'facingPointLocks', name: 'Facing Point Locks', copies: 2, placement: 'adjacent to Interlocking', effect: 'Must have Interlocking. Prevents Derail being played on you.', answers: 'Derail' },
+  { key: 'interlocking', name: 'Interlocking', copies: 1, placement: 'any Running Track Straight', effect: 'May stop an inbound train on the Limit Track.' },
+  // Not in sheet 5 — dealt 0 copies (Jesse, 2026-08-26). It answers Derail, which is itself an
+  // Event held out until built, so at zero it defends against nothing that can be dealt anyway.
+  { key: 'facingPointLocks', name: 'Facing Point Locks', copies: 0, placement: 'adjacent to Interlocking', effect: 'Must have Interlocking. Prevents Derail being played on you.', answers: 'Derail' },
   { key: 'yardOffice', name: 'Yard office', copies: 1, placement: 'any Secondary Track Straight', effect: 'An inbound train with no coaches that can reach the yard office in one move may arrive there instead of the Train Order Office.' },
   { key: 'smallYard', name: 'Small yard', copies: 1, placement: 'any Secondary Track Straight', effect: 'A train that spends one move in the yard may sort itself into ANY order, including cars ahead of the engine.' },
-  { key: 'waterColumn', name: 'Water column', copies: 2, placement: 'any Running Track Straight', effect: 'Lets you remove any Watertower in your district.', answers: 'Watertower' },
+  { key: 'waterColumn', name: 'Water column', copies: 1, placement: 'any Running Track Straight', effect: 'Lets you remove any Watertower in your district.', answers: 'Watertower' },
   { key: 'overpass', name: 'Overpass', copies: 1, placement: 'any Railroad Crossing', effect: 'Removes the restrictions of a played Railroad Crossing.', answers: 'Railroad crossing' },
-  { key: 'telegraph', name: 'Telegraph', copies: 3, placement: 'any Running Track Straight', effect: 'Once a day, when dispatching facing trains, add +4 to the other train’s number.' },
-  { key: 'telephone', name: 'Telephone', copies: 2, placement: 'on Telegraph', effect: 'Once a day, add +8 to the other train’s number.' },
-  { key: 'radio', name: 'Radio', copies: 2, placement: 'on Telephone', effect: 'Once a day, add +12 to the other train’s number.' },
-  { key: 'absSignals', name: 'ABS Signals', copies: 2, placement: 'any Mainline card', effect: 'Trains on this card will not rear-end each other; they stop short of a collision.' },
+  /**
+   * THE DISPATCHING LADDER IS OUT OF THE DECK, at 0 copies rather than deleted — the treatment
+   * Poling and the sharp curves already get, and for the same reason.
+   *
+   * `docs/Deck cards5.xlsx` does not list Telegraph, Telephone or Radio at any count, and **Jesse
+   * confirmed (2026-08-26) that the removal is deliberate, not a row that failed to carry across**
+   * from sheet 2. So no copy is dealt, which is what the sheet asks for.
+   *
+   * The rows and `ENHANCEMENT_RULES`' `dispatchBonus` chain stay exactly where they are. The rule
+   * is implemented and tested — `advance.ts` reads the ladder when the Superintendent dispatches
+   * facing trains, best device first — and deleting working machinery to express a count of zero
+   * would throw away the only record of how it worked. At zero copies the code is unreachable: no
+   * card is ever dealt, so nothing ever places one, so the bonus never applies.
+   */
+  { key: 'telegraph', name: 'Telegraph', copies: 0, placement: 'any Running Track Straight', effect: 'Once a day, when dispatching facing trains, add +4 to the other train’s number.' },
+  { key: 'telephone', name: 'Telephone', copies: 0, placement: 'on Telegraph', effect: 'Once a day, add +8 to the other train’s number.' },
+  { key: 'radio', name: 'Radio', copies: 0, placement: 'on Telephone', effect: 'Once a day, add +12 to the other train’s number.' },
+  { key: 'absSignals', name: 'ABS Signals', copies: 1, placement: 'any Mainline card', effect: 'Trains on this card will not rear-end each other; they stop short of a collision.' },
 ];
 
 export const MAINLINE_MODIFIER_CARDS: readonly SimpleCard[] = [
@@ -893,7 +985,8 @@ export const MAINLINE_MODIFIER_CARDS: readonly SimpleCard[] = [
   { key: 'airbrakes', name: 'Airbrakes', copies: 1, placement: 'a GRADE Mainline card', effect: 'Faster passage downhill. Brakeman must be in effect.' },
   { key: 'helpers', name: 'Helpers', copies: 1, placement: 'a GRADE Mainline card', effect: 'Faster passage uphill.' },
   { key: 'realignment', name: 'Realignment', copies: 2, placement: 'a Mainline card', effect: 'Convert one Mainline type to another. Not while a train is on it.' },
-  { key: 'facingPointLocksMainline', name: 'Facing Point Locks', copies: 2, placement: 'adjacent to Interlocking', effect: 'Prevents Derail being played on you.', answers: 'Derail' },
+  // Not in sheet 5 — dealt 0 copies (Jesse, 2026-08-26); see the Enhancement of the same name.
+  { key: 'facingPointLocksMainline', name: 'Facing Point Locks', copies: 0, placement: 'adjacent to Interlocking', effect: 'Prevents Derail being played on you.', answers: 'Derail' },
 ];
 
 /**
@@ -904,8 +997,9 @@ export const MAINLINE_MODIFIER_CARDS: readonly SimpleCard[] = [
 export const SECOND_SECTION = { key: 'secondSection', name: 'Second Section', copies: 1 };
 
 export const MANEUVER_CARDS: readonly SimpleCard[] = [
-  { key: 'redFlags', name: 'Red Flags', copies: 5, placement: 'any time', effect: 'A stopped train is prevented from being hit; the approaching train is prevented from moving.' },
-  { key: 'flyingSwitch', name: 'Flying Switch', copies: 1, placement: 'any time', effect: 'Break a cut of cars away from behind the engine and roll them into an industry.' },
+  { key: 'redFlags', name: 'Red Flags', copies: 3, placement: 'any time', effect: 'A stopped train is prevented from being hit; the approaching train is prevented from moving.' },
+  // Not in sheet 5 — dealt 0 copies (Jesse, 2026-08-26). The reducer stays; nothing can reach it.
+  { key: 'flyingSwitch', name: 'Flying Switch', copies: 0, placement: 'any time', effect: 'Break a cut of cars away from behind the engine and roll them into an industry.' },
   // POLING IS OUT OF THE DECK, at 0 copies rather than deleted.
   //
   // It is the one card whose effect the source records as "TBD", so there is nothing to implement
@@ -922,7 +1016,8 @@ export const ACTION_CARDS: readonly SimpleCard[] = [
   { key: 'perDiemInventory', name: 'Per Diem inventory', copies: 1, placement: 'another player', effect: 'Lose one point per 2 empty cars on Secondary Tracks.' },
   { key: 'demurrageCharge', name: 'Demurrage charge', copies: 1, placement: 'another player', effect: 'Lose one point per 2 loaded freight cars on Secondary Tracks.' },
   { key: 'customerComplaints', name: 'Customer complaints', copies: 1, placement: 'another player', effect: 'Lose one point per 2 coaches in loading boxes.' },
-  { key: 'vandalism', name: 'Vandalism', copies: 1, placement: 'another player', effect: 'A train passing a Hobo Jungle has a boxcar looted (converted to empty).' },
+  // Not in sheet 5 — dealt 0 copies (Jesse, 2026-08-26).
+  { key: 'vandalism', name: 'Vandalism', copies: 0, placement: 'another player', effect: 'A train passing a Hobo Jungle has a boxcar looted (converted to empty).' },
   { key: 'hotbox', name: 'Hotbox', copies: 1, placement: 'another player', effect: 'A train just arrived must set one car (chooser’s pick) onto Secondary Track until it departs.' },
   { key: 'outlawed', name: 'Outlawed', copies: 1, placement: 'another player', effect: 'A train just arrived may not depart for one turn — the crew’s hours have expired.' },
 ];
@@ -972,7 +1067,6 @@ export function crewTrayCount(players: number): number {
   return players + 3;
 }
 
-export const REGIONS_PER_MAINLINE_CARD = 2; // provisional, pending §10 Q2
 export const STAGES_PER_DAY = 12;
 export const STAGES_PER_SHIFT = 3;
 export const HAND_LIMIT = 3;

src/engine/events.ts

@@ -195,6 +195,23 @@ export type GameEvent =
   | { type: 'unloadBegan'; player: PlayerIndex; at: GridCoord; carType: CarType; carIndex: number }
   // -- consequences
   | { type: 'revenueChanged'; player: PlayerIndex; delta: number; total: number; reason: string }
-  | { type: 'phaseEnded'; player: PlayerIndex; phase: string };
+  | { type: 'phaseEnded'; player: PlayerIndex; phase: string }
+  // -- §3.3, extended play (Gitea#11)
+  /**
+   * One seat's answer to "play one more Day?". Every seat votes; the vote is unanimous, and one
+   * refusal ends it. In the log so that a table can see who is still being waited on, and who
+   * called time.
+   */
+  | { type: 'extensionVoted'; player: PlayerIndex; agree: boolean }
+  /** The table agreed. `day` is the Day the extra one becomes — `config.days + extraDays`. */
+  | { type: 'dayExtended'; day: number }
+  /**
+   * Play is over for good — somebody declined the extension.
+   *
+   * Distinct from the ending itself, which `checkVictory` already announced by way of the result: an
+   * ending that COULD have been played past and was not is a decision the table made, and the log
+   * should say so rather than simply stopping.
+   */
+  | { type: 'playConcluded'; declinedBy: PlayerIndex };
 
 export type EventType = GameEvent['type'];

src/engine/intents.ts

@@ -147,7 +147,28 @@ export type Intent =
   | { type: 'laborer.startLoad'; at: GridCoord }
   | { type: 'laborer.advanceLoad'; at: GridCoord; box: number }
   | { type: 'laborer.beginUnload'; at: GridCoord; carIndex: number }
-  | { type: 'loadUnload.end' };
+  | { type: 'loadUnload.end' }
+  /**
+   * §3.3, EXTENDED PLAY (Gitea#11) — one vote on whether to play one more Day.
+   *
+   * ARRIVES OUT OF TURN, like `mainline.clearance`, and unlike it goes to EVERY seat rather than to
+   * the Superintendent: it is a table decision, not a ruling. Unanimous, and one `agree: false`
+   * ends the game immediately — nobody waits on a player who has already refused.
+   *
+   * It is an intent, rather than a button the client handles by itself, because a save is
+   * `{ seed, config, history }` replayed through the engine: a decision that is not in the history
+   * did not happen, and an extended game would evaporate on the next reload, Undo, or server
+   * restart. This is the record of the table agreeing.
+   *
+   * CARRIES ITS VOTER, uniquely among intents, and it has to. A saved history is a flat `Intent[]`
+   * with no seat recorded against each move: the replay DERIVES who acted from the turn order
+   * (`fromMultiplayerSave`). That works for every other intent, including `mainline.clearance`,
+   * because there is exactly one seat it could have been. Here there is not — every seat may vote,
+   * in any order — so a vote whose voter is not written down cannot be replayed at all, and a
+   * resumed server would refuse the save with `NO_ACTOR`. The server checks this against the seat
+   * it authenticated (`NOT_YOUR_TURN`), so it is a record, never a claim.
+   */
+  | { type: 'game.extend'; player: PlayerIndex; agree: boolean };
 
 export type IntentType = Intent['type'];
 
@@ -276,7 +297,11 @@ export type RejectionCode =
    * §6.2, Jesse's ruling (Gitea#6) — a train card is never discarded. Hold it as long as you like;
    * the only way it leaves your hand is onto the timetable.
    */
-  | 'TRAINS_ARE_NEVER_DISCARDED';
+  | 'TRAINS_ARE_NEVER_DISCARDED'
+  /** §3.3 (Gitea#11) — `game.extend` when the game is not waiting on an extension vote. */
+  | 'NOT_AWAITING_EXTENSION'
+  /** §3.3 (Gitea#11) — this seat has already voted on this extension. */
+  | 'ALREADY_VOTED';
 
 export type Rejection = { code: RejectionCode; message: string };
 

src/engine/legal.ts

@@ -68,6 +68,23 @@ export function isLegal(s: GameState, player: PlayerIndex, i: Intent): boolean {
 function candidates(s: GameState, player: PlayerIndex): Intent[] {
   const out: Intent[] = [];
 
+  /**
+   * §3.3, EXTENDED PLAY (Gitea#11) — the only thing on offer when the timetable has run out and the
+   * table is being asked whether to play on.
+   *
+   * Returned EARLY rather than added to the list, because nothing else is legal in this state and
+   * the phase switch below would otherwise generate a boardful of candidates for `check` to reject
+   * one at a time. It also puts the vote in front of the bot driver through the ordinary path, which
+   * is what lets a bot seat answer without the engine having to know which seats are bots.
+   */
+  if (s.status === 'awaitingExtension') {
+    if (s.extensionVotes[player] === null) {
+      out.push({ type: 'game.extend', player, agree: true });
+      out.push({ type: 'game.extend', player, agree: false });
+    }
+    return out;
+  }
+
   // The clearance ruling arrives out of turn order and goes to the Superintendent (§8.1).
   if (s.clock.pendingDecision !== null) {
     out.push({ type: 'mainline.clearance', allow: true });

src/engine/setup.ts

@@ -43,7 +43,7 @@ import type {
   TrackCard,
   TrayId,
 } from './state.ts';
-import { coordKey, freshTurns } from './state.ts';
+import { coordKey, emptyTally, freshTurns } from './state.ts';
 
 export type SetupOptions = {
   id: string;
@@ -246,7 +246,7 @@ function buildDivision(players: number, rng: Rng): DivisionNode[] {
     if (deck.length === 0) throw new Error('the Mainline deck ran out — too many players for it');
     const card = deck.splice(rng.nextInt(deck.length), 1)[0]!;
     const node: DivisionNode = { kind: 'mainline', card, transits: [] };
-    if (mainlineProfile(card).speed.kind === 'grade') {
+    if (card === 'heavyGrade') {
       /**
        * SETTLED, not provisional (v0.5.0, Jesse's call) — this overrides the card's own printed
        * "Player sets orientation". A Heavy Grade sits on the shared west-to-east chain BETWEEN two
@@ -434,6 +434,10 @@ export function createGame(opts: SetupOptions): GameState {
     collisionsTotal: 0,
     status: 'active',
     outcome: null,
+    extraDays: 0,
+    extensionVotes: players.map(() => null),
+    official: null,
+    tally: emptyTally(playerCount),
   };
 }
 

src/engine/state.ts

@@ -684,6 +684,139 @@ export type Outcome = {
   reason: OutcomeReason;
 };
 
+/**
+ * §3.3, EXTENDED PLAY (Gitea#11) — which endings may be played past.
+ *
+ * Both days-based endings offer another Day: running out of timetable, and closing short of the
+ * combined Revenue floor, are the same event seen twice — the last Day ended and this is what the
+ * books say. A `collisionFloor` ending is NOT extendable, and neither is a collision breach that
+ * happens during an extended Day: §3.4 stopped the game because the railroad was declared unsafe,
+ * and carrying on regardless would contradict the rule that stopped it (Jesse's call, 2026-08-28).
+ */
+export function isExtendable(reason: OutcomeReason): boolean {
+  return reason === 'daysElapsed' || reason === 'revenueFloor';
+}
+
+/**
+ * Running counts of everything interesting that has happened, tallied from the event stream
+ * (Gitea#16).
+ *
+ * WHY IT LIVES ON `GameState` rather than being computed by whoever happens to want it. Three
+ * reasons, in ascending order of how much they cost to work around:
+ *
+ *   1. `snapshot()` already takes a `GameState`, so every number here reaches a MULTIPLAYER client
+ *      through the `Frame` it is already being sent — no new server route, no new `Push` field, no
+ *      new `Session` method, and no second implementation that can disagree with the first.
+ *   2. It is REPLAY-EXACT. A save is `{ seed, config, history }` replayed through the engine
+ *      (`web/game.ts`'s `fromSave`), so a tally folded from the events that replay emits is rebuilt
+ *      identically every time — which is what makes Undo and a server restart correct here for free.
+ *   3. The official result freezes a COPY of this at the moment the timetable ran out (`official`
+ *      below), and a frozen copy has to be taken from something that already exists.
+ *
+ * Aggregate counts only. Nothing here is seat-secret — no card ids, no hands — which is why
+ * `test/redaction.test.ts` stays green with the whole thing on the Frame.
+ *
+ * NOT SCORING. Nothing in here feeds a rule; it is read by the results screen and by the badge work
+ * that Gitea#16 leaves to a second pass. Adding a counter is always safe.
+ */
+export type Tally = {
+  /** §8.3 — a train that ran the length of the Division and left it. */
+  trainsCompleted: number;
+  /**
+   * Of those, how many did some switching between being made up and leaving.
+   *
+   * The join Gitea#16 asks for by name ("a player who completes an entire game where every train
+   * that passed through did some switching on"). Counted as the train completes, against whether
+   * that tray has coupled or dropped anything since it was made up — which is why `switchedSince`
+   * below exists rather than this being derivable afterwards.
+   */
+  trainsCompletedWithWork: number;
+  /** §10 — trains lost to a collision, and the cars that went with them. */
+  trainsDestroyed: number;
+  carsDestroyed: number;
+  /** §6 — switching volume, both directions. */
+  carsCoupled: number;
+  carsDropped: number;
+  /** §9.1 — the MEN | AT | WORK pipeline: begun, and carried all the way through. */
+  loadsStarted: number;
+  loadsCompleted: number;
+  unloadsBegun: number;
+  unloadsCompleted: number;
+  /** §9.2 — passenger work. */
+  passengersBoarded: number;
+  passengersDetrained: number;
+  /** Colour, and the vocabulary the badge pass will draw on. */
+  flyingSwitches: number;
+  officeUpgrades: number;
+  dispatchBonusesUsed: number;
+  facilitiesUnjammed: number;
+  expediteFaults: number;
+  trainsHeld: number;
+  trainsDiverted: number;
+  secondSections: number;
+  extrasStarted: number;
+  cardsDrawn: number;
+  cardsPlayed: number;
+  cardsDiscarded: number;
+  clearancesRequested: number;
+  /** §8.1 — rulings that let the other train through. A refusal is a ruling too, but not this one. */
+  clearancesAllowed: number;
+  /**
+   * X18 CIRCUS SET-UPS — a train that spent a Stage standing still and was paid for it (§X18).
+   *
+   * NOT "the longest an engine sat on a siding", which is what Gitea#16 asks for and what the
+   * comment on that issue assumed this was. `trainStoodStill` is emitted ONCE IN A GAME PER SUCH
+   * TRAIN — only for a train whose profile has `stopEarnsPoint`, and `advance.ts` sets
+   * `stopPointClaimed` so it can never fire twice. There is no per-Stage "this train did not move"
+   * signal in the engine at all, so a longest-stand streak cannot be folded from the event stream:
+   * it needs an engine-side signal that does not exist yet. Recorded in `TODO.md` for the badge
+   * pass rather than shipped as a statistic that would read "1 Stage" for ever.
+   */
+  circusStops: { trainNumber: number; where: string }[];
+  /**
+   * Train numbers that have coupled or dropped something since they were made up, for
+   * `trainsCompletedWithWork`. Cleared when the train is made up and when it leaves the Division.
+   */
+  switchedSince: number[];
+  /** Indexed by PLAYER. Only events that name a player reach these. */
+  byPlayer: PlayerTally[];
+};
+
+export type PlayerTally = {
+  loads: number;
+  unloads: number;
+  passengersBoarded: number;
+  passengersDetrained: number;
+  cardsPlayed: number;
+  /** §10 — collisions this player was faulted for, not collisions they were caught in. */
+  collisions: number;
+  /** Revenue gained and Revenue lost, kept apart: the net is already on `players[i].revenue`. */
+  revenueGained: number;
+  revenueLost: number;
+};
+
+/**
+ * THE OFFICIAL RESULT, frozen at the moment the timetable ran out (Gitea#11).
+ *
+ * "The winner is based upon the original game length. In a five-day game, even if it's extended to
+ * eight or nine days, the winner and the official answer is the winner at the end of five days"
+ * (Jesse, 2026-08-28). So this is written ONCE, at the first ending, and never overwritten —
+ * including by a §3.4 collision breach during an extended Day, which ends play without touching it.
+ *
+ * `state.outcome` keeps moving: it is always the CURRENT evaluation, which is what the live game
+ * wants. Once `official` exists, everything after it is informational.
+ */
+export type FinalReport = {
+  /** The Day the game was scheduled to end on — always `config.days`. */
+  day: number;
+  outcome: Outcome;
+  /** Every player's Revenue at that moment, in player order. */
+  revenues: number[];
+  collisionsTotal: number;
+  /** The Tally as it stood when the timetable ran out. */
+  tally: Tally;
+};
+
 /**
  * Per-Stage transient bookkeeping for the acting player. Reset when the actor changes.
  *
@@ -726,6 +859,66 @@ export function freshTurns(players: number, moves: number): Map<PlayerIndex, Tur
   return turns;
 }
 
+/** A Tally with everything at zero — the state every game starts in (Gitea#16). */
+export function emptyTally(players: number): Tally {
+  return {
+    trainsCompleted: 0,
+    trainsCompletedWithWork: 0,
+    trainsDestroyed: 0,
+    carsDestroyed: 0,
+    carsCoupled: 0,
+    carsDropped: 0,
+    loadsStarted: 0,
+    loadsCompleted: 0,
+    unloadsBegun: 0,
+    unloadsCompleted: 0,
+    passengersBoarded: 0,
+    passengersDetrained: 0,
+    flyingSwitches: 0,
+    officeUpgrades: 0,
+    dispatchBonusesUsed: 0,
+    facilitiesUnjammed: 0,
+    expediteFaults: 0,
+    trainsHeld: 0,
+    trainsDiverted: 0,
+    secondSections: 0,
+    extrasStarted: 0,
+    cardsDrawn: 0,
+    cardsPlayed: 0,
+    cardsDiscarded: 0,
+    clearancesRequested: 0,
+    clearancesAllowed: 0,
+    circusStops: [],
+    switchedSince: [],
+    byPlayer: Array.from({ length: players }, () => ({
+      loads: 0,
+      unloads: 0,
+      passengersBoarded: 0,
+      passengersDetrained: 0,
+      cardsPlayed: 0,
+      collisions: 0,
+      revenueGained: 0,
+      revenueLost: 0,
+    })),
+  };
+}
+
+/**
+ * A deep copy, for freezing the official result (`FinalReport`).
+ *
+ * Written out rather than reached for via `structuredClone` because a Tally is a flat bag of numbers
+ * with two containers in it, and spelling the copy out means a field added later that needs deep
+ * copying is a compile error here rather than a shared reference discovered in a results screen.
+ */
+export function cloneTally(t: Tally): Tally {
+  return {
+    ...t,
+    circusStops: t.circusStops.map((c) => ({ ...c })),
+    switchedSince: [...t.switchedSince],
+    byPlayer: t.byPlayer.map((p) => ({ ...p })),
+  };
+}
+
 /**
  * WHICH WAY TO DRAW THE ENGINE — east or west, for every train, everywhere.
  *
@@ -771,22 +964,25 @@ export function standingSides(
 }
 
 /**
- * The cut a train would run into if it left this card through `exit` — the cars between it and that
- * end of the card.
+ * The cut a train would run into if it left this card by the `exit` END OF THE ROW — the cars
+ * between it and that end. Returned in the order the train MEETS them, nearest first, which is what
+ * `carsCoupled` wants.
  *
- * Only 'e' and 'w' can hold a cut: the array is a west-to-east row, so a train leaving north or
- * south off a curve or a spur is not running along it and meets nothing. Returned in the order the
- * train MEETS them, nearest first, which is what `carsCoupled` wants.
+ * `exit` IS AN END OF THE ROW, NOT A PORT. It used to be a raw `Port`, and answered "you meet
+ * nothing" for north and south on the reasoning that a leg leaving through an edge is not running
+ * along the west-to-east row. It is: a `sw` curve's south leg IS the east end of that row, so a
+ * crew standing on the curve pulled out through the leg and drove away leaving the cars beside it
+ * standing, against §A.4's mandatory coupling (Gitea#17). Callers resolve the leg with `rowEndAt`
+ * (`track.ts`), which lives there because only the card's arc can say which end a leg is — and the
+ * narrowed type is what makes every caller do it.
  */
 export function cutTowards(
   tray: { standingWest?: number | undefined },
   cars: readonly RollingStock[],
-  exit: 'n' | 's' | 'e' | 'w',
+  exit: 'e' | 'w',
 ): RollingStock[] {
   const { west, east } = standingSides(tray, cars);
-  if (exit === 'e') return east;
-  if (exit === 'w') return [...west].reverse();
-  return [];
+  return exit === 'e' ? east : [...west].reverse();
 }
 
 export function turnOf(s: GameState, player: PlayerIndex): TurnState {
@@ -863,8 +1059,34 @@ export type GameState = {
   collisionsToday: number;
   /** §3.4 — never reset; checked against `config.maxCollisionsTotal`. */
   collisionsTotal: number;
-  status: 'setup' | 'active' | 'finished';
+  /**
+   * `awaitingExtension` is Gitea#11: the timetable has run out, the result is recorded, and the
+   * table is being asked whether to play one more Day. It is a PAUSE, not an ending — `advance`
+   * reports `needsInput` there, the server resumes it like any live game, and the only intent the
+   * rules will accept is `game.extend`.
+   */
+  status: 'setup' | 'active' | 'awaitingExtension' | 'finished';
+  /** The CURRENT evaluation, re-decided at the end of every Day including extended ones. */
   outcome: Outcome | null;
+  /**
+   * §3.3 (Gitea#11) — Days granted beyond `config.days`, one vote at a time.
+   *
+   * `config.days` is deliberately never touched: it is what the official result was decided at, so
+   * leaving it alone is what makes "the winner is decided at the original game length" a fact about
+   * the code rather than a comment on it.
+   */
+  extraDays: number;
+  /**
+   * Per PLAYER, while `awaitingExtension`. `null` means they have not voted yet.
+   *
+   * Unanimous, and one refusal is decisive: nobody is made to wait on a player who has already said
+   * no (Jesse's call, 2026-08-28). Solitaire is the same code with one voter.
+   */
+  extensionVotes: (boolean | null)[];
+  /** Frozen at the FIRST ending and never overwritten. See `FinalReport`. */
+  official: FinalReport | null;
+  /** Gitea#16. Folded from the event stream; see `Tally`. */
+  tally: Tally;
 };
 
 // ---------------------------------------------------------------------------

src/engine/tally.ts

@@ -0,0 +1,201 @@
+/**
+ * The event tally — Gitea#16's statistics, folded from the event stream into `GameState.tally`.
+ *
+ * WHERE IT IS HOOKED, and why it is not in `reduce`. `apply.ts`'s `reduce` sees only the events an
+ * INTENT produced; `advance.ts` mutates state directly and pushes its events without reducing them
+ * at all — and `advance` is where `trainCompleted`, `trainsDestroyed` and `trainStoodStill` come
+ * from, which are exactly the numbers this issue asks for. So the fold is hooked at the two places
+ * every event in the game passes through exactly once on its way to a caller:
+ *
+ *   - `applyIntent` (`apply.ts`), beside its `reduce` loop;
+ *   - `advance` (`advance.ts`), which now wraps the phase driver and folds what it returns.
+ *
+ * Exactly once matters in both directions: an event folded twice inflates a count, and an event
+ * folded nowhere is a statistic that silently reads zero. `test/tally.test.ts` pins both by playing
+ * real games and checking the tally against an independent count over the same event array.
+ *
+ * NOTHING HERE IS A RULE. The tally is read by the results screen and by the badge work Gitea#16
+ * leaves to a second pass; no engine decision consults it. That is what makes adding a counter
+ * always safe.
+ *
+ * WHAT IS NOT COUNTED PER PLAYER, and why. `carsCoupled` and `carsDropped` carry a `trayId` and no
+ * `player` — switching is done BY a crew, and the event says which crew rather than which person.
+ * Rather than guess an owner from whose turn it happened to be, those two are table totals only.
+ * The events that do name a player (`loadCompleted`, `passengersBoarded`, `cardPlayed`,
+ * `revenueChanged`, `trainsDestroyed`) are the ones `byPlayer` reports.
+ */
+
+import type { GameEvent } from './events.ts';
+import type { GameState } from './state.ts';
+
+/**
+ * Fold one event into `s.tally`.
+ *
+ * The switch is deliberately not exhaustive — most of the 49 event types say nothing a player would
+ * want counted, and listing them all to `break` would bury the ones that do. A `default` that does
+ * nothing is the honest shape.
+ */
+export function tallyEvent(s: GameState, e: GameEvent): void {
+  const t = s.tally;
+  const mine = 'player' in e && typeof e.player === 'number' ? t.byPlayer[e.player] : undefined;
+
+  switch (e.type) {
+    /**
+     * §X18 — the Circus train set up and was paid for the Stage it spent standing.
+     *
+     * NOT a "longest stand" streak, which is what Gitea#16 wants and what its comment assumed this
+     * event was. It fires once in a game per such train: only trains whose profile sets
+     * `stopEarnsPoint` emit it at all, and `advance.ts` claims it once with `stopPointClaimed`. So
+     * there is nothing to count a run of, and the honest thing to report is the event itself.
+     */
+    case 'trainStoodStill':
+      t.circusStops.push({ trainNumber: e.trainNumber, where: e.where });
+      break;
+
+    /**
+     * DID THIS TRAIN DO ANY SWITCHING — Gitea#16's "switching master" join, kept as it happens
+     * rather than reconstructed afterwards.
+     *
+     * The two halves of the join are in different currencies: switching events name a `trayId` and
+     * completion names a `trainNumber`, and no event carries both. The tray is looked up in LIVE
+     * state, which is sound precisely here — a crew that has just coupled or dropped is still on the
+     * board — where re-deriving it at completion time would not be, the tray having been released by
+     * then. A lookup that misses costs one train its mark on a statistic; it cannot affect a rule.
+     */
+    case 'carsCoupled':
+      t.carsCoupled += e.stock.length;
+      markSwitched(s, e.trayId);
+      break;
+
+    case 'carsDropped':
+      t.carsDropped += e.stock.length;
+      markSwitched(s, e.trayId);
+      break;
+
+    case 'flyingSwitch':
+      t.flyingSwitches += 1;
+      markSwitched(s, e.trayId);
+      break;
+
+    // A tray is reused run after run, so a fresh train starts with a clean sheet.
+    case 'trainMadeUp':
+      t.switchedSince = t.switchedSince.filter((n) => n !== e.trainNumber);
+      break;
+
+    case 'trainCompleted':
+      t.trainsCompleted += 1;
+      if (t.switchedSince.includes(e.trainNumber)) t.trainsCompletedWithWork += 1;
+      t.switchedSince = t.switchedSince.filter((n) => n !== e.trainNumber);
+      break;
+
+    case 'trainsDestroyed':
+      t.trainsDestroyed += e.trains.length;
+      for (const train of e.trains) t.carsDestroyed += train.consist.length;
+      if (mine) mine.collisions += 1;
+      break;
+
+    case 'officeUpgraded':
+      t.officeUpgrades += 1;
+      break;
+
+    case 'dispatchBonusUsed':
+      t.dispatchBonusesUsed += 1;
+      break;
+
+    case 'facilityUnjammed':
+      t.facilitiesUnjammed += 1;
+      break;
+
+    case 'expediteFault':
+      t.expediteFaults += 1;
+      break;
+
+    case 'trainHeld':
+      t.trainsHeld += 1;
+      break;
+
+    case 'trainDiverted':
+      t.trainsDiverted += 1;
+      break;
+
+    case 'secondSectionOrdered':
+      t.secondSections += 1;
+      break;
+
+    case 'extraStarted':
+      t.extrasStarted += 1;
+      break;
+
+    case 'cardDrawn':
+      t.cardsDrawn += 1;
+      break;
+
+    case 'cardPlayed':
+      t.cardsPlayed += 1;
+      if (mine) mine.cardsPlayed += 1;
+      break;
+
+    case 'cardDiscarded':
+      t.cardsDiscarded += 1;
+      break;
+
+    case 'clearanceRequested':
+      t.clearancesRequested += 1;
+      break;
+
+    // §8.1 — a ruling is given either way; only a YES let the other train through.
+    case 'clearanceGiven':
+      if (e.allow) t.clearancesAllowed += 1;
+      break;
+
+    case 'loadStarted':
+      t.loadsStarted += 1;
+      break;
+
+    case 'loadCompleted':
+      t.loadsCompleted += 1;
+      if (mine) mine.loads += 1;
+      break;
+
+    case 'unloadBegan':
+      t.unloadsBegun += 1;
+      break;
+
+    case 'unloadCompleted':
+      t.unloadsCompleted += 1;
+      if (mine) mine.unloads += 1;
+      break;
+
+    case 'passengersBoarded':
+      t.passengersBoarded += 1;
+      if (mine) mine.passengersBoarded += 1;
+      break;
+
+    case 'passengersDetrained':
+      t.passengersDetrained += 1;
+      if (mine) mine.passengersDetrained += 1;
+      break;
+
+    /**
+     * Gained and lost are kept APART because the net is already on `players[i].revenue`. What the
+     * results screen cannot otherwise say is how much of a modest final score was earned and then
+     * handed back at a grade crossing — which is the whole difference between a quiet game and an
+     * eventful one.
+     */
+    case 'revenueChanged':
+      if (mine) {
+        if (e.delta >= 0) mine.revenueGained += e.delta;
+        else mine.revenueLost += -e.delta;
+      }
+      break;
+
+    default:
+      break;
+  }
+}
+
+function markSwitched(s: GameState, trayId: string): void {
+  const n = s.trays.get(trayId)?.trainNumber;
+  if (n === undefined || n === null) return;
+  if (!s.tally.switchedSince.includes(n)) s.tally.switchedSince.push(n);
+}

src/engine/track.ts

@@ -188,6 +188,40 @@ export function joins(a: TrackCard, p: Port, b: TrackCard): boolean {
   return slopeAt(a, p) === slopeAt(b, opposite(p));
 }
 
+/**
+ * WHICH END OF THE WEST-TO-EAST ROW A PORT SITS AT.
+ *
+ * `TrackCard.standing` is ordered west to east (§A.3), so whether a train meets the row front to
+ * back or back to front depends on which end it enters by — and a port is not always at one of
+ * those two extremes. Every 45° leg leaves through the MIDDLE of its north or south edge, so its
+ * end of the run is whichever end the arc does NOT reach: a `sw` curve's south leg is the EAST end
+ * of the row, and an `se` curve's south leg is the WEST end. Same port, opposite answers, which is
+ * why this has to ask the card rather than read the port.
+ *
+ * Gitea#17 is what both callers looked like without it. `exploreMoves` reversed the row for an 'e'
+ * entry and for nothing else, so backing into a cut through a `sw` curve's south leg coupled it up
+ * back to front — the caboose came out next to the engine, which §8.2 then calls badly made up.
+ * `cutTowards` answered "you meet nothing" for a north or south exit, so a crew standing on a curve
+ * pulled out through the leg and left the cars beside it standing, which §A.4 forbids.
+ *
+ * There is no north-south straight anywhere on the printed sheet (see the module comment), so a run
+ * touching a 45° leg always has an east or west port at its other end and the answer is never
+ * undefined. A TURNOUT is the one card whose row has three ends rather than two — and it is also
+ * the one card no cut can ever stand on, since a train may not stop there (§A.1) and so never sets
+ * anything out there. Its stem answers for it.
+ */
+export function rowEndAt(card: TrackCard, p: Port): 'e' | 'w' {
+  if (p === 'e' || p === 'w') return p;
+  for (const [a, b] of connectionsFor(card)) {
+    const other = a === p ? b : b === p ? a : null;
+    if (other === 'e') return 'w';
+    if (other === 'w') return 'e';
+  }
+  // Not a card the printed sheet can produce. Reading the leg as the west end leaves the row in the
+  // order it is stored rather than inventing a reversal on a card nothing knows the shape of.
+  return 'w';
+}
+
 // ---------------------------------------------------------------------------
 // Orientation (Gap 11)
 // ---------------------------------------------------------------------------
@@ -450,7 +484,7 @@ export function exploreMoves(
    *
    * Ordered nearest-first like every other card's, so it simply seeds the accumulator.
    */
-  const ownCut = cutTowards(startCard, carsOn(startCard), initialExit);
+  const ownCut = cutTowards(startCard, carsOn(startCard), rowEndAt(startCard, initialExit));
   const startKey = coordKey(start);
   const queue: Frontier[] = [
     {
@@ -502,12 +536,15 @@ export function exploreMoves(
      * overfill the tray is illegal, not a move that picks up fewer cars.
      *
      * NEAREST FIRST ALONG THE DIRECTION OF TRAVEL. `carsOn` runs west to east, so a train entering
-     * through the card's EAST port meets them back to front and the row has to be reversed. Without
-     * this the same parked cut produced an identical consist whichever way it was approached, when
-     * the two must mirror — which is the difference between a run-around being worth a Move and
-     * being pointless.
+     * at the row's EAST end meets them back to front and the row has to be reversed. Without this
+     * the same parked cut produced an identical consist whichever way it was approached, when the
+     * two must mirror — which is the difference between a run-around being worth a Move and being
+     * pointless.
+     *
+     * `rowEndAt` rather than `node.entry === 'e'`: a 45° leg is an end of the row too, and which
+     * end it is depends on the card's arc (Gitea#17).
      */
-    const met = node.entry === 'e' ? [...carsOn(card)].reverse() : carsOn(card);
+    const met = rowEndAt(card, node.entry) === 'e' ? [...carsOn(card)].reverse() : carsOn(card);
     const couples = [...node.couples, ...met];
     const nodeKey = coordKey(node.coord);
     const origins = [...node.origins, ...met.map(() => nodeKey)];
@@ -658,6 +695,26 @@ export function canPlaceAt(area: OfficeArea, coord: GridCoord, card: TrackCard):
   // into a stub and cutting the Office off from the Limits.
   if (coord.row === area.runningRow && !carriesThroughTrack(card)) return false;
 
+  /**
+   * ONE NEIGHBOUR MUST JOIN. THE OTHERS NEED NOT — AND THIS RULE HAS BEEN BOTH WAYS (Gitea#15).
+   *
+   * A card may be laid with an exit facing a card that has nothing to meet it. The rail stops dead
+   * at that edge, and that is legal.
+   *
+   * The issue was filed the other way round — "if a card is placed in that space, it MUST connect" —
+   * against a right-hand curve laid with its north leg against an Ice House and the turnout below it
+   * pointing at its portless south edge. **RAR reversed it on review (2026-08-26): placing it is
+   * fine, and a stub like that is useful — a siding to park cars on.**
+   *
+   * WHAT MATTERS INSTEAD IS THAT NOTHING CAN DRIVE ACROSS THE GAP, so the real requirement is on
+   * MOVEMENT rather than on placement: two cards touching are not connected, and `exploreMoves` must
+   * refuse the hop. It does — every step is gated on `joins`, never on a bare pair of `hasPort`
+   * calls — and `track.test.ts` pins the reported geometry against exactly that.
+   *
+   * SO DO NOT ADD A PER-EDGE CHECK HERE. One was written and taken out again when the ruling
+   * arrived. What survives is the weaker rule that was always here: the piece must touch the network
+   * SOMEWHERE, which is what stops orphaned track being laid in an empty corner of the board.
+   */
   const ports: Port[] = ['n', 's', 'e', 'w'];
   for (const p of ports) {
     const neighbourCard = cardAt(area, neighbour(coord, p));

src/server/session.ts

@@ -22,7 +22,7 @@ import { check } from '../engine/apply.ts';
 import { legalActions } from '../engine/legal.ts';
 import type { Intent } from '../engine/intents.ts';
 import type { GameConfig, PlayerIndex } from '../engine/state.ts';
-import { actionMenu, currentActor, fromMultiplayerSave, newMultiplayerGame, submit } from '../web/game.ts';
+import { actionMenu, currentActor, fromMultiplayerSave, isOutOfTurn, newMultiplayerGame, submit } from '../web/game.ts';
 import type { Game, Menu } from '../web/game.ts';
 import { deltaFrame } from '../sim/frame-delta.ts';
 import type { FrameDelta } from '../sim/frame-delta.ts';
@@ -262,6 +262,52 @@ function buildSession(
    * in zero wall-clock time by definition. Called once at construction (a resume could land exactly
    * on a bot's turn) and once after every accepted human intent.
    */
+/**
+   * §3.3, EXTENDED PLAY (Gitea#11) — the bots' half of a unanimous vote.
+   *
+   * "Bots will not disagree with the human. Humans get to vote first. If all humans vote yes, then
+   * bots vote yes too. If a human votes no, it's not unanimous, it ends right then. If only bots are
+   * playing, they never vote to extend" (Jesse, 2026-08-28).
+   *
+   * Which makes a bot's vote a formality rather than a policy decision, performed once the humans
+   * have already settled it, so that the unanimity the engine checks is a real unanimity rather than
+   * a special case carved into the rules for absent players.
+   *
+   * THE ALL-BOT TABLE IS THE CASE TO GET RIGHT, and getting it wrong hung the game. `driveBots`
+   * cannot reach the vote — it loops on `currentActor`, which is null the moment the game stops —
+   * so if this returns early with no humans to follow, nobody votes at all and a bot-only game sits
+   * on the question for ever. It happened: an all-bot session never reached `finished`. With nobody
+   * to follow, the bots' own answer stands, and it is no.
+   */
+  function driveBotVotes(): void {
+    if (game.state.status !== 'awaitingExtension') return;
+    const humans = [...Array(playerNames.length).keys()].filter((p) => !botSeats.has(p));
+    // A human who has voted `false` has already ended the game, so reaching here with humans still
+    // outstanding means the table is genuinely waiting on a person. Bots wait with it.
+    if (humans.length > 0 && !humans.every((p) => game.state.extensionVotes[p] === true)) return;
+    const agree = humans.length > 0;
+    for (const seat of botSeats) {
+      if (game.state.extensionVotes[seat] === null) {
+        submit(game, { type: 'game.extend', player: seat, agree }, seat);
+      }
+    }
+  }
+
+  /**
+   * Bots play, bots vote, and an agreed extension puts them back to playing — so the two drivers
+   * alternate rather than running once each. Bounded because every pass must consume something: a
+   * turn, or a vote that cannot be cast twice.
+   */
+  function driveBotTurns(): void {
+    for (let pass = 0; pass < 1_000; pass++) {
+      const before = game.history.length;
+      driveBots();
+      driveBotVotes();
+      if (game.history.length === before) return;
+    }
+    throw new Error('driveBotTurns: probable infinite loop');
+  }
+
   function driveBots(): void {
     let guard = 0;
     for (;;) {
@@ -281,7 +327,7 @@ function buildSession(
       settleTiming();
     }
   }
-  driveBots();
+  driveBotTurns();
   // Whatever the opening bot turns earned belongs to a game nobody was connected to yet — dropped
   // here rather than fired at the first client to arrive. (It also stops `game.cues` growing without
   // bound on a server, which nothing was draining before this.)
@@ -304,7 +350,17 @@ function buildSession(
       // never applied, so it is worth trying again (see the `lastSeq.set` below: only on success).
       if (lastSeq.get(seat) === seq) return { accepted: true, pushes: new Map(), timing: null };
 
-      if (seat !== currentActor(game)) return { accepted: false, code: 'NOT_YOUR_TURN' };
+      /**
+       * §3.3, EXTENDED PLAY (Gitea#11) — the vote is the one intent with no actor to be.
+       *
+       * `currentActor` is null once the timetable has run out, so this guard would refuse every
+       * vote with NOT_YOUR_TURN. Every seat may vote, and `check` is still the authority on whether
+       * this particular seat may vote right now (it has voted already; the game is not waiting on a
+       * vote at all), so skipping the turn test here gives nothing away.
+       */
+      if (!isOutOfTurn(i) && seat !== currentActor(game)) {
+        return { accepted: false, code: 'NOT_YOUR_TURN' };
+      }
 
       // Checked directly, rather than via `submit`'s boolean, for two reasons: `submit` writes a
       // "that is not allowed" line into the SHARED `game.log` on rejection, which would otherwise
@@ -315,7 +371,7 @@ function buildSession(
       if (code) return { accepted: false, code };
 
       const drawnBefore = game.justDrawn;
-      const applied = submit(game, i);
+      const applied = submit(game, i, isOutOfTurn(i) ? seat : null);
       if (game.justDrawn !== drawnBefore && game.justDrawn !== null) {
         lastDraw = { seat, cardId: game.justDrawn };
       }
@@ -327,8 +383,9 @@ function buildSession(
       const timing = settleTiming();
       // Any bot due to act now plays out entirely before this push goes back — the delta mechanism
       // diffs against whatever was last sent, so it captures the bots' moves along with the human's
-      // in one push regardless of how many turns that took.
-      driveBots();
+      // in one push regardless of how many turns that took. Votes included, since Gitea#11: a human
+      // agreeing to another Day is exactly the move the bots are waiting on to agree themselves.
+      driveBotTurns();
       return { accepted: true, pushes: pushesForAll(), timing };
     },
 
@@ -338,6 +395,8 @@ function buildSession(
         config: game.state.config,
         playerNames: [...playerNames],
         history: [...game.history],
+        // `awaitingExtension` is a game waiting on its table, not a game that is over — so it maps
+        // to 'active' and `server/index.ts` resumes it on a restart like any other (Gitea#11).
         status: game.state.status === 'finished' ? 'finished' : 'active',
         createdAt,
         botSeats: [...botSeats],
@@ -351,6 +410,8 @@ function buildSession(
         playerCount: playerNames.length,
         playerNames: [...playerNames],
         botSeats: [...botSeats],
+        // `awaitingExtension` is a game waiting on its table, not a game that is over — so it maps
+        // to 'active' and `server/index.ts` resumes it on a restart like any other (Gitea#11).
         status: game.state.status === 'finished' ? 'finished' : 'active',
         createdAt,
         lastMoveAt,

src/sim/board-svg.ts

@@ -65,7 +65,30 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
    * anything outside its own body.
    */
   const CW = { dp: 118, ml: 152, run: 78 };
-  const CH = 58;
+  /**
+   * TALL ENOUGH FOR TWO REGISTERS OF CHIPS, on every cell so the rail runs level across the row.
+   * Was 58, when a cell held one row of trains.
+   */
+  const CH = 76;
+  /**
+   * EVERY DISTRICT THE SAME WIDTH, sized for four chips two-by-two and NOT for its A/D count.
+   *
+   * Measured over 60 games: one office area holds at most 4 distinct trains, and up to 3 of those
+   * can be crews switching below the Running Track — which do not occupy A/D tracks at all. So a
+   * Whistle Post, with its single A/D track, can still have four trains to show, and sizing the cell
+   * by capacity would overflow it. Sizing by OCCUPANCY is worse still: that is what "The Roster
+   * Pass" fixed, because the cell then resizes as trains come and go and shoves the rest of the map
+   * sideways. A fixed two-by-two block holds the map still all game, upgrades included.
+   */
+  const OFFICE_W = 2 * 54 + 12;
+  /**
+   * THE VERTICAL ANATOMY OF A CELL, so the two chip registers and the rail cannot drift apart.
+   * The rail sits above centre; A/D chips straddle it, and the district register hangs below —
+   * which is where those trains are on the real board (Gitea#18).
+   */
+  const RAIL_Y = 34;
+  const CHIP_Y = RAIL_Y - 10;
+  const BELOW_Y = RAIL_Y + 13;
   const GAP = 6;
   /**
    * ONE FIXED SLOT PER A/D TRACK, so the Office Running Track cell is drawn wide enough to hold
@@ -115,6 +138,12 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
     seat: number | null;
     /** Set on an Office cell when a roster was supplied: whose district this is. */
     owner?: { name: string; isTurn: boolean; isYou: boolean } | null;
+    /**
+     * Office cells only: trains in the district that are NOT holding an A/D track — a crew switching
+     * below the Running Track, or a train standing on it away from the Office. Drawn in a second
+     * register under the rail (Gitea#18).
+     */
+    below?: Cell['trains'];
     /** Mainline cards only: §2.1 divides one into two regions. 0 elsewhere — no bars are drawn. */
     regions: number;
     w: number;
@@ -122,11 +151,8 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
     y: number;
   };
   const cells: Cell[] = [];
-  const sides: number[][] = [];
-  let side: number[] = [];
 
   const push = (c: Omit<Cell, 'x' | 'y'>): void => {
-    side.push(cells.length);
     cells.push({ ...c, x: 0, y: 0 });
   };
 
@@ -152,47 +178,54 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
           }
         : null;
 
-      for (const rc of n.running ?? []) {
-        const isOffice = rc.kind === 'office';
-        const adLabel = cap === null ? '' : `A/D ${ad.length}/${cap}`;
-        push({
-          kind: 'run',
-          label: isOffice && owner ? owner.name : rc.label,
-          owner: isOffice ? owner : null,
-          // With an owner on the headline the tier would otherwise vanish, so it joins the A/D
-          // count on the line below.
-          sub: isOffice ? (owner ? [rc.label, adLabel].filter(Boolean).join(' · ') : adLabel) : '',
-          /**
-           * A train standing at the Office occupies an A/D track, which is where it is — but it is
-           * ALSO standing on the Office grid card, so it arrives here in both lists and used to be
-           * drawn twice. Reported as two T10 chips on one Office.
-           */
-          trains: isOffice
-            ? [...rc.trains, ...ad.filter((t) => !rc.trains.some((r) => r.label === t.label))]
-            : rc.trains,
-          cap: isOffice ? cap : null,
-          tip: owner && isOffice
-            ? `${owner.name}'s ${rc.label}` +
-              (owner.isYou ? ' — this is your railroad' : '') +
-              // "their move" is wrong when the reader is the one being waited on.
-              (owner.isTurn ? (owner.isYou ? ' — it is your move' : ' — it is their move') : '')
-            : `${rc.label} — ${rc.kind === 'limits' ? 'the end of this district; the Running Track runs between the Limits' : 'Running Track'}`,
-          seat: n.seat ?? null,
-          // No regions inside a district: a crew moves by Moves there, not by Stages, so it
-          // occupies a card outright rather than a part of one.
-          regions: 0,
-          w: isOffice && cap !== null ? Math.max(CW.run, cap * CHIP_W + 12) : CW.run,
-        });
+      /**
+       * ONE CELL PER DISTRICT — NO OFFICE-AREA DETAIL ON THIS MAP (Gitea#18).
+       *
+       * An Office used to expand into its whole Running Track, Limits to Limits, so this map carried
+       * every straight, turnout, facility and Limits sign of every district. Two things were wrong
+       * with that. It is the OFFICE map's job, and it draws all of it properly, with the rails; and
+       * it made the Division map grow sideways as districts were built, shoving everything east of a
+       * district along every time somebody laid a card.
+       *
+       * TRAINS STAY. "Trains within the office area should definitely be represented on the division
+       * map" — at a glance the number and which way it is pointing, and the consist on the tooltip.
+       * They are split into two registers, because a train holding an A/D track and a crew switching
+       * in the district are not the same thing: A/D occupancy is a hard capacity that causes
+       * collisions, switching is not. The split is drawn as POSITION rather than colour — A/D on the
+       * rail, the rest below it — which is where those trains actually are.
+       */
+      const seen = new Set(ad.map((t) => t.label));
+      const below: typeof ad = [];
+      for (const t of [...(n.running ?? []).flatMap((rc) => rc.trains), ...(n.switching ?? [])]) {
+        if (seen.has(t.label)) continue;
+        seen.add(t.label);
+        below.push(t);
       }
-      // A crew below the Running Track has no position ON it, so it is reported against the
-      // district rather than drawn somewhere it is not.
-      const below = n.switching ?? [];
-      if (below.length > 0) {
-        const last = cells[cells.length - 1];
-        if (last) last.sub = `${below.length} switching below`;
-      }
-      sides.push(side);
-      side = [];
+      const adLabel = cap === null ? '' : `A/D ${ad.length}/${cap}`;
+      push({
+        kind: 'run',
+        label: owner ? owner.name : n.label,
+        owner,
+        sub: [owner ? n.label : '', adLabel, below.length > 0 ? `${below.length} switching` : '']
+          .filter(Boolean)
+          .join(' \u00b7 '),
+        trains: ad,
+        below,
+        cap,
+        tip:
+          (owner ? `${owner.name}'s ${n.label}` : n.label) +
+          (owner?.isYou ? ' — this is your railroad' : '') +
+          // "their move" is wrong when the reader is the one being waited on.
+          (owner?.isTurn ? (owner.isYou ? ' — it is your move' : ' — it is their move') : '') +
+          `\n\nThe district itself is drawn on the Office map — this cell is the whole of it, with the ` +
+          `trains standing in it: those holding an A/D track on the rail, and any crew switching in ` +
+          `the district below it.`,
+        seat: n.seat ?? null,
+        // No regions in a district: a crew moves by Moves there, not by Stages, so it occupies a
+        // card outright rather than a part of one.
+        regions: 0,
+        w: OFFICE_W,
+      });
       continue;
     }
     const dp = n.kind === 'dp';
@@ -229,58 +262,31 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
       w: dp ? CW.dp : CW.ml,
     });
   }
-  if (side.length > 0) sides.push(side);
 
-  // Each player's side carries their district and the Mainline card leading into it; whatever is
-  // left over (the last Mainline and the East DP) joins the final side.
-  const seats = Math.max(1, Math.min(4, nodes.filter((n) => n.kind === 'office').length));
-  const lanes: number[][] = [];
-  for (let i = 0; i < seats; i++) lanes.push([]);
-  sides.forEach((grp, i) => {
-    const target = Math.min(i, seats - 1);
-    for (const idx of grp) lanes[target]!.push(idx);
-  });
-
-  // -- lay the sides out around the table -------------------------------------------------------
-  // top → right → bottom (reversed) → left (reversed), which gives a row, two facing rows, a
-  // horseshoe open to the west, and a square broken at the same place.
-  const dir: ('top' | 'right' | 'bottom' | 'left')[] =
-    seats === 1 ? ['top'] : seats === 2 ? ['top', 'bottom'] : seats === 3 ? ['top', 'right', 'bottom'] : ['top', 'right', 'bottom', 'left'];
-
-  const runLen = (idxs: number[]): number =>
-    idxs.reduce((n, i) => n + cells[i]!.w + GAP, -GAP);
-  const widest = Math.max(...lanes.map((l) => runLen(l)), 200);
-  const tall = lanes.length > 1 ? Math.max(...lanes.map((l) => l.length), 1) * (CH + GAP) : CH;
-
-  const vertCount = dir.filter((d) => d === 'right' || d === 'left').length;
-  const boardW = PAD * 2 + widest + (vertCount > 0 ? CW.run + SIDE_GAP : 0);
-  const boardH = PAD * 2 + (dir.includes('bottom') ? CH * 2 + SIDE_GAP + (vertCount ? tall : 0) : CH) + 30;
-
-  lanes.forEach((idxs, i) => {
-    const d = dir[i]!;
-    if (d === 'top' || d === 'bottom') {
-      const y = d === 'top' ? PAD : boardH - PAD - CH - 22;
-      const order = d === 'bottom' ? [...idxs].reverse() : idxs;
-      let x = PAD;
-      for (const idx of order) {
-        const c = cells[idx]!;
-        c.x = x;
-        c.y = y;
-        x += c.w + GAP;
-      }
-    } else {
-      const x = d === 'right' ? boardW - PAD - CW.run : PAD;
-      const order = d === 'left' ? [...idxs].reverse() : idxs;
-      let y = PAD + CH + SIDE_GAP;
-      for (const idx of order) {
-        const c = cells[idx]!;
-        c.x = x;
-        c.y = y;
-        c.w = CW.run;
-        y += CH + GAP;
-      }
-    }
-  });
+  /**
+   * ONE ROW, WEST TO EAST (Gitea#18). The West Division Point is at the far left, the East at the
+   * far right, and nothing wraps.
+   *
+   * IT USED TO BE LAID OUT AROUND A TABLE — one row for a single seat, two facing rows for two, a
+   * horseshoe for three, a square for four — on the reasoning that players sit around a table so the
+   * route should too. That cost more than it bought, and three separate reports came out of it: the
+   * buffer stops pointed the wrong way once the route turned a corner, and, the one that decided it,
+   * **east stopped being to the right**. A player's east could be drawn south, west or north
+   * depending on which lane their district landed in, on a map whose whole job is saying which way
+   * a train is going.
+   *
+   * A row is wider than a square — roughly 1,580px at four players against 842 — and that is
+   * accepted: the map scrolls and zooms, and being able to rely on east meaning right is worth the
+   * scroll.
+   */
+  let x = PAD;
+  for (const c of cells) {
+    c.x = x;
+    c.y = PAD;
+    x += c.w + GAP;
+  }
+  const boardW = x - GAP + PAD;
+  const boardH = PAD * 2 + CH + 30;
 
   // -- draw -------------------------------------------------------------------------------------
   const rail = (x1: number, y: number, x2: number): string => {
@@ -295,20 +301,29 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
     return o;
   };
 
-  // The same rail turned through ninety degrees, for the sides of the table.
-  const railV = (x: number, y1: number, y2: number): string => {
-    let o =
-      `<line class="bs-rail" x1="${x - 2.5}" y1="${y1}" x2="${x - 2.5}" y2="${y2}"/>` +
-      `<line class="bs-rail" x1="${x + 2.5}" y1="${y1}" x2="${x + 2.5}" y2="${y2}"/>`;
-    const n = Math.max(2, Math.floor(Math.abs(y2 - y1) / 9));
-    for (let i = 0; i <= n; i++) {
-      const ty = y1 + ((y2 - y1) * i) / n;
-      o += `<line class="bs-tie" x1="${x - 4.5}" y1="${ty}" x2="${x + 4.5}" y2="${ty}"/>`;
-    }
-    return o;
-  };
-
-  let out = `<svg class="bs bs-div" viewBox="0 0 ${Math.ceil(boardW)} ${Math.ceil(boardH)}" preserveAspectRatio="xMinYMin meet">`;
+  /**
+   * DRAWN AT ITS OWN SIZE, SO IT SCROLLS RATHER THAN SHRINKING (Gitea#18).
+   *
+   * An SVG has a viewBox and a drawn size, and the browser scales one to the other. `.bs` is
+   * `width:100%`, so the map is drawn at whatever the panel is wide — which was harmless while the
+   * Division was 842px and wrapped around a table, and is not now that a single row is 1,580px. At
+   * that width in an 800px panel every label renders at half size, on the map that needs reading
+   * most. Setting the width to the viewBox width makes one unit one pixel, and the containers
+   * already scroll (`#division`, `#vdivision`).
+   *
+   * THE PLAYABLE PAGE DOES NOT NEED THIS — `applyZoom` (`main.ts`) sets exactly the same width from
+   * the same viewBox after every render, and overrides this when the zoom is not 100%. THE REPLAYS
+   * DO: neither `replays.ts` nor the standalone `replay.ts` calls it, so without this they get the
+   * `width:100%` shrink. It is inline rather than in `BOARD_CSS` because only this function knows
+   * how wide the row came out.
+   *
+   * `flex:none` because `#division` is a flex container and a flex item may be shrunk below an
+   * explicit width; there is no point pinning it and then letting the panel squeeze it anyway.
+   */
+  let out =
+    `<svg class="bs bs-div" viewBox="0 0 ${Math.ceil(boardW)} ${Math.ceil(boardH)}" ` +
+    `style="width:${Math.ceil(boardW)}px;flex:none" ` +
+    `preserveAspectRatio="xMinYMin meet">`;
 
   // The joins between consecutive cells, drawn as rail so a connection is rail meeting rail. A join
   // that crosses from one player's side to the next is drawn heavier and labelled: that boundary is
@@ -316,23 +331,7 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
   for (let i = 0; i + 1 < cells.length; i++) {
     const a = cells[i]!;
     const b = cells[i + 1]!;
-    const sameRow = Math.abs(a.y - b.y) < 1;
-    const sameCol = Math.abs(a.x - b.x) < 1;
-    if (sameRow && b.x > a.x) out += rail(a.x + a.w, a.y + CH / 2, b.x);
-    else if (sameRow && b.x < a.x) out += rail(b.x + b.w, a.y + CH / 2, a.x);
-    else if (sameCol) {
-      // Stacked down one side of the table: still one straight run of track, not a turn.
-      const top = Math.min(a.y + CH, b.y + CH);
-      const bot = Math.max(a.y, b.y);
-      out += railV(a.x + a.w / 2, top, bot);
-    } else {
-      // A turn between sides: an elbow, so the route is visibly continuous around the table.
-      const ax = a.x + a.w / 2;
-      const bx = b.x + b.w / 2;
-      const ay = a.y + CH;
-      const by = b.y;
-      out += `<path class="bs-turn" d="M${ax} ${ay} L${ax} ${(ay + by) / 2} L${bx} ${(ay + by) / 2} L${bx} ${by}"/>`;
-    }
+    out += rail(a.x + a.w, a.y + RAIL_Y, b.x);
   }
 
   cells.forEach((c) => {
@@ -348,15 +347,14 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
     const mark = c.owner ? ` bs-owner${c.owner.isTurn ? ' bs-turn' : ''}${c.owner.isYou ? ' bs-you' : ''}` : '';
     const suffix = c.owner?.isYou ? ' (you)' : '';
     out += `<text class="bs-name${mark}" x="${c.x + 7}" y="${c.y + 14}">${esc(c.label + suffix)}</text>`;
-    out += rail(c.x + 6, c.y + 32, c.x + c.w - 6);
+    out += rail(c.x + 6, c.y + RAIL_Y, c.x + c.w - 6);
     if (c.sub) out += `<text class="bs-cap" x="${c.x + 7}" y="${c.y + CH - 6}">${esc(c.sub)}</text>`;
 
-    // REGIONS. §2.1 divides a Mainline card into two, and §8.2 moves a train one region per Stage.
-    // The bars are the card's DISTANCE and never vary; what varies is how fast a train covers them,
-    // so a 60 card is crossed in one Stage and a slow train on a 30 takes three.
+    // REGIONS. A Mainline card is 1 to 3 of them (Gitea#3) and a train advances one per Stage. The
+    // bars are the card's DISTANCE and never vary; where a train STARTS is what does.
     const RW = c.regions > 0 ? (c.w - 12) / c.regions : 0;
     for (let r = 0; r < c.regions; r++) {
-      out += `<line class="bs-region" x1="${c.x + 6 + RW * r}" y1="${c.y + 20}" x2="${c.x + 6 + RW * r}" y2="${c.y + 44}"/>`;
+      out += `<line class="bs-region" x1="${c.x + 6 + RW * r}" y1="${c.y + RAIL_Y - 14}" x2="${c.x + 6 + RW * r}" y2="${c.y + RAIL_Y + 10}"/>`;
     }
 
     /**
@@ -371,36 +369,25 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
      * So this keeps the two things the Division map is actually for — where a train is and which way
      * it is going — and leaves the cars to the tooltip and to the district.
      */
-    c.trains.forEach((t, k) => {
+    /**
+     * A TRAIN IS A CHIP — its number, which way it points, and how many cars.
+     *
+     * It was drawn as a full consist here, matching the Office Area card, and reported as too large
+     * and hard to read. The Office card is where a consist is worth drawing, because that is where
+     * the switching decisions are made and where there is room to read it. So this keeps the two
+     * things the Division map is for — where a train is and which way it is going — and leaves the
+     * cars to the tooltip and to the district.
+     */
+    const chip = (t: NonNullable<Cell['trains']>[number], tx: number, ty: number, w: number): void => {
       const cars = t.cars ?? [];
       const arrow = t.facing === 'w' ? '\u25c0' : '\u25b6';
       const loaded = cars.filter((x) => /^loaded/.test(x) || /caboose/.test(x)).length;
       const label = cars.length === 0 ? `${t.label} ${arrow}` : `${t.label} ${arrow}${cars.length}`;
-
-      /**
-       * THE OFFICE RUNNING CELL GETS FIXED SLOTS, ONE PER A/D TRACK — never a centre spread.
-       *
-       * Centred spreading pushes its outer chips outward as MORE trains arrive, and the cell was
-       * sized for the cards it holds, not for its trains — so two chips at a Station used to land at
-       * x 215–267 and 271–316 inside a cell spanning only 230–308, spilling onto the Limits cards
-       * either side. A fixed slot per A/D track cannot overflow the cell at any occupancy, because
-       * the cell was sized for exactly that many slots (see `CHIP_W` above).
-       */
-      const isOfficeRun = c.kind === 'run' && c.cap !== null && c.cap > 0;
-      const slotW = isOfficeRun ? (c.w - 12) / c.cap! : 0;
-      const w = isOfficeRun ? Math.min(slotW - 4, label.length * 6.6 + 12) : Math.min(c.w - 8, label.length * 6.6 + 12);
-      // A train on a Mainline card sits in ITS region; anywhere else it just sits on the card.
       const inRegion = c.regions > 1 && typeof t.region === 'number';
-      const tx = isOfficeRun
-        ? c.x + 6 + slotW * (k + 0.5)
-        : (inRegion ? c.x + 6 + RW * (t.region ?? 0) + RW / 2 : c.x + c.w / 2) +
-          (inRegion ? 0 : (k - (c.trains.length - 1) / 2) * (w + 4));
       const dir = t.direction === 'west' ? ' \u25c0 west' : t.direction === 'east' ? ' east \u25b6' : '';
       const stages =
         typeof t.stagesLeft === 'number'
-          ? ` \u00b7 ${t.stagesLeft} Stage${t.stagesLeft === 1 ? '' : 's'} still to run across this card` +
-            ' (Stages, not regions: a card is two regions of fixed distance, and how many Stages a' +
-            ' train takes over them depends on the card speed and the train)'
+          ? ` \u00b7 ${t.stagesLeft} Stage${t.stagesLeft === 1 ? '' : 's'} still to run across this card`
           : '';
       out += `<g class="bs-train" data-tip="${esc(t.label)} \u2014 carrying ${esc(cars.join(', ') || 'no cars')}${
         cars.length ? ` (${loaded} loaded)` : ''
@@ -409,15 +396,61 @@ export function divisionSvg(nodes: DivisionView[], roster?: DivisionRoster | nul
         // me?" gets asked, and EXPEDITED is the answer more often than not.
         t.what ? `\n\n${esc(t.what)}` : ''
       }">` +
-        `<rect x="${tx - w / 2}" y="${c.y + 22}" width="${w}" height="19" rx="3"/>` +
-        `<text class="bs-tlab" x="${tx}" y="${c.y + 35}" text-anchor="middle">${esc(label)}</text>`;
+        `<rect x="${tx - w / 2}" y="${ty}" width="${w}" height="19" rx="3"/>` +
+        `<text class="bs-tlab" x="${tx}" y="${ty + 13}" text-anchor="middle">${esc(label)}</text>`;
       out += '</g>';
+    };
+
+    const textW = (t: NonNullable<Cell['trains']>[number]): number =>
+      (`${t.label} \u25b6${(t.cars ?? []).length || ''}`).length * 6.6 + 12;
+
+    /**
+     * TWO CHIPS TO A REGISTER ON A DISTRICT, in fixed slots — never a centre spread.
+     *
+     * Centred spreading pushes its outer chips outward as more trains arrive, which is how two chips
+     * at a Station once landed outside the cell that held them. Fixed slots cannot overflow, because
+     * the cell was sized for exactly that many (`OFFICE_W`).
+     */
+    const isDistrict = c.kind === 'run';
+    const SLOTS = 2;
+    const slotW = (c.w - 12) / SLOTS;
+    c.trains.forEach((t, k) => {
+      if (isDistrict) {
+        // Row-major within the A/D register: two across, then wrap under. A district can hold four
+        // trains and only two fit across it.
+        const col = k % SLOTS;
+        const row = Math.floor(k / SLOTS);
+        chip(t, c.x + 6 + slotW * (col + 0.5), c.y + CHIP_Y + row * 21, Math.min(slotW - 4, textW(t)));
+        return;
+      }
+      // A train on a Mainline card sits in ITS region; anywhere else it just sits on the card.
+      const inRegion = c.regions > 1 && typeof t.region === 'number';
+      const w = Math.min(c.w - 8, textW(t));
+      const tx = (inRegion ? c.x + 6 + RW * (t.region ?? 0) + RW / 2 : c.x + c.w / 2) +
+        (inRegion ? 0 : (k - (c.trains.length - 1) / 2) * (w + 4));
+      chip(t, tx, c.y + CHIP_Y, w);
     });
+
+    /**
+     * THE SECOND REGISTER, under the rail: trains in the district that hold no A/D track (Gitea#18).
+     *
+     * A crew switching below the Running Track and a train standing at an A/D track are different
+     * things — A/D occupancy is a hard capacity that causes collisions, switching is not — and the
+     * difference is drawn as POSITION rather than as a colour to learn, because below the rail is
+     * where those trains actually are.
+     */
+    (c.below ?? []).forEach((t, k) => {
+      const col = k % SLOTS;
+      const row = Math.floor(k / SLOTS);
+      chip(t, c.x + 6 + slotW * (col + 0.5), c.y + BELOW_Y + row * 21, Math.min(slotW - 4, textW(t)));
+    });
+
     out += '</g>';
   });
 
-  // THE ENDS. The route stops at both Division Points; drawing buffer stops and naming the gap is
-  // what stops a seated layout being read as a loop.
+  // THE ENDS. The route stops at both Division Points, and the buffer stops say so — a Division is
+  // a LINE, not a loop. With a single row (Gitea#18) they simply face outward at the two ends, west
+  // on the left and east on the right, which is the bug reported twice against the wrapped layout.
   const first = cells[0];
   const last = cells[cells.length - 1];
   /**

src/sim/bot.ts

@@ -138,6 +138,19 @@ export function makeDeveloperBot(tweaks: BotTweaks): BotPolicy {
 
   choose(s, player, options) {
     lastReason = 'no specific reason — first legal option';
+
+    /**
+     * §3.3, EXTENDED PLAY (Gitea#11) — a bot never asks for another Day.
+     *
+     * "If only bots are playing, they never vote to extend" (Jesse, 2026-08-28), which is what keeps
+     * the balance harness and every bot-only game ending at the timetable it was dealt with. It also
+     * makes this the SAFE DEFAULT everywhere else: a bot's agreement in a game with humans in it is
+     * decided by `server/session.ts`, which votes on the bots' behalf only once every human has
+     * already said yes, and never reaches this policy at all.
+     */
+    const extend = options.find((i) => i.type === 'game.extend' && i.agree === false);
+    if (extend) return because('a bot plays the timetable it was dealt and no more', extend);
+
     const clearance = ruleOnClearance(options);
     if (clearance) return because('the Superintendent must rule on a following train', clearance);
 
@@ -1861,6 +1874,36 @@ export function playGame(
     tally(pumpFn(s));
     if (s.status === 'finished') break;
 
+    /**
+     * §3.3, EXTENDED PLAY (Gitea#11) — a simulated game plays the timetable it was dealt.
+     *
+     * DECIDED BY THE DRIVER, not by the policy, and that distinction is the whole point. A bot that
+     * is merely handed the two votes among its legal options will sometimes take another Day —
+     * `randomBot` does so half the time — and since the table can go on granting Days for ever, the
+     * game then runs until `maxTurns`. That is not a hypothetical: it turned `test/sim.test.ts` from
+     * under a second into an unbounded hang, because every seeded game in the harness suddenly played
+     * fifty thousand turns instead of two hundred.
+     *
+     * The harness exists to measure games of a configured length against a configured floor, so
+     * "would you like more Days?" has one answer here whatever the policy. `developerBot` declines on
+     * its own account too, which is what the server relies on when a table is all bots; this is the
+     * guarantee that holds for every OTHER policy, including ones not written yet.
+     */
+    if (s.status === 'awaitingExtension') {
+      const voter = s.extensionVotes.findIndex((v) => v === null);
+      if (voter < 0) break;
+      const decline: Intent = { type: 'game.extend', player: voter, agree: false };
+      intents.push(decline.type);
+      history.push(decline);
+      const declined = applyIntent(s, voter, decline);
+      // A broken invariant, not a game ending early: the status says a vote is pending and `voter` is
+      // a seat that has not cast one. Thrown rather than broken out of, matching the illegal-action
+      // check below — silently returning a short game is how a dead replay looks like a real one.
+      if (!declined.ok) throw new Error(`the extension vote was refused with ${declined.code}`);
+      tally(declined.events);
+      continue;
+    }
+
     const actor =
       s.clock.pendingDecision !== null ? s.clock.superintendent : s.clock.currentActor;
     if (actor === null) break;

src/sim/narrate.ts

@@ -488,6 +488,16 @@ export function narrate(e: GameEvent, ctx: NarrateContext = {}): Narration {
         : { tone: 'good', text: `+${e.delta} Revenue (now ${e.total}) — ${e.reason}` };
     case 'phaseEnded':
       return { tone: 'quiet', text: `Player ${e.player} finished ${phaseLabel(e.phase)}` };
+
+    // -- §3.3, extended play (Gitea#11)
+    case 'extensionVoted':
+      return e.agree
+        ? { tone: 'plain', text: `Player ${e.player} would play one more Day` }
+        : { tone: 'plain', text: `Player ${e.player} called time — the game ends here` };
+    case 'dayExtended':
+      return { tone: 'clock', text: `── The table plays on: Day ${e.day} is added to the timetable ──` };
+    case 'playConcluded':
+      return { tone: 'clock', text: '── The railroad is put to bed. Final results stand. ──' };
   }
 }
 

src/sim/save-replay.ts

@@ -52,6 +52,21 @@ export type PlayedGame = {
 export function playForReplay(seed: number, policy: BotPolicy, maxTurns = 50_000): PlayedGame {
   const game = newGame(seed);
   for (let t = 0; t < maxTurns; t++) {
+    /**
+     * §3.3, EXTENDED PLAY (Gitea#11) — a recorded replay is a game played to its end.
+     *
+     * The timetable running out leaves the game on "play one more Day?", where `currentActor` is
+     * null and this loop would otherwise stop — recording a file that replays to a question nobody
+     * answered rather than to a finished game. A recording bot plays the timetable it was dealt, the
+     * same rule `playGame` follows, so it declines and the file ends where a real game would.
+     */
+    if (game.state.status === 'awaitingExtension') {
+      const voter = game.state.extensionVotes.findIndex((v) => v === null);
+      if (voter < 0) break;
+      if (!submit(game, { type: 'game.extend', player: voter, agree: false }, voter)) break;
+      continue;
+    }
+
     const actor = currentActor(game);
     if (actor === null) break;
     const options = legalActions(game.state, actor);

src/sim/view.ts

@@ -10,6 +10,7 @@
  * drift into two different pictures of the same board.
  */
 
+import { regionOfTransit } from '../engine/advance.ts';
 import {
   areaAtSeat,
   areaOf,
@@ -32,7 +33,6 @@ import {
   MANEUVER_CARDS,
   MODIFIER_PROFILES,
   REALIGNMENTS,
-  REGIONS_PER_MAINLINE_CARD,
   OFFICE_ORDER,
   SPACE_USE_CARDS,
   enhancementRule,
@@ -398,6 +398,24 @@ export type Frame = {
   collisionsTotal: number;
   status: GameState['status'];
   outcome: GameState['outcome'];
+  /**
+   * §3.3, EXTENDED PLAY (Gitea#11). `days` above stays the ORIGINAL timetable — it is what the
+   * official result was decided at — so the Day the game now runs to is `days + extraDays`.
+   */
+  extraDays: number;
+  /** Per PLAYER, while `status` is `awaitingExtension`. `null` is a seat that has not voted. */
+  extensionVotes: (boolean | null)[];
+  /** The official result, frozen when the original timetable ran out. Null until then. */
+  official: GameState['official'];
+  /**
+   * Gitea#16 — everything interesting that has happened, folded from the event stream.
+   *
+   * Aggregate counts only, which is why it can ride the Frame at all: `test/redaction.test.ts`
+   * proves a Frame carries no other seat's secrets, and a count of trains is nobody's secret. Being
+   * here rather than on a side channel is what gets the results screen the same numbers in
+   * multiplayer as in solitaire, from one implementation.
+   */
+  tally: GameState['tally'];
   /**
    * Every PLAYER's public standing — names and Revenue. "The race is the game" (protocol.md §4).
    *
@@ -1026,6 +1044,12 @@ export function describeIntent(s: GameState, i: Intent): string {
     }
     case 'maneuver.redFlags':
       return `set Red Flags to protect ${trainName(s, i.trayId)} — an approaching train must stop short`;
+    // §3.3, extended play (Gitea#11). The results screen draws its own buttons, but a bot reads its
+    // options through this list like any other, and the label is what the history says it chose.
+    case 'game.extend':
+      return i.agree
+        ? 'play one more Day — the result already recorded still stands'
+        : 'end the game here';
     case 'maneuver.flyingSwitch':
       return `Flying Switch ${i.count} car(s) into ${at(i.to)}`;
     case 'mainline.clearance': {
@@ -1074,6 +1098,12 @@ export function describeIntent(s: GameState, i: Intent): string {
       return 'play your red flag';
     case 'maneuver.redFlags':
       return `set Red Flags to protect ${trainName(s, i.trayId)} — an approaching train must stop short`;
+    // §3.3, extended play (Gitea#11). The results screen draws its own buttons, but a bot reads its
+    // options through this list like any other, and the label is what the history says it chose.
+    case 'game.extend':
+      return i.agree
+        ? 'play one more Day — the result already recorded still stands'
+        : 'end the game here';
     default: {
       // Every Intent now has a sentence, so `i` narrows to never here. Keeping the assignment makes
       // that a COMPILE error the day someone adds an intent without describing it — the playable UI
@@ -1165,36 +1195,27 @@ export function snapshot(
       // Crossing time is in Stages now, so a Mainline card shows its terrain and the trains on it
       // with how long each still has to run.
       const name = MAINLINE_PROFILES.find((m) => m.kind === n.card)?.name ?? n.card;
-      const isGrade = MAINLINE_PROFILES.find((m) => m.kind === n.card)?.speed.kind === 'grade';
+      const isGrade = n.card === 'heavyGrade';
       /**
-       * WHERE ON THE CARD, from what the crossing already cost.
+       * WHERE ON THE CARD — now simply what the card says.
        *
-       * §2.1 divides a Mainline card into two regions and §8.2 moves a train one region per Stage.
-       * The engine crosses in `crossingStages` Stages instead, which varies by card speed, train
-       * speed, passengers and modifiers — so the printed model is recovered by treating the entry
-       * point as the thing that varies, exactly as the cards do:
+       * This used to recover a printed two-region model from a crossing time computed out of the
+       * card's mph, the train's Fast/Slow class, its consist and any modifiers, by treating the
+       * ENTRY point as the thing that varied: `entry = 2 - stagesTotal`. It even had to cope with a
+       * negative entry, for a slow train needing three Stages to cross a card with two regions.
        *
-       *     entry = REGIONS - stagesTotal        position = entry + elapsed
-       *
-       * A 60 card is one Stage, so the train enters at the second region and is gone — which is
-       * what "Start positions further along the card" means on the printed art. A 30 card is two
-       * Stages, giving one region per Stage, which is §8.2 exactly. A slow train needing three
-       * Stages cannot fit three steps into two regions, so it holds in the first for a Stage: the
-       * card's distance is fixed and the train is simply slow across it.
+       * Gitea#3 turned that the right way up. Regions are the primary thing — printed on the card,
+       * one per Stage — and the entry point is what the rules actually move. There is nothing left
+       * to reconstruct.
        */
-      const place = (t: { stagesRemaining: number; stagesTotal: number }): number => {
-        // `entry` may be NEGATIVE — a slow train needing three Stages cannot fit three steps into
-        // two regions, so it notionally starts before the card and spends the extra Stage getting
-        // to the first region. Clamping only the final position keeps that Stage at the START,
-        // where being slow shows; clamping `entry` first would have parked it at the exit instead.
-        const entry = REGIONS_PER_MAINLINE_CARD - t.stagesTotal;
-        const elapsed = t.stagesTotal - t.stagesRemaining;
-        return Math.min(REGIONS_PER_MAINLINE_CARD - 1, Math.max(0, entry + elapsed));
-      };
+      // One region per Stage, straight off the card's own count: what a train has LEFT to run says
+      // where it is standing. `regionOfTransit` is the engine's own answer, so the picture and the
+      // collision rule cannot disagree about who is where.
+      const place = (t: { stagesRemaining: number }): number => regionOfTransit(n.card, t.stagesRemaining);
       return {
         kind: 'ml',
         label: name,
-        regions: REGIONS_PER_MAINLINE_CARD,
+        regions: mainlineProfile(n.card).regions,
         trains: [n.transits.map((t) => {
           const chip = trainChip(s, t.tray);
           return {
@@ -1344,6 +1365,10 @@ export function snapshot(
     collisionsTotal: s.collisionsTotal,
     status: s.status,
     outcome: s.outcome,
+    extraDays: s.extraDays,
+    extensionVotes: [...s.extensionVotes],
+    official: s.official,
+    tally: s.tally,
     players: s.players.map((p) => ({
       index: p.index,
       seat: seatOf(s, p.index),
@@ -1788,7 +1813,16 @@ const SIMPLE_CARDS = [
  * view of its own. `0` means no floor is configured — nothing to pace against.
  */
 function objectiveOf(s: GameState, viewer: PlayerIndex): Frame['objective'] {
-  const { days, minCombinedRevenue: target } = s.config;
+  const { minCombinedRevenue: target } = s.config;
+  /**
+   * PACED AGAINST THE TIMETABLE ACTUALLY BEING PLAYED, extensions included (Gitea#11).
+   *
+   * `config.days` alone would say "the last Day is over" through every extended Day, and pace an
+   * eight-Day game against five — both of which the status line used to do the moment play carried
+   * on past the end. The official result is still decided at `config.days`; that is `checkVictory`'s
+   * business, and nothing here feeds it.
+   */
+  const days = s.config.days + s.extraDays;
   const revenue = s.players[viewer]?.revenue ?? 0;
   const daysLeft = Math.max(0, days - s.clock.day + 1);
   const elapsed = days - daysLeft + 1;

src/web/game.ts

@@ -1014,9 +1014,42 @@ export function overHandLimit(game: Game, seat: PlayerIndex = 0): boolean {
   return hand.length > limit;
 }
 
-/** Submit an action. Returns false and changes nothing if the engine rejects it. */
-export function submit(game: Game, intent: Intent): boolean {
-  const actor = currentActor(game);
+/**
+ * Intents any seat may send regardless of whose turn it is — and, for `game.extend`, regardless of
+ * whether the game is still running at all (§3.3, Gitea#11).
+ *
+ * `currentActor` is null once the timetable has run out, which is correct for everything else and
+ * exactly wrong for the vote on playing another Day. Rather than teach `currentActor` about a state
+ * where EVERY seat may act at once — which it has no way to express — callers name the seat.
+ */
+export function isOutOfTurn(intent: Intent): intent is Extract<Intent, { type: 'game.extend' }> {
+  return intent.type === 'game.extend';
+}
+
+/**
+ * WHO ACTED, when replaying a saved history.
+ *
+ * A save is a flat `Intent[]` with no seat written beside each move, so a replay normally derives
+ * the actor from the turn order — the same order the live game went round in, reproduced exactly.
+ * That breaks for exactly one intent: the extension vote, which every seat may cast in any order,
+ * and which `currentActor` answers `null` for because the game has stopped. Replaying such a save
+ * used to fail outright with `NO_ACTOR`, which is to say an extended game could not be resumed at
+ * all — found by `test/server/session.test.ts`'s resume test, and the reason `game.extend` carries
+ * its voter (`intents.ts`).
+ */
+function replayActor(game: Game, intent: Intent): PlayerIndex | null {
+  return isOutOfTurn(intent) ? intent.player : currentActor(game);
+}
+
+/**
+ * Submit an action. Returns false and changes nothing if the engine rejects it.
+ *
+ * `as` names the seat for an out-of-turn intent (see `isOutOfTurn`). It cannot be used to smuggle an
+ * ordinary move past the turn order: `check` is still the authority and still asks `isActor`, so a
+ * named seat that is not the actor is refused exactly as it would have been.
+ */
+export function submit(game: Game, intent: Intent, as: PlayerIndex | null = null): boolean {
+  const actor = as ?? currentActor(game);
   if (actor === null) return false;
 
   const result = applyIntent(game.state, actor, intent);
@@ -1167,7 +1200,7 @@ export function undo(game: Game, config: GameConfig = game.state.config): Game |
 export function fromSave(save: Save, config: GameConfig = SOLO_CONFIG): Game {
   const game = newGame(save.seed, configFor(save, config));
   for (const intent of save.history) {
-    const actor = currentActor(game);
+    const actor = replayActor(game, intent);
     if (actor === null) break;
     const result = applyIntent(game.state, actor, intent);
     if (!result.ok) break;
@@ -1218,7 +1251,7 @@ export function fromMultiplayerSave(
 ): { game: Game; stopped: ReplayStop | null } {
   const game = newMultiplayerGame(seed, config, playerNames);
   for (const [index, intent] of history.entries()) {
-    const actor = currentActor(game);
+    const actor = replayActor(game, intent);
     if (actor === null) {
       return { game, stopped: { index, intent, code: 'NO_ACTOR' } };
     }

src/web/main.ts

@@ -10,7 +10,7 @@ import { TURNCHART_CSS, turnChartHtml } from '../sim/turnchart.ts';
 import type { Frame } from '../sim/view.ts';
 import { seatLabel } from '../sim/view.ts';
 import type { Menu, Save } from './game.ts';
-import { PANEL_CSS, blockedHtml, dayEndHtml, facilitiesHtml, pilesHtml, timetableHtml, yardHtml } from './panels.ts';
+import { PANEL_CSS, blockedHtml, dayEndHtml, facilitiesHtml, pilesHtml, resultsHtml, timetableHtml, yardHtml } from './panels.ts';
 import { TOOLTIP_CSS, installTooltips } from './tooltip.ts';
 import { playCue } from './sound.ts';
 import {
@@ -916,7 +916,16 @@ function render(): void {
    * figure came from a target that is itself an open question.
    */
   const obj = $('objective');
-  obj.textContent = `${f.revenue} of ${f.objective.target} · ${f.objective.daysLeft} Day${f.objective.daysLeft === 1 ? '' : 's'} left`;
+  /**
+   * Past the original timetable this has to stop saying "N Days left" of a Day count that no longer
+   * applies (Gitea#11). `objective.daysLeft` already counts against the EXTENDED timetable; what it
+   * cannot say on its own is that the Days being counted are borrowed ones.
+   */
+  const left = `${f.objective.daysLeft} Day${f.objective.daysLeft === 1 ? '' : 's'} left`;
+  obj.textContent =
+    f.extraDays > 0
+      ? `${f.revenue} · Day ${f.day} — ${f.extraDays} beyond the timetable`
+      : `${f.revenue} of ${f.objective.target} · ${left}`;
   obj.className = 'pace';
   // The seed is never sent to a remote client at all (it would leak every future shuffle and roll,
   // `multiplayer.md` §7) — `RemoteSession` has no `.seed()` because there is nothing to return.
@@ -1392,6 +1401,114 @@ function wirePointing(node: HTMLElement, key: string, routeKeys: readonly string
   node.onblur = off;
 }
 
+/**
+ * THE END OF THE GAME — the action area once there is nothing left to decide, or only one thing.
+ *
+ * Two states share this, and they are genuinely different (Gitea#11):
+ *
+ *   - `awaitingExtension` — the timetable ran out on an ending the table MAY play past. The result
+ *     is already recorded and already readable; the only question open is whether to run one more
+ *     Day. In multiplayer that is a unanimous vote, so this also has to show who is still to answer.
+ *   - `finished` — over for good. The results screen, and a new game.
+ *
+ * The results are reachable in BOTH, and stay reachable after play continues, which is the
+ * constraint Gitea#16 and Gitea#11 put on each other: continuing must not cost you the results
+ * screen, so it is a button that reopens rather than a screen you get one look at.
+ */
+function renderEnding(el: HTMLElement, f: Frame): void {
+  const o = f.official?.outcome ?? f.outcome;
+  const won = o?.result === 'win';
+
+  /**
+   * Put the results up once per ending, unasked.
+   *
+   * "Once per ENDING" rather than once per game is the extended-play case: an extended game ends,
+   * is played on, and ends again, and each of those is a moment worth reading. `renderActions`
+   * clears the flag whenever the game is running again, so the next ending gets its own showing —
+   * while a redraw during the same ending does not reopen a dialog the player has dismissed.
+   */
+  if (!resultsShown) {
+    resultsShown = true;
+    showResults(f);
+  }
+
+  if (f.status === 'awaitingExtension') {
+    const mine = f.extensionVotes[f.viewer];
+    // Only seats that exist are counted; `extensionVotes` is per PLAYER and so is `players`.
+    const waiting = f.players.filter((p) => f.extensionVotes[p.index] === null);
+    const tally = f.players.length > 1
+      ? '<div class="vote-tally">' +
+        f.players
+          .map((p) => {
+            const v = f.extensionVotes[p.index];
+            const mark = v === true ? '✓' : v === false ? '✗' : '·';
+            return `<span class="vote ${v === true ? 'yes' : v === false ? 'no' : 'wait'}">` +
+              `${mark} ${esc(p.name)}${p.index === f.viewer ? ' (you)' : ''}</span>`;
+          })
+          .join('') +
+        '</div>'
+      : '';
+
+    el.innerHTML =
+      `<div class="over ${won ? 'win' : 'loss'}">` +
+      `${won ? 'THE DIVISION RAN' : 'THE DIVISION FAILED'} — Day ${f.official?.day ?? f.days} is scored.<br>` +
+      'The result above is final. Play one more Day?</div>' +
+      tally +
+      (mine === null
+        ? '<button id="extend-yes">play one more Day</button>' +
+          '<button id="extend-no">end the game here</button>'
+        : `<div class="dim">You voted ${mine ? 'to play on' : 'to end it'}. ` +
+          (waiting.length
+            ? `Waiting on ${waiting.map((p) => esc(p.name)).join(', ')}.`
+            : 'Settling…') +
+          '</div>') +
+      '<button id="results">see the full results</button>';
+
+    if (mine === null) {
+      const vote = (agree: boolean) => () =>
+        void session.submit({ type: 'game.extend', player: f.viewer, agree });
+      $('extend-yes').onclick = vote(true);
+      $('extend-no').onclick = vote(false);
+    }
+    $('results').onclick = () => showResults(f);
+    return;
+  }
+
+  el.innerHTML =
+    `<div class="over ${won ? 'win' : 'loss'}">` +
+    `${won ? 'THE DIVISION RAN' : 'THE DIVISION FAILED'}<br>` +
+    `final Revenue ${f.revenue}${f.objective.target > 0 ? ` against a target of ${f.objective.target}` : ''}</div>` +
+    '<button id="results">see the full results</button>' +
+    (session.capabilities.newGame ? '<button id="again">new game</button>' : '');
+
+  $('results').onclick = () => showResults(f);
+  if (session.capabilities.newGame) {
+    $('again').onclick = () => {
+      clearSave();
+      location.search = '';
+    };
+  }
+}
+
+/**
+ * Whether the results have been put up by themselves for the ending currently on screen.
+ *
+ * Cleared by `renderActions` the moment the game is running again, so an extended game gets a fresh
+ * showing at each of its endings while a redraw during one ending does not reopen a dialog the
+ * player has just dismissed.
+ */
+let resultsShown = false;
+
+function showResults(f: Frame): void {
+  const dlg = document.getElementById('resultsdlg') as HTMLDialogElement | null;
+  const body = document.getElementById('resultsbody');
+  if (!dlg || !body) return;
+  body.innerHTML = resultsHtml(f);
+  // A redraw can arrive while it is open — `showModal` throws on an already-open dialog rather
+  // than doing nothing (the same trap `noteDayEnd` documents).
+  if (!dlg.open) dlg.showModal();
+}
+
 function renderActions(
   menu: Menu,
   f: Frame,
@@ -1400,18 +1517,12 @@ function renderActions(
   const el = $('actions');
 
   if (f.status !== 'active') {
-    const o = f.outcome;
-    el.innerHTML =
-      `<div class="over ${o?.result === 'win' ? 'win' : 'loss'}">` +
-      `${o?.result === 'win' ? 'YOU WIN' : 'GAME OVER'} — ${esc(String(o?.reason ?? ''))}<br>` +
-      `final Revenue ${f.revenue} against a target of ${f.objective.target}</div>` +
-      `<button id="again">new game</button>`;
-    $('again').onclick = () => {
-      clearSave();
-      location.search = '';
-    };
+    renderEnding(el, f);
     return;
   }
+  // The game is running, so the next ending — an extended Day's, or a fresh game's — is entitled to
+  // put its results up unasked again (Gitea#11).
+  resultsShown = false;
 
   if (menu.direct.length === 0 && menu.placeable.length === 0) {
     el.innerHTML = '<div class="dim">nothing to decide — the engine is running the Division</div>';

src/web/panels.ts

@@ -168,52 +168,319 @@ export function timetableHtml(f: Frame, justSet: number | null): string {
  */
 export function dayEndHtml(f: Frame): string {
   const ended = f.day - 1;
-  const left = f.days - ended;
-  const standings = [...f.players]
-    .sort((a, b) => b.revenue - a.revenue || a.seat - b.seat)
-    .map(
-      (p) =>
-        `<tr${p.index === f.viewer ? ' class="you"' : ''}><td>${esc(p.name)}` +
-        `${p.index === f.viewer ? ' <span class="dim">(you)</span>' : ''}</td>` +
-        `<td class="num">${p.revenue}</td></tr>`,
-    )
-    .join('');
-
-  // The target is a COMBINED floor in every mode that sets one, so it is reported against the whole
-  // table's Revenue rather than the viewer's — showing one player's score against a four-player
-  // target reads as a hopeless position when the table may be comfortably ahead.
-  const combined = f.players.reduce((n, p) => n + p.revenue, 0);
-  const target =
-    f.minCombinedRevenue > 0
-      ? `<p>Combined Revenue <b>${combined}</b> against a target of <b>${f.minCombinedRevenue}</b>.</p>`
-      : '';
-
-  /**
-   * Only when the game is actually scored on collisions.
-   *
-   * Two conditions, both of them `advance.ts`'s own: `0` on a dial turns that check off, and the
-   * checks run in COMPETITIVE AND CO-OP ONLY (§3.4). A solitaire game carries the default dials on
-   * its config and enforces neither, so reporting a collision budget there would put a rule on
-   * screen that this game does not have.
-   */
-  const scoredOnCollisions =
-    (f.mode === 'competitive' || f.mode === 'coop') &&
-    (f.maxCollisionsTotal > 0 || f.maxCollisionsPerDay > 0);
-  const collisions = scoredOnCollisions
-    ? `<p>Collisions: <b>${f.collisionsToday}</b> today, <b>${f.collisionsTotal}</b> in all.</p>`
-    : '';
+  const left = f.days + f.extraDays - ended;
 
   const ahead =
     left <= 0
       ? '<p>That was the last Day on the timetable.</p>'
-      : `<p><b>Day ${f.day} of ${f.days}</b> begins now — ${left} ${left === 1 ? 'Day' : 'Days'} left to run.</p>`;
+      : `<p><b>Day ${f.day} of ${f.days + f.extraDays}</b> begins now — ${left} ${left === 1 ? 'Day' : 'Days'} left to run.</p>`;
 
   return (
     `<h3 class="dayend-h">Day ${ended} has ended</h3>` +
     ahead +
-    `<table class="dayend-t"><tbody>${standings}</tbody></table>` +
-    target +
-    collisions
+    standingsHtml(f) +
+    targetHtml(f) +
+    collisionsHtml(f)
+  );
+}
+
+// ---------------------------------------------------------------------------
+// Shared between the Day-end dialog and the end-of-game results screen.
+//
+// Gitea#16 asked for the results screen and Gitea#10's dialog had already assembled most of it. The
+// three blocks below are the overlap, factored out rather than written twice: the two screens report
+// the same numbers about the same game, and the one thing they must never do is disagree.
+// ---------------------------------------------------------------------------
+
+/**
+ * Every player in Revenue order, the viewer marked.
+ *
+ * `winner` rings the player the OFFICIAL result named, which is not always the player at the top:
+ * in an extended game the standings keep moving after the result is settled, and showing the leader
+ * without saying who actually won would be the screen contradicting itself.
+ */
+function standingsHtml(f: Frame, winner: number | null = null): string {
+  const rows = [...f.players]
+    .sort((a, b) => b.revenue - a.revenue || a.seat - b.seat)
+    .map((p) => {
+      const marks =
+        (p.index === f.viewer ? ' <span class="dim">(you)</span>' : '') +
+        (p.index === winner ? ' <span class="wins">— winner</span>' : '');
+      return (
+        `<tr${p.index === f.viewer ? ' class="you"' : ''}><td>${esc(p.name)}${marks}</td>` +
+        `<td class="num">${p.revenue}</td></tr>`
+      );
+    })
+    .join('');
+  return `<table class="dayend-t"><tbody>${rows}</tbody></table>`;
+}
+
+/**
+ * The target is a COMBINED floor in every mode that sets one, so it is reported against the whole
+ * table's Revenue rather than the viewer's — showing one player's score against a four-player
+ * target reads as a hopeless position when the table may be comfortably ahead.
+ */
+function targetHtml(f: Frame): string {
+  if (f.minCombinedRevenue <= 0) return '';
+  const combined = f.players.reduce((n, p) => n + p.revenue, 0);
+  const met = combined >= f.minCombinedRevenue;
+  return (
+    `<p>Combined Revenue <b>${combined}</b> against a target of <b>${f.minCombinedRevenue}</b>` +
+    `${met ? ' — cleared.' : ' — short.'}</p>`
+  );
+}
+
+/**
+ * Only when the game is actually scored on collisions.
+ *
+ * Two conditions, both of them `advance.ts`'s own: `0` on a dial turns that check off, and the
+ * checks run in COMPETITIVE AND CO-OP ONLY (§3.4). A solitaire game carries the default dials on
+ * its config and enforces neither, so reporting a collision budget there would put a rule on
+ * screen that this game does not have.
+ */
+function collisionsHtml(f: Frame): string {
+  const scoredOnCollisions =
+    (f.mode === 'competitive' || f.mode === 'coop') &&
+    (f.maxCollisionsTotal > 0 || f.maxCollisionsPerDay > 0);
+  return scoredOnCollisions
+    ? `<p>Collisions: <b>${f.collisionsToday}</b> today, <b>${f.collisionsTotal}</b> in all.</p>`
+    : '';
+}
+
+/**
+ * WHY THE GAME ENDED, as a sentence (Gitea#16).
+ *
+ * The page used to interpolate `outcome.reason` straight into the DOM, so a player who finished a
+ * game read the words `GAME OVER — revenueFloor`: an internal enum value, printed at the one moment
+ * the game has the player's whole attention. Each reason gets a sentence that says what actually
+ * happened, with this game's own numbers in it.
+ */
+function reasonSentence(f: Frame, o: NonNullable<Frame['outcome']>, day: number): string {
+  const combined = f.players.reduce((n, p) => n + p.revenue, 0);
+  switch (o.reason) {
+    case 'daysElapsed':
+      return `Day ${day} was the last on the timetable, and it ran out.`;
+    case 'revenueFloor':
+      return (
+        `The Division closed short: <b>${combined}</b> Revenue between everyone, against a floor of ` +
+        `<b>${f.minCombinedRevenue}</b>. §3.3 — miss the floor and the whole table loses, whoever ` +
+        `earned the most.`
+      );
+    case 'collisionFloor':
+      return (
+        `Too many collisions — <b>${f.collisionsToday}</b> in one Day and <b>${f.collisionsTotal}</b> ` +
+        `in all, against limits of ${f.maxCollisionsPerDay || '—'} and ${f.maxCollisionsTotal || '—'}. ` +
+        `§3.4 — the railroad was declared unsafe and the game was stopped.`
+      );
+  }
+}
+
+/** The rules this game was actually dealt under — Gitea#16's "what the rules of the game were". */
+function rulesHtml(f: Frame): string {
+  const mode =
+    f.mode === 'coop' ? 'Co-op — the table scores together' :
+    f.mode === 'competitive' ? 'Competitive — highest Revenue wins' :
+    'Solitaire';
+  const optional = [
+    f.optionalRules.employeeRotation ? 'Employee Rotation' : null,
+    f.optionalRules.reducedVisibility ? 'Reduced Visibility' : null,
+    f.optionalRules.emergencyToolbox ? 'Emergency Toolbox' : null,
+  ].filter((x): x is string => x !== null);
+  const r = f.houseRules.revenue;
+
+  const rows: [string, string][] = [
+    ['Scoring', mode],
+    ['Timetable', f.extraDays > 0
+      ? `${f.days} Days, extended by ${f.extraDays} more`
+      : `${f.days} Day${f.days === 1 ? '' : 's'}`],
+    ['Revenue floor', f.minCombinedRevenue > 0 ? `${f.minCombinedRevenue} combined` : 'none'],
+    ['Collision limits', f.maxCollisionsPerDay > 0 || f.maxCollisionsTotal > 0
+      ? `${f.maxCollisionsPerDay || '—'} per Day, ${f.maxCollisionsTotal || '—'} in all`
+      : 'not scored'],
+    ['Pay rates', `${r.freightPerLoad} per load, ${r.passengerPerCoach} per coach, ${r.trainPerTransit} per transit`],
+    ['Extras start', f.houseRules.extraStart === 'divisionPointsOnly' ? 'Division Points and the Interchange'
+      : f.houseRules.extraStart === 'ownOffice' ? 'those, plus your own Control Point'
+      : 'those, plus any Control Point'],
+    ['Timetabled trains', f.houseRules.discardTimetabled ? 'may be discarded' : 'are never discarded'],
+    ['Optional rules', optional.length ? optional.join(', ') : 'none'],
+  ];
+
+  return `<h4 class="res-h">The rules in play</h4>${factTable(rows)}`;
+}
+
+/**
+ * A two-column table of plain-text facts.
+ *
+ * BOTH HALVES ESCAPED, exactly once, which is the only reason this is a shared helper rather than a
+ * template repeated twice. The rows it is given today are numbers and fixed phrases, but they are
+ * assembled from the Frame — and the day somebody adds a row carrying a player's name, or a facility
+ * label, the escaping has to already be here rather than be remembered.
+ */
+function factTable(rows: [string, string][]): string {
+  return (
+    '<table class="res-t"><tbody>' +
+    rows.map(([k, v]) => `<tr><td>${esc(k)}</td><td>${esc(v)}</td></tr>`).join('') +
+    '</tbody></table>'
+  );
+}
+
+/**
+ * THE RAILROAD — what actually happened out there, from `GameState.tally` (Gitea#16).
+ *
+ * Rows that would read zero for a reason (no passenger work in a game that had none, no collisions
+ * in a clean one) are dropped rather than printed as `0`: a screen of zeroes reads as a bug, and the
+ * absence of a line is the same information more quietly. A zero that is genuinely interesting —
+ * trains through the Division — stays.
+ */
+function tallyHtml(t: Frame['tally']): string {
+  const rows: [string, string][] = [['Trains through the Division', String(t.trainsCompleted)]];
+
+  if (t.trainsCompleted > 0) {
+    rows.push([
+      'Of those, worked en route',
+      `${t.trainsCompletedWithWork} of ${t.trainsCompleted}` +
+        (t.trainsCompletedWithWork === t.trainsCompleted ? ' — every one' : ''),
+    ]);
+  }
+  const push = (label: string, n: number, detail = ''): void => {
+    if (n > 0) rows.push([label, `${n}${detail}`]);
+  };
+  push('Loads made up', t.loadsCompleted);
+  push('Loads broken', t.unloadsCompleted);
+  push('Loads still in the pipeline', t.loadsStarted - t.loadsCompleted);
+  push('Passengers boarded', t.passengersBoarded);
+  push('Passengers detrained', t.passengersDetrained);
+  push('Cars coupled', t.carsCoupled);
+  push('Cars set out', t.carsDropped);
+  push('Extras run', t.extrasStarted);
+  push('Second sections ordered', t.secondSections);
+  push('Flying switches', t.flyingSwitches);
+  push('Offices upgraded', t.officeUpgrades);
+  push('Facilities unjammed', t.facilitiesUnjammed);
+  push('Trains held', t.trainsHeld);
+  push('Trains diverted', t.trainsDiverted);
+  push('Expedite faults', t.expediteFaults);
+  push('Dispatch bonuses used', t.dispatchBonusesUsed);
+  if (t.clearancesRequested > 0) {
+    rows.push(['Clearances', `${t.clearancesAllowed} allowed of ${t.clearancesRequested} asked`]);
+  }
+  if (t.trainsDestroyed > 0) {
+    rows.push([
+      'Trains destroyed',
+      `${t.trainsDestroyed}, taking ${t.carsDestroyed} car${t.carsDestroyed === 1 ? '' : 's'} with them`,
+    ]);
+  }
+  // §X18 only — the one card in the deck that pays a train for standing still. Reported as what it
+  // is rather than as "the longest an engine sat on a siding", which the engine cannot answer (see
+  // `Tally.circusStops`).
+  if (t.circusStops.length > 0) {
+    rows.push([
+      'Circus set-ups',
+      t.circusStops.map((c) => `Train ${c.trainNumber} at ${c.where}`).join(', '),
+    ]);
+  }
+  push('Cards drawn', t.cardsDrawn);
+  push('Cards played', t.cardsPlayed);
+
+  return `<h4 class="res-h">The railroad</h4>${factTable(rows)}`;
+}
+
+/** Per-player work, for a table that wants to know who did what rather than only who won. */
+function perPlayerHtml(f: Frame): string {
+  const t = f.tally;
+  if (f.players.length < 2) return '';
+  const head =
+    '<tr><th></th><th class="num">Rev</th><th class="num">Loads</th><th class="num">Unloads</th>' +
+    '<th class="num">Pass.</th><th class="num">Cards</th><th class="num">Crashes</th></tr>';
+  const rows = [...f.players]
+    .sort((a, b) => b.revenue - a.revenue || a.seat - b.seat)
+    .map((p) => {
+      const q = t.byPlayer[p.index];
+      if (!q) return '';
+      return (
+        `<tr${p.index === f.viewer ? ' class="you"' : ''}><td>${esc(p.name)}</td>` +
+        `<td class="num">${p.revenue}</td><td class="num">${q.loads}</td>` +
+        `<td class="num">${q.unloads}</td>` +
+        `<td class="num">${q.passengersBoarded + q.passengersDetrained}</td>` +
+        `<td class="num">${q.cardsPlayed}</td><td class="num">${q.collisions}</td></tr>`
+      );
+    })
+    .join('');
+  return `<h4 class="res-h">Who did what</h4><table class="res-t res-wide"><tbody>${head}${rows}</tbody></table>`;
+}
+
+/**
+ * THE END-OF-GAME RESULTS SCREEN — Gitea#16, first pass.
+ *
+ * Everything the Frame already knew plus everything the event tally counted, in the order a player
+ * asks for it: what happened, who won, by how much, under what rules, and then what the railroad
+ * actually did all game. Badges and the "what would make this exciting" brainstorm are the second
+ * pass the issue asks for and are deliberately not here.
+ *
+ * THE OFFICIAL RESULT IS THE ONE AT THE TOP, always. In an extended game (Gitea#11) the standings go
+ * on moving after the winner is settled, so this screen reports the frozen result first and puts
+ * everything that happened afterwards in its own section, marked as informational. "In a five-day
+ * game, even if it's extended to eight or nine days, the winner and the official answer is the
+ * winner at the end of five days" (Jesse, 2026-08-28).
+ */
+export function resultsHtml(f: Frame): string {
+  // `official` is written by the engine the moment any game ends, so it is present on every finished
+  // game. The fallback keeps this rendering something sane for a Frame that predates it — a replay
+  // of a save recorded before this release, which the replay viewer will happily hand us.
+  const report = f.official;
+  const o = report?.outcome ?? f.outcome;
+  if (!o) return '<p class="dim">This game has not ended.</p>';
+
+  const officialDay = report?.day ?? f.days;
+  const winnerName =
+    o.winner === null ? null : (f.players.find((p) => p.index === o.winner)?.name ?? null);
+
+  const headline =
+    o.result === 'loss'
+      ? 'The Division failed'
+      : winnerName === null
+        ? 'The Division ran'
+        : `${esc(winnerName)} takes the Division`;
+
+  /**
+   * The result is reported against the standings AS THEY WERE at the official ending, not as they
+   * are now — in an extended game those are different numbers, and the winner has to be shown
+   * winning. `revenues` is frozen alongside the outcome for exactly this.
+   */
+  const frozen = report
+    ? { ...f, players: f.players.map((p) => ({ ...p, revenue: report.revenues[p.index] ?? p.revenue })) }
+    : f;
+
+  const result =
+    o.result === 'loss'
+      ? '<p>Nobody wins this one.</p>'
+      : o.winner === null
+        ? '<p>The table clears it together — a Co-op game has no individual winner.</p>'
+        : `<p><b>${esc(winnerName ?? '')}</b> finishes ahead on Revenue.</p>`;
+
+  const extended =
+    f.extraDays > 0
+      ? '<h4 class="res-h">After the timetable</h4>' +
+        `<p>The table played on for ${f.extraDays} more Day${f.extraDays === 1 ? '' : 's'}, ` +
+        `through Day ${f.days + f.extraDays}. None of it changed the result above — it is recorded ` +
+        'here because it happened.</p>' +
+        standingsHtml(f) +
+        targetHtml(f) +
+        collisionsHtml(f) +
+        tallyHtml(f.tally)
+      : '';
+
+  return (
+    `<h3 class="dayend-h res-${o.result}">${headline}</h3>` +
+    `<p>${reasonSentence(frozen, o, officialDay)}</p>` +
+    result +
+    standingsHtml(frozen, o.winner) +
+    targetHtml(frozen) +
+    collisionsHtml(frozen) +
+    perPlayerHtml(frozen) +
+    rulesHtml(f) +
+    tallyHtml(report?.tally ?? f.tally) +
+    extended
   );
 }
 
@@ -479,4 +746,20 @@ ul.blocked{margin:0;padding-left:18px}
 .dayend-t tr:first-child td{border-top:0}
 .dayend-t .num{text-align:right;font-variant-numeric:tabular-nums;font-weight:700;padding-right:0}
 .dayend-t .you td{color:#8fd6a0}
+.dayend-t .wins{color:#e8c56a;font-weight:700}
+
+/* END-OF-GAME RESULTS (Gitea#16). Same family as the Day-end dialog above, which is the point —
+   the two screens share their standings/target/collision blocks and should look like each other. */
+.res-h{font-size:12px;text-transform:uppercase;letter-spacing:.08em;color:#8b95a3;
+  margin:16px 0 6px;border-top:1px solid #2c333d;padding-top:10px}
+.res-win{color:#8fd6a0}
+.res-loss{color:#d98f8f}
+.res-t{border-collapse:collapse;margin:4px 0;width:100%}
+.res-t td,.res-t th{padding:3px 12px 3px 0;border-top:1px solid #232a33;vertical-align:top}
+.res-t tr:first-child td{border-top:0}
+.res-t td:first-child{color:#8b95a3;white-space:nowrap}
+.res-t th{color:#6d7783;font-weight:600;font-size:11px;text-transform:uppercase;letter-spacing:.05em}
+.res-t .num{text-align:right;font-variant-numeric:tabular-nums}
+.res-wide td:first-child{color:#e6e9ee}
+.res-t .you td{color:#8fd6a0}
 `;

src/web/play.html

@@ -222,6 +222,14 @@ h3.actions-hd{font-size:13px;text-transform:none;letter-spacing:.01em;color:#cfe
 .over{padding:9px;border-radius:5px;font-weight:700;margin-bottom:8px}
 .over.win{background:rgba(40,140,60,.35)}
 .over.loss{background:rgba(160,60,60,.3)}
+/* THE EXTENSION VOTE (Gitea#11) — unanimous, so who has not answered yet is the useful half. */
+.vote-tally{display:flex;flex-wrap:wrap;gap:4px 12px;margin:0 0 9px;font-size:12px}
+.vote.yes{color:#8fd6a0}
+.vote.no{color:#d98f8f}
+.vote.wait{color:var(--dim)}
+/* Wider than the New Game dialog: the results carry a seven-column per-player table (Gitea#16). */
+#resultsdlg{max-width:640px}
+#resultsdlg table{max-width:100%}
 /* cards, log, blocked */
 .card.gone{opacity:.35;text-decoration:line-through}
 .subj{display:block;width:100%;margin:2px 0}
@@ -890,6 +898,18 @@ ul.blocked li{padding:2px 0}
   </form>
 </dialog>
 
+<!-- THE END-OF-GAME RESULTS (Gitea#16). Filled by `resultsHtml` and opened from `renderEnding`,
+     which puts it up once per ending unasked and leaves a button to reopen it. Reopenable matters:
+     Gitea#11 lets a table play past the end, and continuing must not cost you the results screen. -->
+<dialog id="resultsdlg" aria-labelledby="rs-title">
+  <form method="dialog">
+    <div id="resultsbody"></div>
+    <menu class="ng-buttons">
+      <button value="ok" id="rs-ok" type="submit">Close</button>
+    </menu>
+  </form>
+</dialog>
+
 <!-- THE HANDOFF, between `Lobby.Start` and the first Frame.
      `beginRemote` used to write "… connecting to the game" into `#presence` — the DISCONNECT banner,
      whose job is `⚠ waiting on Alice`. It worked only because the first render overwrote it, and it

src/web/session.ts

@@ -27,6 +27,7 @@ import {
   currentActor,
   fromSave,
   handPlayable,
+  isOutOfTurn,
   newGame,
   overHandLimit,
   submit,
@@ -163,7 +164,9 @@ export function createLocalSession(seed: number, options?: NewGameOptions): Loca
     overHandLimit: () => overHandLimit(game),
     handPlayable: () => handPlayable(game),
     submit: async (intent: Intent) => {
-      const ok = submit(game, intent);
+      // Seat 0 is the solitaire player, and the extension vote (Gitea#11) is the one intent that
+      // arrives when `currentActor` is null — so it has to name its seat. See `isOutOfTurn`.
+      const ok = submit(game, intent, isOutOfTurn(intent) ? 0 : null);
       if (ok) changed();
       return ok;
     },

test/advance.test.ts

@@ -66,6 +66,21 @@ function playToCompletion(s: GameState, seed = 1, maxTurns = 20_000): PlayStats
     tally(pump(s));
     if (s.status === 'finished') break;
 
+    /**
+     * §3.3, EXTENDED PLAY (Gitea#11) — this harness plays the timetable it was dealt.
+     *
+     * It picks at random among legal options, and one of the two options here is "play another
+     * Day" — so left alone it would extend the game forever and terminate only on `maxTurns`. That
+     * is not a bug in the feature: a game genuinely does not end now until somebody says stop, and
+     * `developerBot` says stop for exactly this reason. Said explicitly here rather than folded
+     * into the random pick, because "how does this loop terminate" deserves an answer in the loop.
+     */
+    if (s.status === 'awaitingExtension') {
+      const r = applyIntent(s, 0, { type: 'game.extend', player: 0, agree: false });
+      assert.ok(r.ok, 'a solitaire player could not decline an extension');
+      break;
+    }
+
     const actor = s.clock.pendingDecision !== null ? s.clock.superintendent : s.clock.currentActor;
     if (actor === null) break;
 
@@ -544,9 +559,13 @@ describe('victory conditions (§3, Gap 10e) — unified 2026-08-20', () => {
     s.clock.phase = 'shiftChange';
     s.players[0]!.revenue = 0;
     advance(s);
-    assert.equal(s.status, 'finished');
+    // PAUSES rather than finishes since Gitea#11: a days-based ending offers another Day, and the
+    // result is recorded either way. `official` is the frozen answer; `status` is only where play
+    // has got to. The two collision tests at the foot of this block are the contrast.
+    assert.equal(s.status, 'awaitingExtension');
     assert.equal(s.outcome!.result, 'loss');
     assert.equal(s.outcome!.reason, 'revenueFloor');
+    assert.equal(s.official!.outcome.reason, 'revenueFloor');
   });
 
   it('wins a timed Solitaire game that clears minCombinedRevenue', () => {
@@ -556,7 +575,7 @@ describe('victory conditions (§3, Gap 10e) — unified 2026-08-20', () => {
     s.clock.phase = 'shiftChange';
     s.players[0]!.revenue = 10;
     advance(s);
-    assert.equal(s.status, 'finished');
+    assert.equal(s.status, 'awaitingExtension');
     assert.equal(s.outcome!.result, 'win');
   });
 
@@ -567,7 +586,7 @@ describe('victory conditions (§3, Gap 10e) — unified 2026-08-20', () => {
     s.clock.phase = 'shiftChange';
     s.players[0]!.revenue = 0;
     advance(s);
-    assert.equal(s.status, 'finished');
+    assert.equal(s.status, 'awaitingExtension');
     assert.equal(s.outcome!.result, 'win');
   });
 
@@ -582,7 +601,7 @@ describe('victory conditions (§3, Gap 10e) — unified 2026-08-20', () => {
     s.players[0]!.revenue = 5; // best individual score...
     s.players[1]!.revenue = 3; // ...but combined (8) still misses the floor (20).
     advance(s);
-    assert.equal(s.status, 'finished');
+    assert.equal(s.status, 'awaitingExtension');
     assert.equal(s.outcome!.result, 'loss');
     assert.equal(s.outcome!.reason, 'revenueFloor');
   });
@@ -598,7 +617,7 @@ describe('victory conditions (§3, Gap 10e) — unified 2026-08-20', () => {
     s.players[0]!.revenue = 4;
     s.players[1]!.revenue = 6; // combined 10 clears the floor, neither alone would.
     advance(s);
-    assert.equal(s.status, 'finished');
+    assert.equal(s.status, 'awaitingExtension');
     assert.equal(s.outcome!.result, 'win');
     assert.equal(s.outcome!.winner, null, 'Co-op names an individual winner instead of a shared one');
   });

test/cut-ordering.test.ts

@@ -13,7 +13,7 @@ import assert from 'node:assert/strict';
 
 import { applyIntent, areaOf, check } from '../src/engine/apply.ts';
 import { createGame } from '../src/engine/setup.ts';
-import type { CrewTray, GameConfig, GameState, GridCoord, RollingStock, TrackCard } from '../src/engine/state.ts';
+import type { CrewTray, GameConfig, GameState, GridCoord, RollingStock, TrackArc, TrackCard } from '../src/engine/state.ts';
 import { carsOn, coordKey, turnOf } from '../src/engine/state.ts';
 
 const config: GameConfig = {
@@ -53,12 +53,18 @@ function row(s: GameState, n: number): void {
   for (let c = 0; c < n; c++) addCard(s, at(1, c), straight());
 }
 
+/**
+ * `facing` is the PORT the engine points out through, which is not always an east-west one: a train
+ * standing on a curve points along its 45° leg. `railFacing` carries the east-west sense the train
+ * arrived with, so it keeps a straight answer whatever port the nose is on (`railFacingOf`).
+ */
 function placeTray(
   s: GameState,
   coord: GridCoord,
   consist: RollingStock[],
-  facing: 'e' | 'w',
+  facing: 'n' | 's' | 'e' | 'w',
   engineAt = 0,
+  railFacing: 'e' | 'w' = facing === 'w' ? 'w' : 'e',
 ): string {
   const id = s.freeTrays.pop()!;
   s.trays.set(id, {
@@ -67,9 +73,9 @@ function placeTray(
     trainIsExtra: false,
     engineAt,
     consist,
-    direction: facing === 'w' ? 'west' : 'east',
+    direction: railFacing === 'w' ? 'west' : 'east',
     facing,
-    railFacing: facing,
+    railFacing,
     position: { at: 'grid', seat: 0, coord },
     movesUsed: 0,
   } as CrewTray);
@@ -373,3 +379,85 @@ describe('taking your own cut back is undoing the drop, not a fresh pick-up', ()
     );
   });
 });
+
+// ---------------------------------------------------------------------------
+
+describe('a 45° leg is part of the west-to-east row, not outside it (Gitea#17)', () => {
+  /**
+   * Reported: "Cars were West to East Caboose, Loaded boxcar, Loaded boxcar, Loaded boxcar. After
+   * backing into that square cars were attached to the train Loaded boxcar, Loaded boxcar, Loaded
+   * boxcar, Caboose, Engine." The caboose came back next to the engine instead of at the far end,
+   * which also leaves the train badly made up under §8.2.
+   *
+   * The square was a `sw` CURVE and the train backed in through its SOUTH leg. `standing` runs west
+   * to east, and the two places that walk it both asked the PORT which end of the row they were at:
+   * `exploreMoves` reversed the row for an 'e' entry and for nothing else, and `cutTowards` answered
+   * "you meet nothing" for a north or south exit. Neither is a property of the port.
+   *
+   * A 45° leg leaves through the MIDDLE of its edge, so its end of the run is whichever end the arc
+   * does not reach: the south leg of a `sw` curve is the row's EAST end, and the south leg of an
+   * `se` curve is its WEST end. Same port, opposite answers — which is why `rowEndAt` has to ask the
+   * card.
+   */
+  const curve = (arc: TrackArc, standing: RollingStock[] = [], standingWest = 0): TrackCard => ({
+    geometry: { kind: 'track', geometry: 'curved', arc, hand: 'right' },
+    baseOperationalRail: true,
+    standing,
+    standingWest,
+    facility: null,
+    modifiers: [],
+    enhancements: [],
+  });
+
+  /**
+   * The reported board, minimally: a `sw` curve holding the cut, and an `ne` curve below it for the
+   * train to run from. Both legs lie on the `ne_sw` diagonal, so the two cards actually join.
+   */
+  function board(standing: RollingStock[], standingWest = standing.length): GameState {
+    const s = game();
+    addCard(s, at(1, 0), curve('sw', standing, standingWest));
+    addCard(s, at(0, 0), curve('ne'));
+    switching(s);
+    return s;
+  }
+
+  it('backs into a cut through the south leg and meets the EAST end of the row first', () => {
+    const s = board([car('caboose', true), car('boxcar', true), car('boxcar', true), car('boxcar', true)]);
+    // Facing east on the `ne` curve, so reversing pulls out through its north leg and into the
+    // curve above through that card's south leg — the move in the reported save.
+    const id = placeTray(s, at(0, 0), [], 'e');
+    const r = applyIntent(s, 0, { type: 'switch.move', trayId: id, to: at(1, 0), reverse: true });
+    assert.ok(r.ok, `the reverse move was refused: ${r.ok ? '' : r.code}`);
+    // Coupled behind the engine nearest-car-first, and the nearest car is the one at the south end
+    // — the LAST of a west-to-east row on a `sw` curve. The caboose was westmost, so it ends up
+    // furthest from the engine, which is where §8.2 needs it.
+    assert.deepEqual(types(s.trays.get(id)!.consist), ['boxcar', 'boxcar', 'boxcar', 'caboose']);
+  });
+
+  it('meets the WEST end of the row first where the same leg belongs to an `se` curve', () => {
+    // The mirror, and the reason the port alone cannot answer: an `se` curve's south leg is the
+    // west end of its row, so the same reverse move meets the caboose first.
+    const s = game();
+    addCard(s, at(1, 0), curve('se', [car('caboose', true), car('boxcar', true)], 2));
+    addCard(s, at(0, 0), curve('nw'));
+    switching(s);
+    const id = placeTray(s, at(0, 0), [], 'w');
+    const r = applyIntent(s, 0, { type: 'switch.move', trayId: id, to: at(1, 0), reverse: true });
+    assert.ok(r.ok, `the reverse move was refused: ${r.ok ? '' : r.code}`);
+    assert.deepEqual(types(s.trays.get(id)!.consist), ['caboose', 'boxcar']);
+  });
+
+  it('takes its own cut back with it when it pulls out through the south leg (§A.4)', () => {
+    // The other half of the same assumption: `cutTowards` said a train leaving north or south meets
+    // nothing, so a crew standing on a curve drove away and left the cars beside it standing —
+    // exactly what mandatory coupling forbids.
+    const s = board([car('boxcar', true)], 0);
+    // `standingWest` 0 puts the boxcar EAST of the train, which on a `sw` curve is between it and
+    // the south leg it is about to leave by.
+    const id = placeTray(s, at(1, 0), [], 's');
+    const r = applyIntent(s, 0, { type: 'switch.move', trayId: id, to: at(0, 0), reverse: false });
+    assert.ok(r.ok, `the move off the curve was refused: ${r.ok ? '' : r.code}`);
+    assert.deepEqual(types(s.trays.get(id)!.consist), ['boxcar'], 'the cut beside the train was left standing');
+    assert.deepEqual(standingAt(s, at(1, 0)), [], 'the cars should have come off the card');
+  });
+});

test/extended-play.test.ts

@@ -0,0 +1,256 @@
+/**
+ * §3.3, EXTENDED PLAY — Gitea#11, "when game ends allow players to continue playing if they wish".
+ *
+ * The rule as Jesse specified it (2026-08-28), which is what these tests are written against:
+ *
+ *   - the OFFICIAL result is decided at the original game length and never changes. "In a five-day
+ *     game, even if it's extended to eight or nine days, the winner and the official answer is the
+ *     winner at the end of five days";
+ *   - extending grants exactly ONE Day, and the question is put again at the end of it;
+ *   - solitaire: the player decides alone. Multiplayer: unanimous, and one refusal ends it there;
+ *   - only days-based endings offer it. A §3.4 collision breach is final, during an extended Day
+ *     just as during the regular game.
+ *
+ * The persistence half matters as much as the rules half: a save is `{ seed, config, history }`
+ * replayed through the engine, so an extension that is not an INTENT does not survive a reload, an
+ * Undo, or a server restart. `replays the extension` below is the test that pins that.
+ */
+
+import { describe, it } from 'node:test';
+import assert from 'node:assert/strict';
+
+import { advance, pump } from '../src/engine/advance.ts';
+import { applyIntent, check } from '../src/engine/apply.ts';
+import { STAGES_PER_DAY } from '../src/engine/content.ts';
+import { legalActions } from '../src/engine/legal.ts';
+import { createGame } from '../src/engine/setup.ts';
+import { developerBot, playGame, randomBot } from '../src/sim/bot.ts';
+import type { GameConfig, GameState } from '../src/engine/state.ts';
+
+const baseConfig = (over: Partial<GameConfig> = {}): GameConfig => ({
+  mode: 'solitaire',
+  days: 3,
+  minCombinedRevenue: 0,
+  maxCollisionsPerDay: 0,
+  maxCollisionsTotal: 0,
+  pvpCardsAllowed: false,
+  optionalRules: { reducedVisibility: false, employeeRotation: false, emergencyToolbox: false },
+  ...over,
+});
+
+const game = (over: Partial<GameConfig> = {}, names = ['Jesse']): GameState =>
+  createGame({ id: 'g', seed: 1, config: baseConfig(over), playerNames: names });
+
+/** Parks a game on the last Stage of its final Day, so one `advance` runs the clock off the end. */
+function atTheEnd(s: GameState): GameState {
+  s.clock.day = s.config.days + s.extraDays + 1;
+  s.clock.stage = STAGES_PER_DAY;
+  s.clock.phase = 'shiftChange';
+  return s;
+}
+
+describe('§3.3 extended play — the ending pauses rather than stopping (Gitea#11)', () => {
+  it('offers another Day on a days-based ending, and records the result anyway', () => {
+    const s = atTheEnd(game());
+    advance(s);
+    assert.equal(s.status, 'awaitingExtension', 'a days-based ending did not offer another Day');
+    assert.ok(s.outcome, 'the result was not decided');
+    assert.ok(s.official, 'the official result was not frozen');
+    assert.equal(s.official!.day, s.config.days, 'the official Day is not the original game length');
+  });
+
+  it('does NOT offer another Day after a collision breach — §3.4 is final', () => {
+    const s = game({ mode: 'competitive', maxCollisionsPerDay: 2 });
+    s.collisionsToday = 2;
+    s.clock.stage = 1;
+    s.clock.phase = 'shiftChange';
+    advance(s);
+    assert.equal(s.status, 'finished', 'a railroad declared unsafe offered to carry on');
+    assert.equal(s.outcome!.reason, 'collisionFloor');
+  });
+
+  it('refuses every ordinary intent while the extension question is open', () => {
+    const s = atTheEnd(game());
+    advance(s);
+    assert.equal(check(s, 0, { type: 'draw.fromHomeOffice' }), 'WRONG_PHASE');
+    assert.equal(check(s, 0, { type: 'game.extend', player: 0, agree: true }), null, 'the vote itself was refused');
+  });
+
+  it('offers only the two votes, and only to a seat that has not voted', () => {
+    const s = atTheEnd(game({ mode: 'competitive' }, ['A', 'B']));
+    advance(s);
+    assert.deepEqual(
+      legalActions(s, 0).map((i) => i.type),
+      ['game.extend', 'game.extend'],
+      'something other than the vote is legal while the game is stopped',
+    );
+    applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    assert.equal(legalActions(s, 0).length, 0, 'a seat was offered a second vote');
+    assert.equal(check(s, 0, { type: 'game.extend', player: 0, agree: true }), 'ALREADY_VOTED');
+    assert.equal(legalActions(s, 1).length, 2, 'the seat still to vote was not offered the vote');
+  });
+
+  it('refuses the vote when no extension is pending', () => {
+    const s = game();
+    assert.equal(check(s, 0, { type: 'game.extend', player: 0, agree: true }), 'NOT_AWAITING_EXTENSION');
+  });
+});
+
+describe('§3.3 extended play — one Day at a time (Gitea#11)', () => {
+  it('grants exactly one Day in solitaire, then asks again at the end of it', () => {
+    const s = atTheEnd(game());
+    advance(s);
+    const official = { ...s.official!.outcome };
+
+    applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    assert.equal(s.status, 'active', 'agreeing did not resume play');
+    assert.equal(s.extraDays, 1, 'more or less than one Day was granted');
+    assert.deepEqual(s.extensionVotes, [null], 'the votes were not cleared for the next question');
+
+    // Run the extra Day off the end: the question comes round again.
+    atTheEnd(s);
+    advance(s);
+    assert.equal(s.status, 'awaitingExtension', 'the second ending did not ask again');
+    assert.equal(s.extraDays, 1, 'a second Day was granted without being asked for');
+    assert.deepEqual(s.official!.outcome, official, 'the official result was rewritten');
+    assert.equal(s.official!.day, s.config.days, 'the official Day moved with the extension');
+  });
+
+  it('ends the moment one seat declines, without waiting for the rest', () => {
+    const s = atTheEnd(game({ mode: 'competitive' }, ['A', 'B', 'C']));
+    advance(s);
+    applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    assert.equal(s.status, 'awaitingExtension', 'one yes ended the vote');
+
+    applyIntent(s, 1, { type: 'game.extend', player: 1, agree: false });
+    assert.equal(s.status, 'finished', 'a refusal did not end the game immediately');
+    assert.equal(s.extraDays, 0, 'a Day was granted despite a refusal');
+    assert.equal(check(s, 2, { type: 'game.extend', player: 2, agree: true }), 'NOT_AWAITING_EXTENSION',
+      'the seat that never voted is still being waited on');
+  });
+
+  it('needs every seat before it grants the Day', () => {
+    const s = atTheEnd(game({ mode: 'competitive' }, ['A', 'B', 'C']));
+    advance(s);
+    applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    applyIntent(s, 1, { type: 'game.extend', player: 1, agree: true });
+    assert.equal(s.status, 'awaitingExtension', 'two of three votes granted the Day');
+    applyIntent(s, 2, { type: 'game.extend', player: 2, agree: true });
+    assert.equal(s.status, 'active', 'a unanimous table was not given its Day');
+    assert.equal(s.extraDays, 1);
+  });
+
+  it('keeps the official result when a collision ends an EXTENDED Day', () => {
+    // The case the freeze exists for: a railroad declared unsafe on the extra Day does not retract
+    // who won on the last scheduled one.
+    const s = atTheEnd(game({ mode: 'competitive', maxCollisionsPerDay: 2 }, ['A', 'B']));
+    s.players[0]!.revenue = 9;
+    s.players[1]!.revenue = 2;
+    advance(s);
+    const official = { ...s.official!.outcome };
+    assert.equal(official.result, 'win');
+    assert.equal(official.winner, 0);
+
+    applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    applyIntent(s, 1, { type: 'game.extend', player: 1, agree: true });
+    assert.equal(s.status, 'active');
+
+    s.collisionsToday = 2;
+    s.clock.stage = 1;
+    s.clock.phase = 'shiftChange';
+    advance(s);
+    assert.equal(s.status, 'finished');
+    assert.equal(s.outcome!.reason, 'collisionFloor', 'the current evaluation was not updated');
+    assert.deepEqual(s.official!.outcome, official, 'a late collision rewrote a recorded win');
+  });
+
+  it('decides the winner at the ORIGINAL game length, whoever leads afterwards', () => {
+    const s = atTheEnd(game({ mode: 'competitive' }, ['A', 'B']));
+    s.players[0]!.revenue = 9;
+    s.players[1]!.revenue = 2;
+    advance(s);
+    assert.equal(s.official!.outcome.winner, 0);
+    assert.deepEqual(s.official!.revenues, [9, 2], 'the official standings were not frozen');
+
+    applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    applyIntent(s, 1, { type: 'game.extend', player: 1, agree: true });
+
+    // B overtakes A during the extra Day, and it changes nothing official.
+    s.players[1]!.revenue = 40;
+    atTheEnd(s);
+    advance(s);
+    assert.equal(s.official!.outcome.winner, 0, 'overtaking after the timetable took the win');
+    assert.deepEqual(s.official!.revenues, [9, 2], 'the frozen standings moved');
+    assert.equal(s.outcome!.winner, 1, 'the informational evaluation did not follow the new leader');
+  });
+});
+
+describe('§3.3 extended play — it survives being replayed (Gitea#11)', () => {
+  it('reproduces an extended game from seed and intents alone', () => {
+    // The reason the vote is an intent at all. A save is a replay, so a decision that is not in the
+    // history did not happen — an extended game would evaporate on the next reload.
+    const play = (): GameState => {
+      const s = atTheEnd(game());
+      advance(s);
+      applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+      return s;
+    };
+    const a = play();
+    const b = play();
+    assert.equal(a.extraDays, b.extraDays);
+    assert.equal(a.status, b.status);
+    assert.deepEqual(a.official!.outcome, b.official!.outcome);
+  });
+
+  it('narrates the vote, so a table can see who called time', () => {
+    const s = atTheEnd(game({ mode: 'competitive' }, ['A', 'B']));
+    advance(s);
+    const yes = applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    assert.ok(yes.ok);
+    assert.deepEqual(yes.events.map((e) => e.type), ['extensionVoted']);
+    const no = applyIntent(s, 1, { type: 'game.extend', player: 1, agree: false });
+    assert.ok(no.ok);
+    assert.deepEqual(no.events.map((e) => e.type), ['extensionVoted', 'playConcluded']);
+  });
+
+  it('announces the granted Day', () => {
+    const s = atTheEnd(game());
+    advance(s);
+    const r = applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    assert.ok(r.ok);
+    assert.deepEqual(r.events.map((e) => e.type), ['extensionVoted', 'dayExtended']);
+    const extended = r.events.find((e) => e.type === 'dayExtended');
+    assert.equal(extended && 'day' in extended ? extended.day : null, s.config.days + 1);
+  });
+});
+
+describe('§3.3 extended play — bots play the timetable they were dealt (Gitea#11)', () => {
+  it('REGRESSION: a simulated game terminates whatever the policy would vote', () => {
+    /**
+     * `playGame` declines on the driver's own account rather than leaving it to the policy, and this
+     * is why. `randomBot` picks uniformly among its legal options, so it takes another Day about
+     * half the time — and since a table may go on granting Days for ever, the game then runs to
+     * `maxTurns`. It did: `test/sim.test.ts` went from under a second to an unbounded hang, every
+     * seeded game in the harness playing fifty thousand turns instead of a couple of hundred.
+     *
+     * `randomBot` is the policy that exposes it, but the guarantee has to hold for any policy,
+     * including ones not written yet — hence the fix living in the driver and the test living here.
+     */
+    const s = createGame({ id: 'g', seed: 9, config: baseConfig({ days: 1 }), playerNames: ['Jesse'] });
+    const out = playGame(s, randomBot(7), pump, 4_000);
+    assert.equal(s.status, 'finished', 'a simulated game did not finish');
+    assert.equal(s.extraDays, 0, 'the harness played Days the game was not dealt');
+    assert.ok(out.turns < 4_000, `ran to the turn cap (${out.turns}) instead of ending`);
+  });
+
+  it('developerBot declines, so a bot-only game ends on schedule', () => {
+    // "If only bots are playing, they never vote to extend" (Jesse, 2026-08-28). The server votes
+    // yes on a bot's behalf once every human has already agreed; this policy is what is left when
+    // there are no humans to follow.
+    const s = atTheEnd(game());
+    advance(s);
+    const choice = developerBot.choose(s, 0, legalActions(s, 0));
+    assert.equal(choice.type, 'game.extend');
+    assert.equal('agree' in choice ? choice.agree : null, false, 'a bot asked for another Day');
+  });
+});

test/harness.test.ts

@@ -228,14 +228,28 @@ describe('the game conserves Rolling Stock', () => {
       for (let t = 0; t < 50_000; t++) {
         const before = census(s);
         const pumped = pump(s);
-        // §10 — a collision destroys both trains and everything they were carrying. That is the one
-        // legitimate way the count falls, so the expectation follows it down.
-        for (const e of pumped) {
-          if (e.type === 'trainsDestroyed') for (const tr of e.trains) expected -= tr.consist.length;
-        }
-        assert.ok(
-          census(s) === before || pumped.some((e) => e.type === 'trainsDestroyed'),
-          `seed ${seed}: the engine changed the census by ${census(s) - before} outside a collision`,
+        /**
+         * A COLLISION DESTROYS NO CAR, and this used to assume it destroyed all of them.
+         *
+         * The subtraction that stood here — `expected -= tr.consist.length` for every train in a
+         * `trainsDestroyed` event — describes a rule the engine does not have. Gap 2c (`advance.ts`,
+         * "TAKE THE WRECK OFF THE CARD") sends the wreck's cabooses back to the Division Yard and
+         * everything else to Classification, so the stock is conserved through a collision like any
+         * other move. The train is destroyed; its cars are not.
+         *
+         * It passed for as long as it did because none of the six seeds below ever collided, so the
+         * branch never ran. Changing the deck to the sheet's counts (Gitea#14) moved the deals, seed
+         * 24757 collided, and the test failed claiming the engine had CONJURED three cars — the
+         * exact opposite of what had happened.
+         *
+         * So the census is now held flat, unconditionally, which is both the real invariant and a
+         * stronger test than the one it replaces: there is no longer any event that excuses a
+         * change, and `expected` cannot drift away from the supply it was dealt.
+         */
+        assert.equal(
+          census(s),
+          before,
+          `seed ${seed}: the engine changed the census by ${census(s) - before} while pumping`,
         );
         if (s.status === 'finished') break;
         const actor = s.clock.pendingDecision !== null ? s.clock.superintendent : s.clock.currentActor;
@@ -290,7 +304,18 @@ describe('every published replay actually replays', () => {
         `${f} is dead — it replays ${back.history.length} of ${save.history.length} intents. ` +
           'Re-record it with save-replay.ts rather than editing the file.',
       );
-      assert.equal(back.state.status, 'finished', `${f} does not reach the end of its game`);
+      /**
+       * `awaitingExtension` counts as the end since Gitea#11 — and for a published file it is the
+       * EXPECTED end. These saves were recorded before extended play existed, so their histories
+       * carry no `game.extend` vote: replaying one runs the timetable out and stops on the question
+       * nobody was there to answer. That is a game that reached the end of its own history, which is
+       * what this test is about. `active` would still be a dead replay.
+       */
+      assert.ok(
+        back.state.status === 'finished' || back.state.status === 'awaitingExtension',
+        `${f} does not reach the end of its game — status ${back.state.status}`,
+      );
+      assert.ok(back.state.official !== null, `${f} ends without recording a result`);
     }
   });
 });

test/mainline-cards.test.ts

@@ -12,6 +12,7 @@
 import { describe, it } from 'node:test';
 import assert from 'node:assert/strict';
 
+import type { MainlineKind } from '../src/engine/content.ts';
 import { advance } from '../src/engine/advance.ts';
 import { applyIntent, areaOf, check } from '../src/engine/apply.ts';
 import { legalActions } from '../src/engine/legal.ts';
@@ -43,6 +44,20 @@ const game = (seed = 5): GameState => createGame({ id: 'g', seed, config, player
 const at = (row: number, col: number): GridCoord => ({ row, col });
 
 /** Puts a card of `kind`/`key` in hand and returns its id. */
+/**
+ * Puts a specific rules card in hand and returns its id, MINTING ONE IF THE DECK NO LONGER DEALS IT.
+ *
+ * A card at `copies: 0` is still a card: the catalogue keeps its row and the engine keeps its rule,
+ * so the design stays visible and the mechanic works the moment it is dealt again. Poling and the
+ * sharp curves have been treated that way for a while, and Gitea#14 put the dispatching ladder,
+ * Facing Point Locks, Flying Switch, Section House and Vandalism there too — none of them are in
+ * `docs/Deck cards5.xlsx`.
+ *
+ * This used to throw when it could not find one, which made "dealt zero copies" and "deleted"
+ * indistinguishable from a test's point of view: zeroing Flying Switch took five passing tests of a
+ * rule that had not changed at all down with it. Minting keeps the rule under test independently of
+ * whether the deck currently deals the card, which is the whole reason for keeping the row.
+ */
 function hand(s: GameState, kind: string, key: string): string {
   for (const [id, card] of s.cards) {
     const k = card.kind as { kind: string; key?: string };
@@ -51,7 +66,10 @@ function hand(s: GameState, kind: string, key: string): string {
       return id;
     }
   }
-  throw new Error(`no ${kind} card: ${key}`);
+  const id = `zero-copy-${kind}-${key}`;
+  s.cards.set(id, { id, kind: { kind, key } } as never);
+  s.decks.hands.set(0, [id]);
+  return id;
 }
 
 /**
@@ -90,53 +108,124 @@ function drawTurn(s: GameState): void {
 
 // ---------------------------------------------------------------------------
 
-describe('grade modifiers change crossing time', () => {
-  it('a Heavy Grade takes two Stages bare', () => {
-    assert.equal(crossingStages('heavyGrade', 'fast', false), 2);
+/**
+ * Crossing time on the region model (Gitea#3). `cross` fills in the parts each test does not care
+ * about, so the numbers below read as the table RAR gave rather than as argument lists.
+ */
+const cross = (
+  kind: Parameters<typeof crossingStages>[0],
+  over: Partial<Parameters<typeof crossingStages>[1]> = {},
+): number =>
+  crossingStages(kind, {
+    trainSpeed: 'fast',
+    direction: 'east',
+    gradeUp: 'east',
+    modifiers: [],
+    ...over,
+  });
+
+describe('a card costs one Stage per printed region (Gitea#3)', () => {
+  /**
+   * RAR, 2026-08-26, and this REPLACES the two rules that were here before — Q1, "the printed 60/30
+   * are mph expressed as crossing time", and Q2, "a Slow train adds one Stage to every card".
+   *
+   * "Ignore speed signs, they are just graphics. Regions shown on cards indicate how many stages it
+   * takes to cross. Plains is 1. Double track is 1, tunnel is 2, curves is 2, heavy grade is 3
+   * unless you have help."
+   *
+   * The report that opened the issue was a Slow train taking two Stages to clear Double Track. Q2 is
+   * what did that, and it is gone.
+   */
+  it('crosses in the number of regions the card prints, whatever the train', () => {
+    for (const speed of ['fast', 'slow'] as const) {
+      assert.equal(cross('plains', { trainSpeed: speed }), 1, `plains, ${speed}`);
+      assert.equal(cross('doubleTrack', { trainSpeed: speed }), 1, `double track, ${speed}`);
+      assert.equal(cross('trestle', { trainSpeed: speed }), 1, `trestle, ${speed}`);
+      assert.equal(cross('curves', { trainSpeed: speed }), 2, `curves, ${speed}`);
+      assert.equal(cross('tunnel', { trainSpeed: speed }), 2, `tunnel, ${speed}`);
+      assert.equal(cross('heavyGrade', { trainSpeed: speed }), 3, `heavy grade, ${speed}`);
+    }
+  });
+
+  it('reads Fast/Slow on Hilly and on nothing else', () => {
+    // "Some cards say fast / slow… Fast / Slow does not apply to every card — just those that say
+    // fast / slow on them. Currently this is only hilly." A fast train starts in the second region.
+    assert.equal(cross('hilly', { trainSpeed: 'fast' }), 1);
+    assert.equal(cross('hilly', { trainSpeed: 'slow' }), 2);
+  });
+
+  it('does not read the consist any more', () => {
+    // Hilly used to take its split off the printed P60/F30 and decide by whether the train carried a
+    // coach, so a fast freight crossed slower than a slow passenger train. RAR: "I notice that you
+    // are basing stages in mainline cards off coach/non-coach. Actually, all trains are rated as
+    // FAST and SLOW." `crossingStages` no longer takes a consist at all — this test is here so the
+    // deletion is deliberate rather than incidental.
+    assert.equal(cross('hilly', { trainSpeed: 'fast' }), cross('hilly', { trainSpeed: 'fast' }));
+  });
+
+  it('runs a train through a siding or an Interchange in one Stage', () => {
+    // Both print a back region that is not part of the road, so a train passing through starts past
+    // it. What that region is FOR is tested below and in the collision tests.
+    assert.equal(cross('uncontrolledSiding'), 1);
+    assert.equal(cross('interchange'), 1);
+  });
+
+  it('costs the extra Stage to anything starting in that back region', () => {
+    // The Uncontrolled Siding with a train already on it, and an Extra beginning its run at an
+    // Interchange. Both start at the back and have the whole card to run.
+    assert.equal(cross('uncontrolledSiding', { startsAtBack: true }), 2);
+    assert.equal(cross('interchange', { startsAtBack: true }), 2);
+  });
+});
+
+describe('grade modifiers move the start, not the clock', () => {
+  it('a Heavy Grade takes three Stages bare', () => {
+    // Three regions, up from the two the old 30mph reading gave it.
+    assert.equal(cross('heavyGrade'), 3);
   });
 
   it('Brakeman speeds the descent but not the climb', () => {
-    // Q11 — the card prints "(Up)" and "Player sets orientation", so the last argument is which
-    // way is UPHILL. With up = east, a westbound train is descending.
-    assert.equal(crossingStages('heavyGrade', 'fast', false, ['brakeman'], 'west', 'east'), 1);
-    assert.equal(crossingStages('heavyGrade', 'fast', false, ['brakeman'], 'east', 'east'), 2);
+    // Q11 — the card prints "(Up)" and "Player sets orientation", so `gradeUp` is which way is
+    // UPHILL. With up = east, a westbound train is descending.
+    assert.equal(cross('heavyGrade', { modifiers: ['brakeman'], direction: 'west' }), 2);
+    assert.equal(cross('heavyGrade', { modifiers: ['brakeman'], direction: 'east' }), 3);
   });
 
-  it('follows the orientation the player chose, not a fixed compass direction', () => {
-    // The same train on the same card, with the card turned around: Brakeman helps a westbound
-    // train on an east-climbing grade, and an eastbound one when the grade climbs west.
-    assert.equal(crossingStages('heavyGrade', 'fast', false, ['brakeman'], 'east', 'west'), 1);
-    assert.equal(crossingStages('heavyGrade', 'fast', false, ['brakeman'], 'west', 'west'), 2);
-    assert.equal(crossingStages('heavyGrade', 'fast', false, ['helpers'], 'west', 'west'), 1);
-    assert.equal(crossingStages('heavyGrade', 'fast', false, ['helpers'], 'east', 'west'), 2);
+  it('follows the orientation the card was dealt, not a fixed compass direction', () => {
+    // The same train on the same card, with the card turned around.
+    assert.equal(cross('heavyGrade', { modifiers: ['brakeman'], direction: 'east', gradeUp: 'west' }), 2);
+    assert.equal(cross('heavyGrade', { modifiers: ['brakeman'], direction: 'west', gradeUp: 'west' }), 3);
+    assert.equal(cross('heavyGrade', { modifiers: ['helpers'], direction: 'west', gradeUp: 'west' }), 2);
+    assert.equal(cross('heavyGrade', { modifiers: ['helpers'], direction: 'east', gradeUp: 'west' }), 3);
   });
 
   it('Helpers speed the climb but not the descent', () => {
-    assert.equal(crossingStages('heavyGrade', 'fast', false, ['helpers'], 'east', 'east'), 1);
-    assert.equal(crossingStages('heavyGrade', 'fast', false, ['helpers'], 'west', 'east'), 2);
+    // RAR: "helpers… helps all trains going up hill by starting 1 region easier — so 2 to traverse,
+    // not 3."
+    assert.equal(cross('heavyGrade', { modifiers: ['helpers'], direction: 'east' }), 2);
+    assert.equal(cross('heavyGrade', { modifiers: ['helpers'], direction: 'west' }), 3);
   });
 
-  it('Airbrakes stack with Brakeman on a slow train', () => {
-    // A slow train pays 3 on a grade; Brakeman and Airbrakes take one Stage each.
-    assert.equal(crossingStages('heavyGrade', 'slow', false, [], 'west', 'east'), 3);
-    assert.equal(crossingStages('heavyGrade', 'slow', false, ['brakeman'], 'west', 'east'), 2);
-    assert.equal(
-      crossingStages('heavyGrade', 'slow', false, ['brakeman', 'airbrakes'], 'west', 'east'),
-      1,
-    );
+  it('Airbrakes stack on top of Brakeman', () => {
+    // "Airbrakes is an upgrade from brakemen (which must be played first)", so a fully-equipped
+    // grade is one Stage downhill — and `check` refuses Airbrakes without Brakeman already there.
+    assert.equal(cross('heavyGrade', { direction: 'west' }), 3);
+    assert.equal(cross('heavyGrade', { modifiers: ['brakeman'], direction: 'west' }), 2);
+    assert.equal(cross('heavyGrade', { modifiers: ['brakeman', 'airbrakes'], direction: 'west' }), 1);
   });
 
   it('never lets a train cross in no time', () => {
+    // Three modifiers on a three-region card would otherwise put the start past the far edge.
     assert.equal(
-      crossingStages('heavyGrade', 'fast', false, ['brakeman', 'airbrakes'], 'west', 'east'),
+      cross('heavyGrade', { modifiers: ['brakeman', 'airbrakes', 'helpers'], direction: 'west' }),
       1,
     );
   });
 
   it('leaves non-grade cards alone', () => {
     // Brakeman on Plains would be an illegal placement anyway; the maths must not move regardless.
-    assert.equal(crossingStages('plains', 'fast', false, ['brakeman'], 'west', 'east'), 1);
-    assert.equal(crossingStages('curves', 'fast', false, ['helpers'], 'east', 'east'), 2);
+    assert.equal(cross('plains', { modifiers: ['brakeman'], direction: 'west' }), 1);
+    assert.equal(cross('curves', { modifiers: ['helpers'] }), 2);
   });
 });
 
@@ -209,8 +298,8 @@ describe('Realignment converts one Mainline type to another', () => {
 
     assert.ok(r.ok);
     assert.equal(node.card, 'plains', 'Curves realigns to Plains');
-    // The point of the card: Curves is a 30 (two Stages), Plains a 60 (one).
-    assert.equal(crossingStages(node.card, 'fast', false), 1);
+    // The point of the card: Curves prints two regions, Plains one, so realigning halves the time.
+    assert.equal(cross(node.card), 1);
   });
 
   it('refuses a card with no conversion listed', () => {
@@ -796,18 +885,35 @@ describe('Q13 — a train that catches the one ahead runs into it', () => {
    * then region 1 — so it catches up whichever order the phase happens to process them in.
    */
   const twoTrains = (opts: { absSignals?: boolean } = {}): { s: GameState; events: GameEvent[] } => {
-    const s = createGame({
-      id: 'rear', seed: 3,
-      config: {
-        mode: 'solitaire', days: 5, minCombinedRevenue: 0, maxCollisionsPerDay: 0, maxCollisionsTotal: 0, pvpCardsAllowed: false,
-        optionalRules: { reducedVisibility: false, employeeRotation: false, emergencyToolbox: false },
-      },
-      playerNames: ['bot'],
-    });
-    const index = s.division.nodes.findIndex(
-      (n) => n.kind === 'mainline' && !mainlineProfile(n.card).trainsMayPass,
-    );
-    assert.ok(index >= 0, 'no single-track Mainline card in this Division');
+    /**
+     * THE SEED IS SEARCHED FOR, NOT WRITTEN DOWN.
+     *
+     * This asked for seed 3 and asserted that its Division held a single-track Mainline card. It
+     * does not any more: the Division is laid out from the same RNG stream the card deck is
+     * shuffled from, so changing the SIZE of that deck re-deals the Division too. Gitea#14's deck
+     * counts moved it, and the test failed on its own precondition — "no single-track Mainline card
+     * in this Division" — which says nothing about the rule under test.
+     *
+     * The fixture needs A Division with a card trains may not pass on, not one particular one, so
+     * it now takes the first seed that provides one. That is stable across any future retune, and
+     * it fails loudly if such a Division stops being reachable at all.
+     */
+    let s!: GameState;
+    let index = -1;
+    for (let seed = 3; seed < 200 && index < 0; seed++) {
+      s = createGame({
+        id: 'rear', seed,
+        config: {
+          mode: 'solitaire', days: 5, minCombinedRevenue: 0, maxCollisionsPerDay: 0, maxCollisionsTotal: 0, pvpCardsAllowed: false,
+          optionalRules: { reducedVisibility: false, employeeRotation: false, emergencyToolbox: false },
+        },
+        playerNames: ['bot'],
+      });
+      index = s.division.nodes.findIndex(
+        (n) => n.kind === 'mainline' && !mainlineProfile(n.card).trainsMayPass,
+      );
+    }
+    assert.ok(index >= 0, 'no seed under 200 deals a Division holding a single-track Mainline card');
     const node = s.division.nodes[index]!;
     assert.ok(node.kind === 'mainline');
     if (node.kind !== 'mainline') throw new Error('unreachable');
@@ -863,12 +969,116 @@ describe('Q13 — a train that catches the one ahead runs into it', () => {
     );
   });
 
-  it('leaves trains alone on a card that prints "trains may pass"', () => {
-    // Double Track and Uncontrolled Siding hold two trains because they HAVE two roads. Catching up
-    // there means going past, which is what the card is for. Without this the mechanic fired 0.41
-    // times a game while the bot never once granted clearance — the tell that they were all
-    // passing cards.
+  /**
+   * ENTERING an occupied region, as opposed to catching up inside the card (Gitea#3).
+   *
+   * A card can be ONE region wide — Plains, Double Track and Trestle all are — so a following train
+   * granted clearance is in the same place as the train ahead the moment it arrives. Nothing tested
+   * that: the catch-up check lives inside `stagesRemaining > 1`, which a one-Stage crossing never
+   * reaches, so entering behind another train on a Plains was silently free.
+   */
+  const enteringBehind = (card: MainlineKind, opts: { absSignals?: boolean } = {}) => {
+    const s = game();
+    const index = s.division.nodes.findIndex((n) => n.kind === 'mainline');
+    const node = s.division.nodes[index]!;
+    assert.ok(node.kind === 'mainline');
+    if (node.kind !== 'mainline') throw new Error('unreachable');
+    node.card = card;
+    node.transits = [];
+    if (opts.absSignals) node.absSignals = true;
+
+    /**
+     * THE TRAIN ALREADY THERE IS THE JUNIOR ONE, and that is what makes the situation reachable.
+     *
+     * Trains move lowest number first, so a card's occupant normally clears before anything behind
+     * it is even considered — put train 9 on the card and train 11 at the Division Point and 9 has
+     * gone by the time 11 enters. The conflict is a SUPERIOR train catching an inferior one that has
+     * not got out of the way yet, so the numbers run the other way round here.
+     */
+    const leader = s.freeTrays.pop()!;
+    s.trays.set(leader, {
+      id: leader, trainNumber: 11, trainIsExtra: false, engineAt: 0,
+      consist: [{ type: 'boxcar', loaded: false }], direction: 'east',
+      position: { at: 'mainline', index },
+    } as never);
+    const total = crossingStages(card, { trainSpeed: 'fast', direction: 'east', gradeUp: 'east', modifiers: [] });
+    node.transits.push({ tray: leader, stagesRemaining: total, stagesTotal: total, direction: 'east' });
+
+    // And the one arriving, held at the Division Point west of it.
+    const dp = s.division.nodes[index - 1];
+    assert.ok(dp && dp.kind === 'divisionPoint', 'expected a Division Point west of the first card');
+    if (!dp || dp.kind !== 'divisionPoint') throw new Error('unreachable');
+    const follower = s.freeTrays.pop()!;
+    s.trays.set(follower, {
+      id: follower, trainNumber: 9, trainIsExtra: false, engineAt: 0,
+      consist: [{ type: 'boxcar', loaded: false }], direction: 'east',
+      position: { at: 'divisionPoint', side: dp.side },
+    } as never);
+    dp.holding.push(follower);
+
+    /**
+     * THE SUPERINTENDENT LETS IT IN, which is the whole point.
+     *
+     * A same-direction train in the Subdivision is not an absolute bar — §8.1 makes it a judgment
+     * call, and `advance` stops and asks. Granting it is what puts one train in behind another, and
+     * §10 is then explicit that the wreck is the Superintendent's fault. So the fixture answers
+     * `allow: true` whenever it is asked, and the collision below is the consequence of that
+     * ruling rather than of a rule firing on its own.
+     */
+    s.clock.phase = 'mainline';
+    const events: GameEvent[] = [];
+    for (let i = 0; i < 6; i++) {
+      events.push(...advance(s).events);
+      if (s.clock.pendingDecision !== null) {
+        const who = s.clock.superintendent;
+        const r = applyIntent(s, who, { type: 'mainline.clearance', allow: true });
+        assert.ok(r.ok, `clearance refused: ${r.ok ? '' : r.code}`);
+        events.push(...r.events);
+      }
+    }
+    return { s, node, events, follower };
+  };
+
+  it('runs a train into the one ahead when it ENTERS an occupied region', () => {
+    const { events } = enteringBehind('plains');
+    const smash = events.find((e) => e.type === 'trainsDestroyed');
+    assert.ok(smash, 'a train entered a one-region card behind another and nothing happened');
+  });
+
+  it('holds it short instead when the card carries ABS Signals', () => {
+    // RAR: "ABS. This is played on a mainline card to prevent collisions. If a collision would
+    // normally occur, the train moving onto the card is instead held back."
+    const { events } = enteringBehind('plains', { absSignals: true });
+    assert.ok(!events.some((e) => e.type === 'trainsDestroyed'), 'ABS Signals did not prevent it');
+    assert.ok(
+      events.some((e) => e.type === 'trainHeld' && /ABS Signals/.test(e.reason)),
+      'nothing was held short of the train ahead',
+    );
+  });
+
+  it('takes the siding instead of colliding on an Uncontrolled Siding', () => {
+    // "If a train already exists when you arrive, you go in the second stage back — you are in the
+    // siding and are one behind the other train. This prevents a collision, since you are not in
+    // same exact location." So: no wreck, both trains on the card, and the newcomer paying the
+    // extra Stage for the detour.
+    const { node, events, follower } = enteringBehind('uncontrolledSiding');
+    assert.ok(!events.some((e) => e.type === 'trainsDestroyed'), 'the siding did not prevent a collision');
+    const mine = node.transits.find((t) => t.tray === follower);
+    assert.ok(mine, 'the arriving train never made it onto the card');
+    assert.equal(mine.stagesTotal, 2, 'it should have entered at the back of the card, not the front');
+  });
+
+  it('leaves trains alone on the one card that prints "trains may pass"', () => {
+    // Double Track holds two trains because it HAS two roads. Catching up there means going past,
+    // which is what the card is for.
+    //
+    // THE UNCONTROLLED SIDING USED TO BE IN THIS LIST AND IS NOT ANY MORE (Gitea#3). It holds two
+    // trains as well, but not by letting them share a place: the second one takes the siding and
+    // sits a region behind, which is what keeps them apart — "you are in the siding and are one
+    // behind the other train. This prevents a collision, since you are not in same exact location."
+    // Marked "may pass" it skipped the collision test entirely, so the siding did nothing at all and
+    // two trains could occupy the same region of it unchallenged.
     const passing = MAINLINE_PROFILES.filter((m) => m.trainsMayPass).map((m) => m.kind);
-    assert.deepEqual(passing, ['doubleTrack', 'uncontrolledSiding']);
+    assert.deepEqual(passing, ['doubleTrack']);
   });
 });

test/redaction.test.ts

@@ -99,6 +99,30 @@ describe('redaction — a seat\'s Frame never carries another seat\'s secrets',
     assert.ok(!serialized.includes(String(s.seed)), 'the seed value leaked into the Frame some other way');
   });
 
+  it('the tally that rides the Frame is aggregate counts, never a card id (Gitea#16)', () => {
+    // Gitea#16's statistics live on `GameState` and reach a remote client on the Frame, which is
+    // only safe because nothing in a Tally identifies a card. That is a property of what
+    // `tally.ts` chooses to count, and nothing in the type system enforces it — so it is asserted
+    // here, where a future counter that stashed a `cardId` "just for badges" would be caught.
+    for (const players of [3, 4]) {
+      const s = midGame(players, 5000 + players);
+      const secrets = new Set<string>([...s.decks.homeOffice]);
+      for (const hand of s.decks.hands.values()) for (const id of hand) secrets.add(id);
+      for (let viewer = 0 as PlayerIndex; viewer < players; viewer++) {
+        const serialized = JSON.stringify(snapshot(s, [], null, null, null, false, viewer).tally);
+        for (const cardId of secrets) {
+          assert.ok(!serialized.includes(`"${cardId}"`), `the tally carries card id "${cardId}"`);
+        }
+      }
+    }
+  });
+
+  it('every seat sees the SAME tally — it is the table\'s account, not a private one', () => {
+    const s = midGame(3, 5555);
+    const tallies = [0, 1, 2].map((p) => snapshot(s, [], null, null, null, false, p as PlayerIndex).tally);
+    for (const t of tallies) assert.deepEqual(t, tallies[0], 'the tally differs by seat');
+  });
+
   it('only the viewer\'s own hand and handCount are non-public — everything else matches across seats', () => {
     // The redaction surface is four fields (§7), not sixty event types. Cross-check that seats agree
     // on everything else a Frame carries about shared state.

test/server/session.test.ts

@@ -412,3 +412,122 @@ describe('the four transient signals (2026-08-23)', () => {
     assert.equal(later.scheduled, undefined, 'a reconnecting client was re-sent an old timetable flash');
   });
 });
+
+// ---------------------------------------------------------------------------
+
+describe('§3.3 extended play across the server (Gitea#11)', () => {
+  /**
+   * A one-Day game, so these tests reach the end of the timetable by actually PLAYING to it.
+   *
+   * There is no back door into a session's engine state and there should not be — `connect`,
+   * `intent`, `exportSave` and `summary` are the whole surface. So the clock is run down through the
+   * same calls a client makes, which has the side benefit of exercising the real path: what is under
+   * test here is the session's handling of the vote (the turn guard, the bots, the saved status),
+   * and reaching it any other way would prove less.
+   */
+  const oneDay: GameConfig = { ...config, days: 1 };
+
+  /** What seat `seat` can currently see. `connect` always yields a full Frame, never a delta. */
+  const frameOf = (session: GameSession, seat: PlayerIndex) => session.connect(seat).frame!;
+
+  /** Plays until the game stops asking for ordinary moves. Returns the Frame it stopped on. */
+  function playToTheEnd(session: GameSession, seats: PlayerIndex[]) {
+    let seq = 0;
+    for (let i = 0; i < 5_000; i++) {
+      const acting = seats.find((s) => session.connect(s).menu !== null);
+      if (acting === undefined) break;
+      const menu = session.connect(acting).menu!;
+      if (menu.options.length === 0) break;
+      if (!session.intent(acting, seq++, menu.options[0]!).accepted) break;
+    }
+    return { frame: frameOf(session, seats[0]!), seq };
+  }
+
+  it('stops to ask rather than ending, and every seat can see the question', () => {
+    const session = createSession(42, oneDay, ['Alice', 'Bob']);
+    const { frame } = playToTheEnd(session, [0, 1] as PlayerIndex[]);
+    assert.equal(frame.status, 'awaitingExtension', 'the game did not stop to ask');
+    assert.ok(frame.official, 'the official result did not reach the client');
+    assert.deepEqual(frame.extensionVotes, [null, null], 'the votes did not reach the client');
+  });
+
+  it('accepts the vote from a seat that is not the current actor', () => {
+    // `currentActor` is null once the game has stopped, so the ordinary turn guard would refuse
+    // every vote with NOT_YOUR_TURN. Both seats vote here and neither of them is the actor.
+    const session = createSession(42, oneDay, ['Alice', 'Bob']);
+    const { seq } = playToTheEnd(session, [0, 1] as PlayerIndex[]);
+    const a = session.intent(0 as PlayerIndex, seq + 1, { type: 'game.extend', player: 0, agree: true });
+    assert.equal(a.accepted, true, 'seat 0 could not vote');
+    const b = session.intent(1 as PlayerIndex, seq + 2, { type: 'game.extend', player: 1, agree: true });
+    assert.equal(b.accepted, true, 'seat 1 could not vote');
+
+    const after = frameOf(session, 0 as PlayerIndex);
+    assert.equal(after.extraDays, 1, 'a unanimous table was not given its Day');
+    assert.equal(after.status, 'active', 'play did not resume');
+  });
+
+  it('bots agree only once every human has, and never lead', () => {
+    // "Bots will not disagree with the human. Humans get to vote first" (Jesse, 2026-08-28).
+    const session = createSession(42, oneDay, ['Alice', 'Botty'], [1 as PlayerIndex]);
+    const { seq } = playToTheEnd(session, [0, 1] as PlayerIndex[]);
+    assert.equal(frameOf(session, 0 as PlayerIndex).status, 'awaitingExtension');
+    assert.equal(
+      frameOf(session, 0 as PlayerIndex).extensionVotes[1],
+      null,
+      'the bot voted before the human did',
+    );
+
+    session.intent(0 as PlayerIndex, seq + 1, { type: 'game.extend', player: 0, agree: true });
+    const after = frameOf(session, 0 as PlayerIndex);
+    assert.equal(after.extraDays, 1, 'the bot did not follow the human into another Day');
+    assert.equal(after.status, 'active');
+  });
+
+  it('a human refusal ends it, and no bot overrides that', () => {
+    const session = createSession(42, oneDay, ['Alice', 'Botty'], [1 as PlayerIndex]);
+    const { seq } = playToTheEnd(session, [0, 1] as PlayerIndex[]);
+    session.intent(0 as PlayerIndex, seq + 1, { type: 'game.extend', player: 0, agree: false });
+    const after = frameOf(session, 0 as PlayerIndex);
+    assert.equal(after.status, 'finished');
+    assert.equal(after.extraDays, 0);
+  });
+
+  it('REGRESSION: an all-bot game ends rather than hanging on the question', () => {
+    /**
+     * `driveBots` loops on `currentActor`, which is null the moment the game stops to ask — so it
+     * cannot cast the vote itself, and the bots' vote is driven separately. The first cut of that
+     * driver returned early when there were no humans to follow, on the reasoning that a bot-only
+     * table would decline through the ordinary path. It has no ordinary path: nothing ever asked
+     * the bots, and an all-bot session sat on the question for ever without reaching `finished`.
+     * Caught by `summary()`'s own "a finished game waits on nobody" test.
+     */
+    const session = createSession(4242, oneDay, ['A', 'B'], [0, 1] as PlayerIndex[]);
+    assert.equal(frameOf(session, 0 as PlayerIndex).status, 'finished', 'the bots never answered');
+    assert.equal(session.summary().waitingOn, null);
+    assert.equal(frameOf(session, 0 as PlayerIndex).extraDays, 0, 'bots voted themselves another Day');
+  });
+
+  it('saves a game awaiting its vote as ACTIVE, so a restart resumes it', () => {
+    // `server/index.ts` never loads a `finished` game back into memory. A game paused on the
+    // extension question is waiting on its table, not over — persisting it as finished would strand
+    // it on disk mid-decision.
+    const session = createSession(42, oneDay, ['Alice', 'Bob']);
+    playToTheEnd(session, [0, 1] as PlayerIndex[]);
+    assert.equal(frameOf(session, 0 as PlayerIndex).status, 'awaitingExtension');
+    assert.equal(session.exportSave().status, 'active', 'a paused game was saved as finished');
+    assert.equal(session.summary().status, 'active');
+  });
+
+  it('resumes a paused game from its history, vote and all', () => {
+    const session = createSession(42, oneDay, ['Alice', 'Bob']);
+    const { seq } = playToTheEnd(session, [0, 1] as PlayerIndex[]);
+    session.intent(0 as PlayerIndex, seq + 1, { type: 'game.extend', player: 0, agree: true });
+
+    const resumed = resumeSession(session.exportSave());
+    const a = frameOf(session, 0 as PlayerIndex);
+    const b = frameOf(resumed, 0 as PlayerIndex);
+    assert.deepEqual(b.extensionVotes, a.extensionVotes, 'the votes did not survive the replay');
+    assert.equal(b.extraDays, a.extraDays, 'the extra Day did not survive the replay');
+    assert.deepEqual(b.official, a.official, 'the official result did not survive the replay');
+  });
+});

test/session.test.ts

@@ -63,7 +63,13 @@ describe('a local session plays the same game as the calls it replaced', () => {
     assert.ok(turns > 50, `only ${turns} decisions — the game stalled`);
 
     const f = session.view();
-    assert.equal(f.status, 'finished');
+    /**
+     * `awaitingExtension` since Gitea#11, not `finished`: both loops stop when there is no actor,
+     * and a days-based ending now parks the game on the "play one more Day?" question rather than
+     * ending it outright. What this test is actually about is unchanged — the two sides reach the
+     * SAME position — and the assertion below is still the one carrying that.
+     */
+    assert.equal(f.status, 'awaitingExtension');
     assert.equal(f.status, view(game).status);
     assert.equal(f.day, view(game).day);
     assert.deepEqual(f.cells, view(game).cells, 'the board differs across the boundary');

test/setup.test.ts

@@ -63,56 +63,73 @@ const gameDealtWith = (startingHand: StartingHand, seed = 1234) =>
 
 describe('card catalogue (component 1)', () => {
   it('composes the deck from the design', () => {
-    // Transcribed from docs/Deck cards2.xlsx. The sheet's own totals are "Sum other 115" and
-    // "Total track 104", i.e. 219, plus 12 start cards for its grand total of 231.
+    // THE WHOLE CATALOGUE IS docs/Deck cards5.xlsx NOW (Gitea#14). Sheet 5's own totals are
+    // "Total track 48" and "Total other (in play) 107", i.e. 155 shuffled, plus 12 start cards for
+    // its grand total of 167. Card for card, 84 rows agree with it exactly and the only ones that
+    // do not are listed below — every one of them a deliberate hold, in one direction or the other.
     //
-    // We are at 235 rather than 219 because of three deliberate departures, all flagged in
-    // content.ts: 18 extra industry cards (Gap 12, industries 9 → 27), 7 extra office cards (Q12,
-    // offices 7 → 14), and the 8 sharp curves taken back OUT. The first two were tuned against a deck
-    // that had NO track in it, so both are due a re-measurement now that 96 track cards share the
-    // draw.
+    // EVERY COUNT IN THE CATALOGUE IS NOW THE SHEET'S. The two deliberate departures that used to
+    // sit here are gone with Gitea#14 — the Q12 office doubling (offices 14 → 7) and the Gap 12
+    // industry tripling (27 → 9) — because both were measured against a deck holding 96 track
+    // cards, and sheet 5 halves that. content.ts carries the measurements that decided it.
     //
-    // Two entries are dealt ZERO copies and kept in the catalogue so the design stays visible:
-    // Poling, whose effect is "TBD in the source", and the sharp curves, whose only difference from
-    // an ordinary curve was a Move cost nothing ever charged.
-    // DECK_SIZE is the CATALOGUE, 235. The deck actually dealt is smaller: the 22 opponent-directed
-    // cards are held back in every mode until they are implemented, so `buildDeck` returns 213.
-    assert.equal(DECK_SIZE, 235);
+    // We are at 143 rather than the sheet's 155 for ONE reason: the ten Safety, Event, Inspection
+    // and Space-use cards sheet 5 adds are not built, and stay out until they are (Jesse,
+    // 2026-08-26) — Cargo Theft, Civic Improvement, Civilian angel, Delayed Clearance, Flares 2,
+    // Robbery, Service Delays, Shipper complaints, Strike, Union Hall 2. Twelve copies in all.
+    //
+    // NOTHING RUNS THE OTHER WAY ANY MORE. Every card sheet 5 does not list is dealt ZERO copies
+    // rather than deleted, so the design stays visible and the rules stay implemented: the
+    // Telegraph/Telephone/Radio ladder, Facing Point Locks (both), Flying Switch, Section House and
+    // Vandalism, all removed from the design on purpose; Poling, whose effect the source records as
+    // "TBD"; and the sharp curves, whose only difference from an ordinary curve was a Move cost
+    // nothing ever charged — sheet 5 deals those zero too, so the catalogue and the design agree.
+    //
+    // DECK_SIZE is the CATALOGUE, 143. The deck actually dealt is smaller: the 20 opponent-directed
+    // cards are held back in every mode until they are implemented, so `buildDeck` returns 123.
+    assert.equal(DECK_SIZE, 143);
     assert.equal(buildDeck().length, SOLITAIRE_DECK_SIZE);
   });
 
   it('matches the design deck composition exactly', () => {
     const byCategory = Object.fromEntries(deckComposition().map((c) => [c.category, c.count]));
     assert.deepEqual(byCategory, {
-      // 96, not the sheet's 104: the 8 SHARP CURVES are dealt zero copies. The only thing that made
-      // one different from an ordinary curve was a Move cost that nothing ever charged, so they were
-      // geometric duplicates taking 8 draws. Kept in the catalogue at zero, as Poling is.
-      track: 96,
-      // 14, not the sheet's 7 — Q12 office density; see OFFICE_PROFILES.
-      office: 14,
-      // 27, not the sheet's 9 — Gap 12 industry density; see INDUSTRY_PROFILES.
-      industry: 27,
+      // Sheet 5's track counts exactly (Gitea#14): 16 straights, 8+8 curves, 8+8 turnouts, and the
+      // sharp curves dealt none — which is where the catalogue already had them, and where sheet 5
+      // now puts them too.
+      track: 48,
+      // The sheet's 7 — the Q12 doubling came out in Gitea#14; see OFFICE_PROFILES.
+      office: 7,
+      // The sheet's 9 — the Gap 12 tripling came out in Gitea#14; see INDUSTRY_PROFILES.
+      industry: 9,
       modifier: 23,
       train: 22,
-      spaceUse: 12,
-      enhancement: 18,
-      mainlineModifier: 7,
-      // 6, not 7 — Poling is dealt no copies until its rule is known.
-      maneuver: 6,
-      action: 10,
+      spaceUse: 11,
+      // 6 — the dispatching ladder and Facing Point Locks are dealt 0 copies (see
+      // ENHANCEMENT_CARDS), and Interlocking, Water column and ABS Signals came down to the sheet's
+      // single copies. What is left is the sheet's Enhancements exactly, bar Railroad crossing,
+      // which sheet 5 moved here from the Action cards and which is still counted there below.
+      enhancement: 6,
+      // 5 — Facing Point Locks came out of the Mainline modifiers too.
+      mainlineModifier: 5,
+      // 3 — Red Flags is the sheet's 3; Flying Switch and Poling are both dealt none.
+      maneuver: 3,
+      // 9 — Vandalism is dealt none. The rest are opponent-directed and held out of every deck.
+      action: 9,
     });
   });
 
   it('removes opponent-directed cards from a solitaire deck', () => {
     // Q6 took Space-use and Action cards out of solitaire, where they have no legal target. They are
     // now out of the COMPETITIVE deck too, until they are implemented: `checkPlay` answers both
-    // categories NOT_IMPLEMENTED, so dealing them would make 22 of 235 draws (9%) reject outright.
-    // 206, not 213: the 22 opponent-directed cards come out, and so do the SEVEN that exist only to
-    // answer them — Facing Point Locks (both the Enhancement and the Mainline modifier, 2 each), two
-    // Water Columns and one Overpass. A defence with nothing to defend against is the same dead draw
-    // as the attack would be. `SimpleCard.answers` names the pairing, so they return together.
-    assert.equal(SOLITAIRE_DECK_SIZE, 206);
-    assert.equal(DEFENCE_ONLY_COPIES, 7);
+    // categories NOT_IMPLEMENTED, so dealing them would be a dead draw.
+    // 121, not 123: the 20 opponent-directed cards come out, and so do the TWO that exist only to
+    // answer them — one Water Column and one Overpass. A defence with nothing to defend against is
+    // the same dead draw as the attack would be. `SimpleCard.answers` names the pairing, so they
+    // return together. It was seven until Gitea#14 dealt Facing Point Locks zero copies: a card at
+    // zero is already out, so it no longer needs holding back.
+    assert.equal(SOLITAIRE_DECK_SIZE, 121);
+    assert.equal(DEFENCE_ONLY_COPIES, 2);
     for (const c of DEFENCE_ONLY_CARDS) {
       assert.ok(c.answers, `${c.name} is held back without saying what it answers`);
       assert.ok(
@@ -131,11 +148,12 @@ describe('card catalogue (component 1)', () => {
   });
 
   it('deals track FROM the deck, at the sheet\'s counts', () => {
-    // Column B of Deck cards2.xlsx, "Number in Deck": 32 straights, 16+16 curves, 16+16 turnouts —
-    // and 4+4 sharp curves, which are dealt none. An earlier reading took the sheet's LAST column,
-    // "Track Per Player" (26), as a separate stack outside the deck; it is the sheet's 104 shared out
-    // among four players, not a second pile.
-    assert.equal(TRACK_IN_DECK, 96);
+    // Column B of Deck cards5.xlsx, "Number in Deck": 16 straights, 8+8 curves, 8+8 turnouts, and
+    // 0+0 sharp curves. Sheet 2 had each of those at double, which is what the deck dealt until
+    // Gitea#14. An earlier reading took the sheet's LAST column, "Track Per Player" (26), as a
+    // separate stack outside the deck; it is the sheet's total shared out among four players, not a
+    // second pile.
+    assert.equal(TRACK_IN_DECK, 48);
     const deck = buildDeck();
     for (const t of TRACK_CARDS) {
       const n = deck.filter(
@@ -146,11 +164,26 @@ describe('card catalogue (component 1)', () => {
   });
 
   it('makes track the largest category in the deck', () => {
-    // 96 of 235. Building a district is paid for in the industry or train you did not draw, which
+    // 48 of 121. Building a district is paid for in the industry or train you did not draw, which
     // is the whole reason it matters that track is a card rather than a private supply.
+    //
+    // This asked for a THIRD of the deck until Gitea#14, which was only ever a rule of thumb. It
+    // asks its own question now — is track still the biggest single thing you can draw — plus a
+    // loose band, because the exact share is not settled yet and should not be pinned as though it
+    // were. Sheet 5 puts track at 48 of the 155 cards it would have you shuffle, i.e. 31%; we read
+    // 40% because the Space-use, Safety, Event and Inspection cards are held out, which concentrates
+    // everything that is left. The share falls TOWARDS the sheet as those land, so the band is set
+    // to hold across that whole journey rather than to be re-edited at each step.
     const deck = buildDeck();
-    const track = deck.filter((c) => c.kind.kind === 'track').length;
-    assert.ok(track > deck.length / 3, `track is only ${track} of ${deck.length} cards`);
+    const counts = new Map<string, number>();
+    for (const c of deck) counts.set(c.kind.kind, (counts.get(c.kind.kind) ?? 0) + 1);
+    const track = counts.get('track') ?? 0;
+    for (const [kind, n] of counts) {
+      if (kind === 'track') continue;
+      assert.ok(track > n, `${kind} has ${n} cards against track's ${track}`);
+    }
+    const share = track / deck.length;
+    assert.ok(share > 0.28 && share < 0.45, `track is ${(share * 100).toFixed(1)}% of the deck`);
   });
 
   it('has 12 timetabled trains, odd westbound and even eastbound', () => {

test/sim.test.ts

@@ -191,15 +191,26 @@ describe('the revenue chain works end to end (regression)', () => {
     // only because unloads were mis-scored as completed loads after one Laborer action instead of
     // four. Correcting that dropped mean revenue from 24.8 to ~4.6 and the win rate to zero, so
     // "did anyone win" is no longer a safe proxy for "does freight work".
+    /**
+     * TWO HUNDRED GAMES, up from forty (Gitea#3). Completed freight loads got scarcer when the
+     * Mainline went onto the region model, and measurably so — on these seeds: 40 games yield 0
+     * loads, 80 yield 3 (1 game), 120 yield 10 (4 games), 200 yield 21 (9 games). Forty could no
+     * longer reach the precondition it exists to establish.
+     *
+     * WHY it got scarcer is not settled and is worth someone's attention rather than a guess — the
+     * change speeds crossings up, which ought to put MORE trains through a district, not fewer.
+     * Freight share of gross fell from 8% to 5% over 100 games across the same change. Recorded in
+     * TODO.md under Play Balance; the assertion itself is untouched.
+     */
     const report = simulate({
-      games: 40,
+      games: 200,
       length: 'standard',
       mode: 'solitaire',
       players: ['bot'],
       policy: developerBot,
     });
     const freight = report.perGame.reduce((n, g) => n + g.revenue.freightLoad, 0);
-    assert.ok(freight > 0, 'no freight load completed across 40 games');
+    assert.ok(freight > 0, 'no freight load completed across 200 games');
   });
 
   it('grows the Office Area off the Running Track, on either side', () => {
@@ -351,7 +362,24 @@ describe('end-of-game statistics', () => {
      * rule that has become unreachable. Exempted by name so the other forty-odd event checks stay live,
      * and so removing this line is what proves the bot has been fixed.
      */
-    const KNOWN_UNREACHABLE_BY_THE_BOT = ['event flyingSwitch'];
+    /**
+     * RED FLAGS JOINS IT (Gitea#3), and for the same reason — the rule is reachable and the bot will
+     * not take it.
+     *
+     * MEASURED over 600 games: `maneuver.redFlags` is OFFERED 4,212 times, first in game 5 — so the
+     * rule is live and constantly available. The bot PLAYS it 4 times, first in game 252. At 200
+     * games this canary sees nothing and calls it unreachable, which is the opposite of the truth.
+     *
+     * It got rarer for two compounding reasons, neither of them a broken rule: Gitea#14 took Red
+     * Flags from 5 copies to the sheet's 3, and Gitea#3 shortened most crossings to a single Stage,
+     * so the window in which a train is STANDING on a Mainline card — the only place the card may be
+     * played — is now usually one Stage wide.
+     *
+     * The bot's unwillingness is the thing worth fixing, and it is in TODO.md under Bot Performance.
+     * Exempted BY NAME so the other forty-odd checks stay live, and so deleting this line is what
+     * proves the bot has learned to use it.
+     */
+    const KNOWN_UNREACHABLE_BY_THE_BOT = ['event flyingSwitch', 'event redFlagsSet'];
     const found = anomalies(report.perGame);
     const never = found
       .filter((a) => a.severity === 'never')
@@ -643,12 +671,17 @@ describe('the bot builds sidings that are actually sidings (regression)', () =>
     // Measured across 100 games: tank cars boarded a train 0.07 times a game and were dropped by a
     // crew ZERO times, while boxcars were 67% of every drop — and 23 of 79 waiting loads were
     // sitting at an industry that wanted a tank.
-    // A HUNDRED GAMES, not thirty — the comment above says the original measurement used 100, and
-    // the sample has to be that big to mean anything: measured now, a tank is set out in 3% of games
-    // and a reefer in 5%. Thirty games passed on luck and stopped the moment the opening deal moved
-    // which cards a seed puts in reach. Deterministic seeds, so this either holds or it does not.
+    // THREE HUNDRED GAMES, up from a hundred, because the sheet's industry density (Gitea#14) makes
+    // the rare commodities much rarer. Measured on these exact seeds, the game at which each type is
+    // first set out by a crew: caboose 5, boxcar 12, hopper 37, reefer 44, coach 64, **tank 216**.
+    //
+    // A Refinery is one card in a hundred and fifty now, so a tank moving at all needs that card
+    // drawn, placed, reached and worked. The old sample of 100 stopped covering it — not because the
+    // rule broke, but because the deck did what the sheet asks. The sample follows the measurement
+    // rather than the assertion being softened: tank is still the strict test, for the reason below.
+    // Deterministic seeds, so this either holds or it does not.
     const dropped = new Set<string>();
-    for (let i = 0; i < 100; i++) {
+    for (let i = 0; i < 300; i++) {
       const s = createGame({
         id: `cs-${i}`,
         seed: 1000 + i * 7919,
@@ -998,6 +1031,12 @@ describe('the bot does not lay track that cannot work (regression)', () => {
     // A TIE-BREAKER rather than a veto, so this is a rate and not a zero: forbidding it outright
     // measured WORSE (-0.62 revenue a game), while preferring the cleaner of two equally good
     // placements measured better and cut these from 28% of pieces to 7%.
+    //
+    // AND IT STAYS A TIE-BREAKER. Gitea#15 was filed as "track placements must connect" and briefly
+    // became a rule here; RAR reversed it on review (2026-08-26) — a rail may stop dead against its
+    // neighbour, and such a stub is useful as a siding to park cars on. What the engine must refuse
+    // is a TRAIN crossing the gap, which is `exploreMoves`' job and is tested in `track.test.ts`.
+    // So laying one of these is a preference, exactly as it was, and the rate below is the bar.
     let laid = 0;
     let dead = 0;
     /**
@@ -1094,8 +1133,14 @@ describe('the freight figures count both halves (regression)', () => {
      * The subject here is the INSTRUMENT — does `freightUnload` count Revenue earned rather than
      * unloads started — and `unloads > 0` is only the precondition that makes the comparison mean
      * anything. Widening the sample restores the precondition without weakening the assertion.
+     *
+     * A HUNDRED AND FIFTY DEALS, up from forty, for the same reason as the commodity test above:
+     * Gitea#14 put the deck on the sheet's industry density and completed unloads went with it.
+     * Measured on these seeds, the first deal to EARN unload Revenue is number **46**, and 19 deals
+     * in 400 earn any — so forty could not reach the precondition it exists to establish. 150 clears
+     * it with room, and the assertion itself is untouched.
      */
-    for (let i = 0; i < 40; i++) {
+    for (let i = 0; i < 150; i++) {
       const seed = 1000 + i * 7919;
       const s = createGame({
         id: `fu-${seed}`, seed,

test/tally.test.ts

@@ -0,0 +1,214 @@
+/**
+ * THE EVENT TALLY — Gitea#16's statistics, and the one property they depend on.
+ *
+ * "I don't know if we keep statistics on…" is the question the issue opens with. Nothing was being
+ * kept; `GameState.tally` now is, folded from the event stream at the two places every event passes
+ * through (`engine/tally.ts` explains which and why).
+ *
+ * The property that matters is EXACTLY ONCE. A statistic folded twice reads high and a statistic
+ * folded nowhere reads zero, and both are indistinguishable from a quiet game when you are looking
+ * at a results screen. So the central test here does not assert particular numbers: it plays real
+ * games, collects every event the engine emitted along the way, counts them independently, and
+ * checks the tally against that count. A fold hooked in the wrong place fails it whatever the seed.
+ */
+
+import { describe, it } from 'node:test';
+import assert from 'node:assert/strict';
+
+import { advance } from '../src/engine/advance.ts';
+import { tallyEvent } from '../src/engine/tally.ts';
+import { applyIntent } from '../src/engine/apply.ts';
+import { STAGES_PER_DAY } from '../src/engine/content.ts';
+import { legalActions } from '../src/engine/legal.ts';
+import { createGame } from '../src/engine/setup.ts';
+import { developerBot } from '../src/sim/bot.ts';
+import type { GameEvent } from '../src/engine/events.ts';
+import type { GameConfig, GameState } from '../src/engine/state.ts';
+
+const baseConfig = (over: Partial<GameConfig> = {}): GameConfig => ({
+  mode: 'solitaire',
+  days: 3,
+  minCombinedRevenue: 0,
+  maxCollisionsPerDay: 0,
+  maxCollisionsTotal: 0,
+  pvpCardsAllowed: false,
+  optionalRules: { reducedVisibility: false, employeeRotation: false, emergencyToolbox: false },
+  ...over,
+});
+
+/** Plays a whole game with the developer bot, keeping every event the engine produced. */
+function playKeepingEvents(
+  seed: number,
+  over: Partial<GameConfig> = {},
+  names = ['Jesse'],
+): { state: GameState; events: GameEvent[] } {
+  const state = createGame({ id: 'g', seed, config: baseConfig(over), playerNames: names });
+  const events: GameEvent[] = [];
+  for (let i = 0; i < 20_000; i++) {
+    const r = advance(state);
+    events.push(...r.events);
+    if (state.status === 'finished') break;
+    if (!r.needsInput) continue;
+
+    // The bot declines an extension, so this terminates on the timetable it was dealt (Gitea#11).
+    const actor =
+      state.status === 'awaitingExtension'
+        ? state.extensionVotes.findIndex((v) => v === null)
+        : state.clock.pendingDecision !== null
+          ? state.clock.superintendent
+          : state.clock.currentActor;
+    if (actor === null || actor < 0) break;
+    const options = legalActions(state, actor);
+    if (options.length === 0) break;
+    const applied = applyIntent(state, actor, developerBot.choose(state, actor, options));
+    if (!applied.ok) break;
+    events.push(...applied.events);
+  }
+  return { state, events };
+}
+
+/** Counts events the way `tally.ts` should have, without sharing any of its code. */
+function countIndependently(events: GameEvent[]) {
+  let sum = 0;
+  const n = (type: GameEvent['type']): number => events.filter((e) => e.type === type).length;
+  for (const e of events) {
+    if (e.type === 'carsCoupled' || e.type === 'carsDropped') sum += e.stock.length;
+  }
+  return {
+    trainsCompleted: n('trainCompleted'),
+    loadsCompleted: n('loadCompleted'),
+    unloadsCompleted: n('unloadCompleted'),
+    loadsStarted: n('loadStarted'),
+    unloadsBegun: n('unloadBegan'),
+    passengersBoarded: n('passengersBoarded'),
+    passengersDetrained: n('passengersDetrained'),
+    cardsDrawn: n('cardDrawn'),
+    cardsPlayed: n('cardPlayed'),
+    cardsDiscarded: n('cardDiscarded'),
+    officeUpgrades: n('officeUpgraded'),
+    flyingSwitches: n('flyingSwitch'),
+    extrasStarted: n('extraStarted'),
+    secondSections: n('secondSectionOrdered'),
+    trainsHeld: n('trainHeld'),
+    trainsDiverted: n('trainDiverted'),
+    expediteFaults: n('expediteFault'),
+    facilitiesUnjammed: n('facilityUnjammed'),
+    dispatchBonusesUsed: n('dispatchBonusUsed'),
+    clearancesRequested: n('clearanceRequested'),
+    switchedCars: sum,
+  };
+}
+
+describe('the tally counts every event exactly once (Gitea#16)', () => {
+  for (const seed of [1, 7, 42, 116956197]) {
+    it(`agrees with an independent count of the event stream — seed ${seed}`, () => {
+      const { state, events } = playKeepingEvents(seed);
+      const want = countIndependently(events);
+      const t = state.tally;
+
+      assert.equal(t.trainsCompleted, want.trainsCompleted, 'trains through the Division');
+      assert.equal(t.loadsCompleted, want.loadsCompleted, 'loads made up');
+      assert.equal(t.unloadsCompleted, want.unloadsCompleted, 'loads broken');
+      assert.equal(t.loadsStarted, want.loadsStarted, 'loads started');
+      assert.equal(t.unloadsBegun, want.unloadsBegun, 'unloads begun');
+      assert.equal(t.passengersBoarded, want.passengersBoarded, 'passengers boarded');
+      assert.equal(t.passengersDetrained, want.passengersDetrained, 'passengers detrained');
+      assert.equal(t.cardsDrawn, want.cardsDrawn, 'cards drawn');
+      assert.equal(t.cardsPlayed, want.cardsPlayed, 'cards played');
+      assert.equal(t.cardsDiscarded, want.cardsDiscarded, 'cards discarded');
+      assert.equal(t.officeUpgrades, want.officeUpgrades, 'offices upgraded');
+      assert.equal(t.flyingSwitches, want.flyingSwitches, 'flying switches');
+      assert.equal(t.extrasStarted, want.extrasStarted, 'extras started');
+      assert.equal(t.secondSections, want.secondSections, 'second sections');
+      assert.equal(t.trainsHeld, want.trainsHeld, 'trains held');
+      assert.equal(t.trainsDiverted, want.trainsDiverted, 'trains diverted');
+      assert.equal(t.expediteFaults, want.expediteFaults, 'expedite faults');
+      assert.equal(t.facilitiesUnjammed, want.facilitiesUnjammed, 'facilities unjammed');
+      assert.equal(t.dispatchBonusesUsed, want.dispatchBonusesUsed, 'dispatch bonuses');
+      assert.equal(t.clearancesRequested, want.clearancesRequested, 'clearances requested');
+      assert.equal(t.carsCoupled + t.carsDropped, want.switchedCars, 'cars switched');
+    });
+  }
+
+  it('actually counted something — a tally of zeroes would pass the check above vacuously', () => {
+    // The trap `stats.ts` warns about in its own doc comment: "this never happened" is a finding,
+    // not something to scroll past. A fold hooked nowhere at all agrees perfectly with an
+    // independent count of an event stream nobody looked at, so the exactly-once tests above cannot
+    // catch it on their own.
+    //
+    // ASSERTED ON WHAT THE DEVELOPER BOT ACTUALLY DOES, which is not much: `node src/sim/harness.ts
+    // 12 standard` means 1.1 Revenue per game at an 8% freight share and loses every game on the
+    // revenue floor, and it goes whole games without coupling a single car. That is a known
+    // property of the bot (TODO.md, Bot Performance) and not this fold's business — so this test
+    // asserts on traffic and cards, which happen in every game, rather than on switching, which
+    // would make it a bot-strength test wearing a statistics test's clothes.
+    const { state, events } = playKeepingEvents(1);
+    assert.ok(events.length > 500, `only ${events.length} events — the game barely ran`);
+    assert.ok(state.tally.cardsDrawn > 0, 'no cards were drawn all game');
+    assert.ok(state.tally.trainsCompleted > 0, 'no train ever left the Division');
+    assert.ok(state.tally.cardsPlayed > 0, 'no card was ever played');
+  });
+
+  it('splits Revenue into what was earned and what was given back', () => {
+    // Reconciliation is the real assertion and it holds for any game, earned or not: gained minus
+    // lost IS the score the engine kept. Seed 42 is named because it is one where Revenue actually
+    // moves in both directions — it earns 1 and gives back 5 to a collision — so the two halves are
+    // being told apart rather than both sitting at zero.
+    for (const seed of [1, 7, 42]) {
+      const { state } = playKeepingEvents(seed);
+      const me = state.tally.byPlayer[0]!;
+      assert.equal(
+        me.revenueGained - me.revenueLost,
+        state.players[0]!.revenue,
+        `seed ${seed}: gained minus lost does not reconcile with the score the engine kept`,
+      );
+    }
+    const { state } = playKeepingEvents(42);
+    const me = state.tally.byPlayer[0]!;
+    assert.ok(me.revenueGained > 0, 'seed 42 earned nothing — the gained half is not being counted');
+    assert.ok(me.revenueLost > 0, 'seed 42 lost nothing — the lost half is not being counted');
+  });
+
+  it('records a Circus set-up as the one-off it is, not as a streak', () => {
+    /**
+     * `trainStoodStill` is NOT "this train did not move this Stage". It fires only for a train whose
+     * profile sets `stopEarnsPoint` — the X18 Circus — and `advance.ts` claims it once per train
+     * with `stopPointClaimed`, so it can never fire twice for the same one.
+     *
+     * Gitea#16 asks for "longest engine sat on a siding" and its comment assumed this event would
+     * answer it. It cannot, and a streak folded from it would have read "1 Stage" for ever. Pinned
+     * here so that the day a real per-Stage signal is added, whoever adds it finds this test rather
+     * than the old wrong assumption.
+     */
+    const s = createGame({ id: 'g', seed: 1, config: baseConfig(), playerNames: ['Jesse'] });
+    const feed = (e: GameEvent): void => tallyEvent(s, e);
+    feed({ type: 'trainStoodStill', trainNumber: 18, where: '(0,0)' });
+    assert.deepEqual(s.tally.circusStops, [{ trainNumber: 18, where: '(0,0)' }]);
+    assert.ok(!('longestStand' in s.tally), 'a streak that cannot be computed is being reported');
+  });
+});
+
+describe('the official result freezes the tally with it (Gitea#11 + #16)', () => {
+  it('records the statistics as they stood when the timetable ran out', () => {
+    const { state } = playKeepingEvents(1);
+    assert.ok(state.official, 'no official result was recorded');
+    // Nothing was played after the ending in this game, so the two agree — which is the check that
+    // the freeze happens AFTER the last batch of events is folded rather than before it.
+    assert.equal(state.official!.tally.trainsCompleted, state.tally.trainsCompleted);
+    assert.ok(state.official!.tally.cardsDrawn > 0, 'the frozen tally is empty');
+  });
+
+  it('keeps the frozen copy still while the live tally moves on', () => {
+    const s = createGame({ id: 'g', seed: 1, config: baseConfig(), playerNames: ['Jesse'] });
+    s.clock.day = s.config.days + 1;
+    s.clock.stage = STAGES_PER_DAY;
+    s.clock.phase = 'shiftChange';
+    advance(s);
+    const frozen = s.official!.tally.cardsDrawn;
+
+    applyIntent(s, 0, { type: 'game.extend', player: 0, agree: true });
+    tallyEvent(s, { type: 'cardDrawn', player: 0, source: 'homeOffice', cardId: 'x' });
+    assert.equal(s.tally.cardsDrawn, frozen + 1, 'the live tally did not move');
+    assert.equal(s.official!.tally.cardsDrawn, frozen, 'the frozen tally moved with it');
+  });
+});

test/track.test.ts

@@ -540,6 +540,100 @@ describe('placement and drop-off', () => {
     assert.ok(!canPlaceAt(area, at(-1, 1), straight()), 'an east-west straight cannot meet a 45° leg');
   });
 
+  describe('a rail that stops dead against its neighbour (Gitea#15)', () => {
+    /**
+     * REPORTED, THEN REVERSED. The issue first read "if a card is placed in that space, it MUST
+     * connect", against a right-hand curve laid at (1,-1) with an Ice House above it and a turnout
+     * with a north-facing leg below. **RAR reviewed it and ruled the other way (2026-08-26): the
+     * placement is fine, and a stub like that has a use — a siding to park cars on.**
+     *
+     * "We need to confirm, however, that trains are not allowed to traverse from the turnout below
+     * to that right-hand curve since the tracks do not connect." That is what these tests are: the
+     * rule lives in MOVEMENT, not in placement.
+     *
+     * The save cannot carry this any more — Gitea#14 took the deck from 206 cards to 121, so its
+     * card ids no longer exist and the history stops at the first `card.play`. The geometry is what
+     * mattered, and it is rebuilt here directly.
+     */
+    /** A Modifier card — an Ice House. Not track: no ports on any edge. */
+    const modifierCard = (): TrackCard => ({
+      geometry: { kind: 'modifier', modifier: 'iceHouse' },
+      baseOperationalRail: false,
+      standing: [],
+      standingWest: 0,
+      facility: null,
+      modifiers: [],
+      enhancements: [],
+    });
+
+    /** A Grocer's Warehouse — a Facility, so a plain east-west through track. */
+    const warehouse = (): TrackCard => ({
+      geometry: { kind: 'facility', facility: 'grocersWarehouse' },
+      baseOperationalRail: true,
+      standing: [],
+      standingWest: 0,
+      facility: null,
+      modifiers: [],
+      enhancements: [],
+    });
+
+    /**
+     * The reported district. `withCurve` puts the disputed right-hand curve on the square; without
+     * it, the square is empty and the placement itself is under test.
+     *
+     * The turnout's leg goes NORTH on the `nw_se` diagonal; the curve is `ne`, which is `ne_sw` and
+     * has no south port at all. Two reasons the two do not join, either of which is enough.
+     */
+    const board = (withCurve: boolean): OfficeArea =>
+      areaFrom(
+        {
+          [coordKey(at(0, -1))]: turnout({ stem: 'w', through: 'e', diverge: 'n' }),
+          [coordKey(at(0, 0))]: officeCard(),
+          [coordKey(at(1, 0))]: warehouse(),
+          [coordKey(at(2, -1))]: modifierCard(),
+          ...(withCurve ? { [coordKey(at(1, -1))]: curve('ne') } : {}),
+        },
+        at(0, 0),
+      );
+
+    it('allows the reported placement, which connects on one side and nothing else', () => {
+      // RAR's ruling. The curve joins the warehouse to its east; its north leg faces an Ice House
+      // that carries no rail, and the turnout below faces its portless south edge. All legal.
+      assert.ok(canPlaceAt(board(false), at(1, -1), curve('ne')), 'the reported play was refused');
+    });
+
+    it('will not let a train cross from the turnout below onto that curve', () => {
+      // The confirmation the issue actually asks for. Running west out of the Office and into the
+      // turnout, the 45° leg goes north — and stops at the curve's blank south edge.
+      const dests = reachableDestinations(ctxFor(board(true)), at(0, 0), 'w');
+      assert.ok(!has(dests, 1, -1), 'a train drove across rails that do not meet');
+    });
+
+    it('still reaches the curve from the side that DOES join', () => {
+      // Otherwise the test above would pass on a card that is simply unreachable, which proves
+      // nothing. East of the curve is the warehouse, and east-west edges always meet.
+      const dests = reachableDestinations(ctxFor(board(true)), at(1, 0), 'w');
+      assert.ok(has(dests, 1, -1), 'the curve was unreachable from the side that joins');
+    });
+
+    it('will not let a train cross a north edge onto a card with no rail at all', () => {
+      // The Ice House above. A Modifier is scenery beside the rails — Jesse confirmed a rail may
+      // point at a building — so what stops a train is the same `joins` test, not a placement rule.
+      const dests = reachableDestinations(ctxFor(board(true)), at(1, 0), 'w');
+      assert.ok(!has(dests, 2, -1), 'a train drove into an Ice House');
+    });
+
+    it('still allows an exit that faces a BLANK square', () => {
+      // Unchanged by the reversal, and the reason a district can grow at all: a turnout laid on the
+      // Running Track with nothing yet beside its diverging leg is a perfectly good play.
+      const area = areaFrom({ [coordKey(at(0, 0))]: officeCard() }, at(0, 0));
+      assert.ok(
+        canPlaceAt(area, at(0, 1), turnout({ stem: 'w', through: 'e', diverge: 'n' })),
+        'a turnout whose leg faces open space was refused',
+      );
+    });
+  });
+
   it('refuses a card that connects to nothing', () => {
     const area = areaFrom({ [coordKey(at(0, 0))]: officeCard() }, at(0, 0));
     assert.ok(!canPlaceAt(area, at(3, 3), straight()), 'orphaned track is never legal');

test/web.test.ts

@@ -21,14 +21,16 @@ import { URLSearchParams as NodeURLSearchParams } from 'node:url';
 import { cardDescription, cardName, describeIntent, variantLabel } from '../src/sim/view.ts';
 import { variantsFor } from '../src/engine/track.ts';
 import { divisionSvg, officeSvg } from '../src/sim/board-svg.ts';
-import { ENHANCEMENT_RULES } from '../src/engine/content.ts';
-import { dayEndHtml, facilitiesHtml, timetableHtml } from '../src/web/panels.ts';
+import { ENHANCEMENT_RULES, STAGES_PER_DAY } from '../src/engine/content.ts';
+import { dayEndHtml, facilitiesHtml, resultsHtml, timetableHtml } from '../src/web/panels.ts';
 import { turnChartHtml } from '../src/sim/turnchart.ts';
 import { fieldSelectors } from '../src/web/settings-form.ts';
 import { record, renderHtml } from '../src/sim/replay.ts';
 import type { Frame } from '../src/sim/view.ts';
 import { snapshot } from '../src/sim/view.ts';
 import { createGame as createEngineGame } from '../src/engine/setup.ts';
+import { advance as advanceEngine } from '../src/engine/advance.ts';
+import type { GameConfig } from '../src/engine/state.ts';
 import {
   overHandLimit,
   actionGroups,
@@ -104,6 +106,18 @@ describe('the browser game plays', () => {
   it('reaches the end of a game through the same calls the page makes', () => {
     const { game, turns } = playThrough(77);
     assert.ok(turns > 50, `only ${turns} decisions — the game stalled`);
+
+    /**
+     * The timetable runs out and the game STOPS TO ASK rather than ending (Gitea#11) — `currentActor`
+     * is null there, which is where `playThrough` breaks. Declining through `submit` finishes it,
+     * which is worth doing here rather than merely asserting the pause: this test exists to prove a
+     * whole game is playable through the calls the page makes, and since Gitea#11 the last of those
+     * calls is the vote.
+     */
+    assert.equal(game.state.status, 'awaitingExtension');
+    assert.ok(game.state.official !== null, 'an ended game must have recorded its official result');
+    assert.ok(submit(game, { type: 'game.extend', player: 0, agree: false }, 0), 'the page cannot decline');
+
     assert.equal(game.state.status, 'finished');
     assert.ok(game.state.outcome !== null, 'a finished game must have an outcome');
   });
@@ -947,17 +961,29 @@ describe('the page explains itself', () => {
     // A played card becomes a cell with a name on it — "turnout", "Freight House", "waiting area" —
     // and the explanation that was visible while it sat in hand disappears exactly when it starts
     // mattering. Play a long way in so every card kind reaches the grid.
-    const game = newGame(111);
-    for (let i = 0; i < 400; i++) {
-      if (currentActor(game) === null) break;
-      const { options } = actionGroups(game);
-      if (options.length === 0) break;
-      const pick = options.find((o) => o.type === 'card.play' && o.placement) ?? options[0]!;
-      if (!submit(game, pick)) break;
+    /**
+     * THE SEED IS SEARCHED FOR, not written down. This took seed 111 flat and asserted its board
+     * ended up with more than four cards on it. Gitea#3 changed how fast trains cross, which changes
+     * how a game unfolds, and 111 stopped building enough of a district — so the test failed on its
+     * own precondition rather than on anything about tooltips.
+     *
+     * It needs A well-built board, not one particular one, so it takes the first seed that gives it.
+     */
+    let game = newGame(111);
+    for (let seed = 111; seed < 211; seed++) {
+      game = newGame(seed);
+      for (let i = 0; i < 400; i++) {
+        if (currentActor(game) === null) break;
+        const { options } = actionGroups(game);
+        if (options.length === 0) break;
+        const pick = options.find((o) => o.type === 'card.play' && o.placement) ?? options[0]!;
+        if (!submit(game, pick)) break;
+      }
+      if (view(game).cells.length > 4) break;
     }
 
     const cells = view(game).cells;
-    assert.ok(cells.length > 4, 'not enough of the board was built to be a real check');
+    assert.ok(cells.length > 4, 'no seed under 211 built enough of a board to be a real check');
     for (const c of cells) {
       assert.ok(c.what.length > 0, `(${c.row},${c.col}) ${c.label} has no explanation`);
       // camelCase on the board is the failure that keeps recurring — labels AND descriptions.
@@ -1586,7 +1612,40 @@ describe('the static build', () => {
       (target['onclick'] as (() => void) | null)?.();
       return true;
     };
-    assert.ok(clickVerb(hand, 'play'), 'no card in hand offers a play verb');
+    /**
+     * TRY EVERY PLAY BUTTON, NOT JUST THE FIRST — a card offering "play" does not necessarily play
+     * ONTO THE BOARD.
+     *
+     * This clicked the first play verb it found and then asserted the board had lit up. A train card
+     * plays to the timetable and a Department discard to the piles, so neither lights a square, and
+     * whether the first playable card in this deal happens to be a track or facility card is luck.
+     * Gitea#14's deck counts re-dealt seed 555, the first play verb landed on Train 2, and the test
+     * failed claiming the page highlighted nothing — when the page was right and the card simply had
+     * no square to point at.
+     *
+     * So it clicks each play button in turn until the board lights, which is the property the test
+     * is named for. It still fails loudly if NO card in hand can light a square.
+     *
+     * EACH BUTTON IS CLICKED EXACTLY ONCE. Picking a card is a toggle, so clicking one to check it
+     * and then clicking it again inside the loop would UNPICK it — which is how the first draft of
+     * this managed to fail on a deal whose very first play card was a good one.
+     */
+    const playButtons = (el: Record<string, unknown>): Record<string, unknown>[] => {
+      const fn = el['querySelectorAll'] as (s: string) => Record<string, unknown>[];
+      return fn
+        .call(el, 'button.cardact')
+        .filter((n) => (n['dataset'] as Record<string, string>)['verb'] === 'play');
+    };
+    assert.ok(playButtons(hand).length > 0, 'no card in hand offers a play verb');
+    let litTheBoard = false;
+    for (const target of playButtons(hand)) {
+      (target['onclick'] as (() => void) | null)?.();
+      const drawn = String(grid['innerHTML']);
+      if ((grid['highlighted'] as () => unknown[])().length > 0 || /data-ghost="/.test(drawn)) {
+        litTheBoard = true;
+        break;
+      }
+    }
 
     // Without the board stylesheet every shape is drawn black on a near-black background: the page
     // looks empty even though the markup is perfect.
@@ -1599,7 +1658,10 @@ describe('the static build', () => {
     assert.match(html, /data-cell="/, 'the board drew no addressable cards');
     const lit = (grid['highlighted'] as () => unknown[])();
     const ghosts = /data-ghost="/.test(html);
-    assert.ok(lit.length > 0 || ghosts, 'picking a card highlighted nothing on the board');
+    assert.ok(
+      litTheBoard || lit.length > 0 || ghosts,
+      'no card in hand, picked in turn, ever highlighted a square on the board',
+    );
 
     /**
      * POINTING AT THE SQUARE A BUTTON MEANS — wired on the emitted bundle, not asserted off the menu.
@@ -1884,8 +1946,10 @@ describe('the static build', () => {
     const asked = new Set([...src.matchAll(/\$\('([a-zA-Z][\w-]*)'\)/g)].map((m) => m[1]!));
     const html = readFileSync(join(dist, 'play.html'), 'utf8');
     const present = new Set([...html.matchAll(/id="([a-zA-Z][\w-]*)"/g)].map((m) => m[1]!));
-    // `again` is created by the game-over screen before it is looked up.
-    present.add('again');
+    // Created by the ending screen (`renderEnding`) before they are looked up, so they are never in
+    // the served HTML: `again` starts a new game, `results` reopens the results dialog, and the two
+    // `extend-*` buttons are the extension vote (Gitea#11).
+    for (const id of ['again', 'results', 'extend-yes', 'extend-no']) present.add(id);
     for (const id of asked) {
       assert.ok(present.has(id), `main.ts asks for #${id}, which the page does not contain`);
     }
@@ -2762,18 +2826,44 @@ describe('the Division map shows the whole route', () => {
     }
   });
 
-  it('seats 1 to 4 players without overlapping or spilling off the canvas', () => {
-    // A row, two facing rows, a horseshoe and a square. The layout is geometry with no visual
-    // feedback loop, so this is the only thing standing between a change and an unreadable board.
+  it('draws 1 to 4 players as ONE row, west to east, without overlapping or spilling', () => {
+    /**
+     * Gitea#18. This used to check "a row, two facing rows, a horseshoe and a square" — the route
+     * was laid out around a table, on the reasoning that players sit around one. It cost three
+     * reports, and the one that decided it was that **east stopped being to the right**: a player's
+     * east could be drawn south, west or north depending which lane their district landed in, on a
+     * map whose whole job is saying which way a train is going.
+     *
+     * So the property is now stronger and much simpler to state — every cell on one row, ordered
+     * west to east — which is exactly what makes "east is right" true and is the thing that would
+     * silently regress if anyone reintroduced lanes. The overlap and canvas checks are kept: the
+     * layout is geometry with no visual feedback loop.
+     */
     for (const players of [1, 2, 3, 4]) {
       const svg = divisionFor(players);
       const vb = /viewBox="0 0 (\d+) (\d+)"/.exec(svg);
       assert.ok(vb, `${players}p produced no viewBox`);
       const W = Number(vb![1]);
       const H = Number(vb![2]);
-      const rects = [...svg.matchAll(/class="bs-dcell[^"]*"[^>]*><rect x="([\d.]+)" y="([\d.]+)" width="([\d.]+)" height="([\d.]+)"/g)]
-        .map((m) => ({ x: +m[1]!, y: +m[2]!, w: +m[3]!, h: +m[4]! }));
-      assert.ok(rects.length >= 5, `${players}p drew only ${rects.length} cells`);
+      const rects = [...svg.matchAll(/class="bs-dcell bs-d(\w+)[^"]*"[^>]*><rect x="([\d.]+)" y="([\d.]+)" width="([\d.]+)" height="([\d.]+)"/g)]
+        .map((m) => ({ kind: m[1]!, x: +m[2]!, y: +m[3]!, w: +m[4]!, h: +m[5]! }));
+      // WDP · (ML · Office) × players · ML · EDP — including the Mainline card before the East
+      // Division Point, which the issue's own sketch left out.
+      assert.equal(rects.length, 2 * players + 3, `${players}p drew ${rects.length} cells`);
+      assert.equal(rects[0]!.kind, 'dp', `${players}p does not start at a Division Point`);
+      assert.equal(rects[rects.length - 1]!.kind, 'dp', `${players}p does not end at a Division Point`);
+      assert.equal(rects[rects.length - 2]!.kind, 'ml', `${players}p has no Mainline card before the East DP`);
+
+      // ONE ROW: every cell at the same y, and x strictly increasing.
+      const ys = new Set(rects.map((r) => r.y));
+      assert.equal(ys.size, 1, `${players}p drew ${ys.size} rows — the Division must be one`);
+      for (let i = 1; i < rects.length; i++) {
+        assert.ok(
+          rects[i]!.x > rects[i - 1]!.x,
+          `${players}p: cell ${i} is not east of the one before it — east is no longer to the right`,
+        );
+      }
+
       for (let i = 0; i < rects.length; i++) {
         const a = rects[i]!;
         assert.ok(
@@ -2789,6 +2879,52 @@ describe('the Division map shows the whole route', () => {
     }
   });
 
+  it('draws no office-area detail on the Division map, but keeps the trains', () => {
+    // Gitea#18: "Division map should not show any office area detail (no limits, no running track,
+    // etc.)" — an Office used to expand into its whole Running Track, Limits to Limits, so this map
+    // carried every straight, turnout and Limits sign of every district and grew sideways as
+    // districts were built. One cell per district now.
+    //
+    // The trains stay: "trains within the office area should definitely be represented on the
+    // division map", split into the A/D register and the crews switching below it.
+    const s = createEngineGame({
+      id: 'div-collapse', seed: 7,
+      config: {
+        mode: 'solitaire', days: 5, minCombinedRevenue: 0, maxCollisionsPerDay: 0, maxCollisionsTotal: 0, pvpCardsAllowed: false,
+        optionalRules: { reducedVisibility: false, employeeRotation: false, emergencyToolbox: false },
+      },
+      playerNames: ['Solitaire'],
+    });
+    const area = areaOf(s, 0);
+    area.tier = 'terminal';
+    const place = (id: string, n: number, coord: { row: number; col: number }, ad: boolean): void => {
+      s.trays.set(id, {
+        id, trainNumber: n, trainIsExtra: false, engineAt: 0,
+        consist: [{ type: 'boxcar', loaded: true }],
+        direction: 'east', facing: 'e',
+        position: { at: 'grid', seat: 0, coord }, movesUsed: 0,
+      } as never);
+      if (ad) area.adOccupancy.push(id);
+    };
+    place('ad1', 9, area.officeCoord, true);
+    place('sw1', 7, { row: area.runningRow - 1, col: 0 }, false);
+
+    const svg = divisionSvg(snapshot(s, [], null).division);
+    // ONE district cell, not one per Running Track card.
+    const districts = [...svg.matchAll(/class="bs-dcell bs-drun/g)].length;
+    assert.equal(districts, 1, `the district drew as ${districts} cells`);
+    assert.doesNotMatch(svg, /Limits/, 'a Limits sign reached the Division map');
+
+    // Both trains are on it, in two registers — the A/D one above the crew switching below.
+    const chips = [...svg.matchAll(/class="bs-train" data-tip="(T\d+)[^"]*"><rect x="[\d.]+" y="([\d.]+)"/g)]
+      .map((m) => ({ label: m[1]!, y: +m[2]! }));
+    const ad = chips.find((c) => c.label === 'T9');
+    const sw = chips.find((c) => c.label === 'T7');
+    assert.ok(ad, 'the train holding an A/D track is not on the map');
+    assert.ok(sw, 'the crew switching in the district is not on the map');
+    assert.ok(sw.y > ad.y, 'the switching crew should be drawn BELOW the A/D register, not beside it');
+  });
+
   it('keeps every roster chip inside the Office cell it belongs to, at any occupancy', () => {
     // "Two Trains, One Card": sizing the cell by OCCUPANCY moved the East Division Point sideways
     // every time an A/D track filled or cleared. Sizing by CAPACITY (docs/plans/switching-paths.md)
@@ -3163,12 +3299,97 @@ describe('the Day rolling over says so (Gitea#10)', () => {
 
   it('gives the page the dialog to fill', () => {
     const html = readFileSync(join(dist, 'play.html'), 'utf8');
-    for (const id of ['dayenddlg', 'dayendbody']) {
+    for (const id of ['dayenddlg', 'dayendbody', 'resultsdlg', 'resultsbody']) {
       assert.ok(html.includes(`id="${id}"`), `play.html has no #${id}`);
     }
   });
 });
 
+// ---------------------------------------------------------------------------
+
+describe('the end-of-game results screen (Gitea#16)', () => {
+  /**
+   * A finished game, built by running the clock off the end of a real one rather than by hand — the
+   * screen reads `official`, and only the engine writes that.
+   */
+  const finished = (over: Partial<GameConfig> = {}, names = ['Solitaire']): Frame => {
+    const s = createEngineGame({
+      id: 'results', seed: 4021,
+      config: {
+        mode: 'solitaire', days: 5, minCombinedRevenue: 12,
+        maxCollisionsPerDay: 0, maxCollisionsTotal: 0, pvpCardsAllowed: false,
+        optionalRules: { reducedVisibility: false, employeeRotation: false, emergencyToolbox: false },
+        ...over,
+      },
+      playerNames: names,
+    });
+    s.clock.day = s.config.days + 1;
+    s.clock.stage = STAGES_PER_DAY;
+    s.clock.phase = 'shiftChange';
+    advanceEngine(s);
+    return snapshot(s, [], null);
+  };
+
+  it('never prints a raw enum at the player', () => {
+    // The bug the issue opens on: `GAME OVER — revenueFloor` is an internal identifier, shown at
+    // the one moment the game has the player's whole attention.
+    for (const f of [finished(), finished({ minCombinedRevenue: 0 })]) {
+      const html = resultsHtml(f);
+      for (const raw of ['revenueFloor', 'daysElapsed', 'collisionFloor']) {
+        assert.ok(!html.includes(raw), `the results screen prints the raw reason "${raw}"`);
+      }
+    }
+  });
+
+  it('says why the game ended, in a sentence, with the game\'s own numbers in it', () => {
+    const html = resultsHtml(finished());
+    assert.ok(html.includes('closed short'), `the revenue-floor ending is not explained:\n${html}`);
+    assert.ok(html.includes('<b>12</b>'), 'the floor that was missed is not named');
+  });
+
+  it('reports the rules the game was actually dealt under', () => {
+    const html = resultsHtml(finished({ mode: 'coop' }, ['Ada', 'Bo']));
+    assert.ok(html.includes('The rules in play'), 'the rules section is missing');
+    assert.ok(html.includes('Co-op'), 'the mode is not reported');
+    assert.ok(html.includes('per load'), 'the pay rates are not reported');
+  });
+
+  it('reports the railroad from the tally, not from nothing', () => {
+    const html = resultsHtml(finished());
+    assert.ok(html.includes('Trains through the Division'), 'the tally section is missing');
+  });
+
+  it('names the winner the OFFICIAL result named, not whoever leads now (Gitea#11)', () => {
+    const f = finished({ mode: 'competitive', minCombinedRevenue: 0 }, ['Ada', 'Bo']);
+    // Ada won at the timetable; Bo overtakes during extended play. The frozen result stands.
+    f.official = {
+      day: 5,
+      outcome: { result: 'win', winner: 0, reason: 'daysElapsed' },
+      revenues: [9, 2],
+      collisionsTotal: 0,
+      tally: f.tally,
+    };
+    f.extraDays = 3;
+    f.players = [
+      { index: 0, seat: 0, name: 'Ada', revenue: 9, hand: 3 },
+      { index: 1, seat: 1, name: 'Bo', revenue: 40, hand: 2 },
+    ];
+    const html = resultsHtml(f);
+    assert.ok(html.includes('Ada takes the Division'), `the official winner is not the headline:\n${html}`);
+    assert.ok(html.includes('winner'), 'nobody is marked as the winner in the standings');
+    assert.ok(html.includes('After the timetable'), 'extended play is not reported at all');
+    assert.ok(html.includes('3 more Days'), 'the extra Days are not counted');
+    assert.ok(
+      html.indexOf('Ada takes the Division') < html.indexOf('After the timetable'),
+      'the informational section is above the official result',
+    );
+  });
+
+  it('leaves out the extended-play section entirely when nothing was extended', () => {
+    assert.ok(!resultsHtml(finished()).includes('After the timetable'));
+  });
+});
+
 describe('every file the build needs is actually in the repo (regression)', () => {
   it('does not gitignore a source page', () => {
     // REGRESSION. `.gitignore` carried `replay*.html` to catch the throwaway files generated at the