v0.8.0.15 wrote a Quickstart for a tester who has never played and then left it
in docs/, where a tester does not look — reachable only by somebody who already
has the repository. Nobody handed the box had it. build-web.ts now copies it to
dist/quickstart.md and the splash page offers it under the three doors, as a
line rather than a fourth door: reading the guide is not a way to play, and
giving it equal weight in that grid would say it is.
COPIED, NEVER RE-WRITTEN. The Markdown document stays the one copy. A
hand-written HTML twin drifts from it on the first edit, which is the failure
#15a was raised about and precisely what the v0.8.0.15 pass spent itself
undoing. It is served as PLAIN TEXT, which is honest rather than good — tables
render as pipes and the links do not click. Rendering it into a styled page
wants a small Markdown converter and is filed as TODO #109; build-web.ts's
comment names that number rather than gesturing at "the next step", so the file
and the worklist cannot drift the way the references just did.
Two things had to be true and tsc checks neither, so both are tests. The href on
the splash page and the filename the build writes are two strings with nothing
connecting them: rename the document and the build quietly publishes nothing
while the page keeps offering a link that 404s. And a .md file must not arrive
as a download — the server's MIME fallback is application/octet-stream, which a
browser saves instead of displaying, so the link would have handed a tester a
file to save rather than a page to read. '.md' is in http.ts's table now, and
the test reads that table out of the source rather than asserting on a copy of
it, which would pass while the real one was wrong.
VERIFIED AGAINST A RUNNING SERVER, not only compiled: 200,
text/plain; charset=utf-8, the guide's own first lines, and the splash link
resolving.
THE sortsCars COMMENT. Asked after v0.8.0.15 whether everything now agreed, and
the audit turned up one place that did not — the field's own doc comment named
the card's printed text as though it were the flag's meaning. Nothing reads it
to permit a sort; its two readers, resolveExtraStart in apply.ts and the
enumeration in legal.ts, both ask whether this is the one Mainline card with a
Yard Limit and therefore the one an Extra may be made up and started on. Comment
only, and worth the bump because of where it is: it is what a developer reads
before using the flag, and it is the likeliest source of the sentence v0.8.0.15
had to correct off the board. The name is kept for its link to the card face and
the comment now says outright that the name is not the meaning.
The five references and docs/design.md read v0.8.0.16. They describe this build
because the audit re-checked them against it, not because the number was swept
forward — a stamp bumped without a reading is worth less than none.
1001 fast tests and 35 sim tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MUizFYCMHRWhbWwXhp7WPR
The four hand-written references brought up to the game as it actually runs,
ahead of the next testing round, plus a Quickstart to hand a tester who has never
played. They had not been touched since v0.6.2 — a month and two minor versions —
and each now says at the top which build it describes.
ONE LIVE BUG CAME OUT OF THE PASS. mainlineDescription told players "Cars may be
sorted into any new order here" on the Interchange. It is the printed capability
and has never been implemented: nothing reads sortsCars to permit a sort, and its
one live use is marking the card an Extra may be made up on, because it is the
Mainline card with a yard. That sentence is not only documentation — view.ts
renders it as a Mainline card's `what`, so it is what a player reads on the
board, and the generated reference printed a "Sorts cars: yes" column beside it.
A card advertising a button that does not exist sends a player hunting for it and
then concluding the game is broken.
What the documents had wrong, all of it verified against the code rather than
read for tone: the victory model in the Rules book (firstToTarget /
highestAfterDays and the target-bearing length presets stopped existing in
2026-08 — it is a free days count and a combined floor of 3 x players x days);
"there is no lobby, no server, no multiplayer"; crossing time in mph rather than
regions; Extras launched automatically eastbound; the Uncontrolled Siding listed
as a passing card; industry track length taken from the box count; and a
"Sister Trains" optional rule that never existed. The Home deck's counts table
came out under TODO #15a — Jesse's own ruling that counts move with balance —
and it had been wrong for a month, which is the argument made twice.
The Home and Mainline deck references are restructured to explain how a deck is
USED and to defer every per-card table to rules/as-built.md. Duplicating it by
hand is precisely the drift #15a was raised about: as-built needed no correction
beyond the Interchange, because build:cards regenerates it and a test fails when
the checked-in file disagrees. Everything hand-maintained around it had drifted;
it had not.
docs/design.md, the index everything starts from, said v0.4.3, "what is not: the
server", and 493 tests. It now also lists the player-facing references, which it
never has, so the Quickstart is findable at all.
999 fast tests and 35 sim tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DmdqqCNoiqE7GBo6wthBnR
Gitea#33. A session token is the only identity the game has, and it lives in
exactly one place the player controls: their browser's localStorage, scoped to
the origin they joined at. Lose it — a cleared profile, a private window, a
different browser — and the seat is unreachable while the game runs on and the
session sits intact on disk. Reported from the table: of two humans in one game
the host reloaded straight back in, the joiner met an empty lobby.
Diagnosed before it was fixed, and two server-side theories of mine were
retracted on the evidence: no storage key changed in 0.8.0.11, nothing in the
app deletes the secret or name, create and join both call persistSession, that
game's sessions.json held both seats, and it resumed with 80 intents replayed.
Both players used the same URL, so it was not a second origin either.
The fix is a recovery link. An administrator mints a code for a named seat
(admin-gated: deciding somebody lost a seat is a judgement no route can make);
the player opens the link and the page trades the code for the token over a
POST, then strips it from the address bar. The link never carries the token —
lobby-and-sessions.md §1 says keep it out of URLs, and a recovery link is
exactly what gets pasted into a chat. Single use, 30-minute expiry, held in
memory because a restart dropping them is the right failure.
server/claims.ts is a pure store, so single use, lazy expiry and one identical
answer for unknown/spent/expired codes are tested rather than asserted. The
admin game listing gained seatedPlayers — the seats a human holds a token for,
read from the session map rather than guessed from player names — so the
StartOS action can offer real players instead of bot chairs.
No rule changed: `git diff v0.8.0.11..v0.8.0.12 -- src/engine/` is empty, so
games in progress resume.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017nnuCv8UodHucFfx3LWEoX
Arrivals name whose Office they reached, and no longer tell every seat they can
work the train. The turn chart follows the animation queue, so being five behind
looks five behind across the whole screen rather than half of it. Pause sits
beside Skip and preserves the dwell a held step still owed. A one-render look at
another player's Office Area. The district summary counts the board being shown.
LIMITS is printed beneath its card instead of through its border. The Mainline
region divider is visible. Only Hilly mentions FAST/SLOW, because it is the only
card that reads it. A passenger Modifier on a Whistle Post reports itself dormant
rather than claiming the facility "only receives". An automatic phase says what
the Division is doing instead of answering by negation. The version appears once
in the header rather than twice on every .s9pk. Save files carry the join code,
the Stage and the date.
The New Train phase, reviewed before being changed: the make-up panel now says
what the train STILL needs rather than only what its card calls for, explains
that a player adds one car before the round passes on, marks the train being
loaded on the Division map, and gives an addable car in the yard the same amber
every other clickable thing on the page wears.
Reasoning, measurements and the reports behind each are in CHANGELOG.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017nnuCv8UodHucFfx3LWEoX
Both repositories allow anonymous clone — checked rather than assumed: info/refs
for git-upload-pack answers 200 for each, git-receive-pack answers 401. So
everything committed here is public, and tracked files are supposed to carry
placeholders rather than real hosts.
Ten mentions added while building v0.8.0 are now "the test server" or "the target
hardware", across CHANGELOG.md, the common-board plan's three deferral banners,
sim/pacing.ts, and two test files. Prose and comments only, no behaviour; the
quotes are untouched, because what was said about bot pacing is the part worth
keeping.
Left alone deliberately: nineteen older mentions in entries about v0.7.5, v0.7.6
and v0.7.8 and in TODO.md, since rewriting a changelog after the fact makes the
record less true; and scripts/deploy-web.ts, where the host is the functional
default for FB_URL rather than prose — turning that into a required variable
changes how deploying works and wants deciding on its own.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X6cF1iYvJ1kNmzYBzu4QX6
contradiction
Two things found by playing v0.8.0.1, neither in the mechanism itself.
?pace= never worked. index.html's doors are play.html?lobby and
play.html?solitaire, so arriving through the splash replaces the query string and
the play page only ever saw ?lobby — a whole game was played at 1x while believing
it was at 7x. v0.8.0 shipped that parameter as the only way to change speed and the
game's own front door destroyed it. There is a control on the play screen now,
beside zoom, persisted per viewer; the doors carry pace through as well, so the URL
lever is honest for handing two playtesters different speeds. PACE_LEVELS moved to
sim/pacing.ts with DWELL and MAX_PACE — the tuning surface in one file, and
testable. The committed default is unchanged: what it should be is a question for a
game played at a speed that took effect.
And the Day-end dialog said "0 today, 2 in all". advance.ts increments the Day and
then zeroes collisionsToday, and noteDayEnd() fires when the Day goes up — so the
dialog reporting the Day that just finished was drawn from the very frame in which
that Day's count was reset. Reproduced on four of five seeds before changing
anything. The count is captured at the rollover now; it is not derivable on the
client, because in multiplayer the push announcing the new Day is the same push
that carries the reset. And "today" was the wrong word regardless: it names the Day
instead — "Collisions: 2 on Day 1, 2 in all".
Unrelated to v0.8.0 — that one has been wrong since the dialog was built for
Gitea#10, and needed somebody to play a Day with a collision in it and then read
the summary.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X6cF1iYvJ1kNmzYBzu4QX6
TODO #13, #15 and #18 — Gitea#20 steps 2-4 pointed at a seated player's own screen.
Every accepted intent, and every automatic phase that does anything, becomes an
ordered presentation step. A bot's whole switching turn used to land in one push;
now it arrives as a run of steps, the district panel follows whoever is acting,
and a [N behind] … [Skip] row says how far the board is from the game.
Solitaire runs the same path — one collector inside submit(), which both session
kinds already funnel through — which is where its automatic phases finally get a
visible beat.
Dwell is assigned by kind: switching holds the screen, turn bookkeeping costs
nothing, and the clock turning over earns the beat. Tunable per viewer without a
rebuild, and off entirely at pace 0.
Also: switching was the one class of action logging unattributed, and now names
its train. Reasoning, measurements and the three things that turned out wrong are
in CHANGELOG.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X6cF1iYvJ1kNmzYBzu4QX6
Housekeeping before v0.8.0: the answer to "anything else that should be
looked at first". One real hole, one stale document, and my own leavings.
#102 — `npm test` passed green on a type error. `pretest` ran
`scripts/build-web.ts`, which invokes `tsc --ignoreConfig` against three
web entry points, so it saw only what those three transitively import and
under a WEAKER configuration than tsconfig.json — no
`noUncheckedIndexedAccess`, no `exactOptionalPropertyTypes`,
`--types ''`. It never saw `src/server/` or a single file under `test/`.
Demonstrated rather than argued: a planted
`const DELIBERATE_TYPE_ERROR: number = 'not a number';` in
src/server/session.ts gives `npm run typecheck` a TS2322 and `npm test` a
clean `# fail 0`. `pretest` is `tsc --noEmit && node
scripts/build-web.ts` now, and the same error exits 1 with the tests
never running.
This mattered THIS week rather than generally: v0.8.0 is steps 2-7 of the
common board — display stream, credentials, persistence, Chromium
supervisor — which is almost entirely src/server/, exactly the half the
test command could not see.
#103 — the plan had drifted from the code it is the source for.
docs/plans/jitsi-common-board.md was written 2026-08-27, still said "No
implementation has been performed", and is what steps 2-7 get built from.
Step 1 shipped across four releases since, so every "current code
finding" under it described a fault that is now fixed — a document
reading as present tense and nine days stale sends the next reader to fix
things twice.
Measured: its PublicFrame sketch lists four properties never built
(protocolVersion, config, scoring, deckCounts) and omits 28 that exist,
and the shape is the real difference — the implementation is FLAT where
the plan grouped things into objects, so a renderer written from the
sketch would not compile. The plan now says so at the top and at step 1,
names src/sim/view.ts and the redaction allow-list as the authority,
keeps the original sketch for its reasoning, and calls out
`protocolVersion` as unbuilt rather than dropping it quietly — step 2 is
the reconnecting display stream and is the first thing that would want
one.
One step-1 item is STRUCK OFF rather than built: "add the Red Flag holder
to the public player projection". The premise does not hold here.
`decks.redFlags` is written once, in setup.ts, from
`optionalRules.emergencyToolbox`, and never again — `redFlag.play` emits
`phaseEnded` and does not spend it — so every player holds one or none
does, decided before the deal. A per-player `redFlagHeld` would be one
already-public option copied N times, while telling every reader of the
common board that it varies by player and might change mid-game. Worse
than the absence. Pinned by test so it is not re-raised from the plan.
Four dead imports removed, all mine: `HAND_LIMIT` left unused in
apply.ts, view.ts and web/game.ts when 0.7.9.6 consolidated the three
copies of the §6.2 test, and `actingPlayer` in web/game.ts, dead since
0.7.9.5 made `currentActor` delegate. Finding them re-measured #46:
`tsc --noUnusedLocals` now reports 40, up from 29 on 2026-08-30. That
entry's "without the flag this list simply regrows" is a measurement
rather than a forecast now. The other 36 and the flag stay open.
946 tests pass, up from 943. No behaviour changes: three new tests pin an
invariant, and the rest is a build command, dead imports and a document.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y5boPxP6JHRYMm8adXaF5R
Both faults are in what 0.7.9.4 had just built, and both are the same
shape: a second copy of an answer that agreed with the first until it
didn't.
#96 — the §3.3 vote has no actor, and the screen named one anyway. The
vote is PARALLEL: every un-voted seat may vote at any moment, in any
order, one refusal ends it, and `apply.ts` says where it accepts one that
there is no actor to be. The turn chart named the last seat to move
before the timetable ran out — no more claim on the vote than anybody
else — directly above a tally correctly showing three seats outstanding.
The cause is worth more than the symptom. `currentActor(game)`
(`web/game.ts`) guarded on `status !== 'active'`; `currentActorOfState`
(`sim/view.ts`), added the same day in #95 and the one the frame calls,
did not, so it handed back whatever `clock.currentActor` was left
holding. The view now carries the guard and `currentActor` delegates to
it. That matters more than the tidiness: `currentActor` is what REFUSES
an intent, so a screen answering differently tells the table to wait on a
player the server would turn away.
The fourth of this class after Gitea#21, #22 and #94 — but the first
found by asking a view helper its question in a state the game is not
`active` in, which is the generalisation and is cheaper than finding the
fifth the same way.
#97 — narration reaches a seat once, by one path. `Frame.lines` carried
the whole log on every push to every seat, and nothing read it:
`RemoteSession` accumulates from `push.lines` alone and its `lines()`
returns that accumulator, so the log was serialised into every frame,
grew all game, and was discarded on arrival while `linesSince` sent the
same text correctly beside it.
The duplicate was masking a bug rather than merely wasting bandwidth.
`connect()` cleared `lastFrame` but not `sentLines`, so a reconnecting
seat was told "nothing new since your last push" while the browser it
answered had just reloaded from an EMPTY accumulator — the history panel
came back blank, mid-game, with the server holding the whole log. So the
two halves are one change, and the plan's instruction taken alone ("stop
passing the full game log into `frameFor()`") would have deleted a real
behaviour rather than a duplicate.
Every remaining reader of `Frame.lines` was checked before the field was
emptied: all of them are the solitaire and replay path, which builds
Frames through `snapshot()` directly and never goes near a session.
One test was wrong before the code was. The first draft of the reconnect
test connected inside its own fixture, so both sides of the comparison
were the empty array and it passed against the broken server. Each test
now asserts its premise is non-empty before comparing.
Also: `docs/plans/jitsi-common-board.md` is committed. It was never added
— not ignored, just missed — while TODO.md cites it twice as the plan for
all of v0.8.0 and the last two releases were built from it, so a clone
got a TODO pointing at a file that did not exist.
917 tests pass, up from 909.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y5boPxP6JHRYMm8adXaF5R
The generated card reference now carries the half TODO #15a said was the
point of generating it: every Enhancement's `live` / `dormantSolo` /
`unbuilt` status — whether its printed effect actually resolves yet — and
the opponent-directed Action and Space-use cards, none of which is dealt
in any deck. A transcription cannot carry either fact.
Card counts are removed throughout, as ruled: they move with play balance
so a document printing them is stale on the next retune. Where a count
matters it is a yes/no "is this dealt at all", which is a fact about the
design rather than the current tuning. #88 closes with it — it asked
whether `card-reference.md`'s industry rows were stale, deliberately
without rewriting them since Laborer counts are a balance decision. They
are; nothing in the engine changed; that file is simply no longer where
anyone looks. The balance question it guarded is #70.
Save compatibility becomes a general rule in `README.md` § Design notes
rather than a fact restated per version: a save is a list of moves and
reopens by being re-played through the CURRENT rules, so any change that
makes a once-legal move illegal stops an older one there — a deck change
being the likeliest breaker. It fails safe every time. #40 generalised,
#32's version-specific note dropped, #52 carries the ruling that
versioned replays are a post-1.0 question.
#94 opened: a Red Flag set out at an Office's Limits holds the next train
from that side and is drawn nowhere. `DivisionView`'s office node has no
`redFlag` field and `board-svg.ts` never mentions one, so after the single
log line announcing it there is nothing on screen. Same class as Gitea#21
and #22, and already on the common board's step 1 list.
No engine change. 897 tests pass, unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01E3Qk7uresKCHksdZajXCLg
Both leaks were found while planning the common board (Gitea#20 step 1),
and both are live multiplayer bugs with or without that display, so they
are fixed now rather than with 0.8.0.
`game.log` is one shared list and `linesSince(seat)` slices it with no
per-seat filter, so every line reaches every player. It carried the SEED
in the opening line of each multiplayer game — the whole future of the
deal — and the NAME OF A CARD DRAWN BLIND from the face-down Home Office
deck. Solitaire deliberately keeps both: a one-seat table has nobody to
leak to, the seed is what a bug report quotes, and a player's own history
naming their own draw is the record. A Department slot is face up and
stays named. The drawer still learns their card through `justDrawn`,
which already goes to that seat alone.
Neither was found by a test. Every test in `redaction.test.ts` passes an
empty log, so the whole of narration has sat outside the redaction net
since the net was built. Both now have tests there; TODO #91 carries what
is still owed and supersedes #78, which described a gap that had already
been closed and never mentioned this one.
`docs/rules/` had no current description of the game, and `content.ts`
named `card-reference.md` as the file that carries what the cards say —
a file whose own banner says not to use its numbers, describing the
v0.4.5 deck where 3/4 is a Mail-Express with three coaches. Every file in
that directory is a deliberate historical record, so none of them is
rewritten. `as-built.md` is new and GENERATED from the same catalogues
the engine instantiates from, with a test that re-runs the generator and
fails when the checked-in file disagrees. A hand-written replacement
would have drifted the same way, for the same reason.
TODO.md: #32 closed — the playtest migration note did its job and the
jump is made; the durable fact it carried is kept. #78 retired in favour
of #91. The "play it at a table" section now records that 0.7.4-0.7.9
were test-run without change requests, and that more testing comes at the
end of the 0.7.9 series.
897 tests pass, up from 891.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01E3Qk7uresKCHksdZajXCLg
Gitea#17 — a 45° leg is an end of the west-to-east row, so backing into a cut
through a curve's south leg no longer couples it back to front. The same
assumption left a crew's own cut standing when it pulled out through a leg,
which is the "cars left behind" report we had failed to reproduce.
Gitea#14 — every count is docs/Deck cards5.xlsx. Track halved, and the Q12
office doubling and Gap 12 industry tripling both come out with it: they were
measured against a deck with twice the track, and keeping them at the sheet's
track count wipes out the reefer chain entirely. 84 rows now match card for
card; the ten Safety, Event and Inspection cards it adds are not built and are
held out. Cards the sheet no longer lists are dealt zero copies rather than
deleted, so their rules stay implemented.
Gitea#15 — RAR reversed it: a rail may stop dead against its neighbour and the
placement is legal. What must hold is that no train crosses the gap, which was
already true and is now pinned against the reported board.
Gitea#3 — the printed speeds are scenery. A card costs one Stage per printed
region and where a train STARTS is what varies; Fast/Slow is read on Hilly
alone. Entering a one-region card behind another is a collision now, which is
what ABS exists to prevent, and ABS no longer holds trains silently.
Gitea#18 — the Division draws as one row, west to east, with no office-area
detail. East is finally always to the right.
Closes#3Closes#14Closes#15Closes#17Closes#18
Four issues off the Gitea tracker, all of them things a player saw at the board. Reasoning for
every item, and what was verified how: CHANGELOG.md.
- Gitea#8: X22 Pee-Dee refused every caboose, including the one it was made up with, so setting it
out stranded the train. All six cabooses are minted loaded because §2.2's "coloured is loaded,
white is empty" doubles as a piece count in the supply table; one read of the flag took that
literally. A caboose carries the crew, not freight, so it is never a load.
- Gitea#10: a Day turns over inside the phases that run themselves, so it passes between one click
and the next — and both transient signals fade before a player reading the board notices. A modal
stops and waits, carrying the standings, the Days left and the combined target. Suppressed on the
first frame, on Undo stepping back across a rollover, and on the Day the game ends.
- Gitea#9, which SUPERSEDES Gitea#6 from three days ago: a Timetabled train may be tossed face-up
to a Department slot, where a rival may pick it up — the second half of the ruling needed no code,
since that is where every discard already goes. An Extra still may not. A New Game setting on this
line (discardTimetabled, on by default), the plain rule on the 0.4.9 line.
- Gitea#2 is not an engine bug: the rules are implemented exactly, and running the coach pool dry is
Jesse's ruling to keep — "part of the strategy". What was wrong is that the game said nothing. A
blocked platform now gives its reason, from the engine's own predicate, including how many coaches
are stranded in Classification and what brings them back.
The same four ship as v0.4.9g on the playtest line.
Closes#2Closes#8Closes#9Closes#10
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FLnYR4XtXQNamYJXGYT8oC
The multiplayer set-up, the lobby, the start of a game, and four signals a remote client had never
been sent. Reasoning, the preset table and what was verified how: CHANGELOG.md.
- Co-op, Competitive, Cutthroat, Solitaire and Custom, on both screens, from one shared block —
they had drifted, and each was missing a question the other asked.
- A player reads the whole rule set before taking a seat, may leave a lobby or a running game, and
keeps a seat across a reload. The host may clear a chair. The browser remembers every game it is
in, not just the last one.
- The start of a game is drawn: a handoff beat, an announcement, the code and type in the header.
- Sound, the timetable flash, announcements and the just-drawn badge now reach a remote client;
justDrawn goes to the seat that drew it and nobody else.
- Played on StartOS, which found the rest: an Extra belongs to the player who played it, the board
never named the Superintendent, bot seats were reported as absent players, and rule section
numbers are out of every string a player reads.
Also carries the previous session's Heavy Grade documentation work — asked again, answer unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016JczK5i33ZNSf2PtzZqdhS
Three more from the v0.4.9e gameplay-testing round, filed as Gitea issues, plus two bugs found
underneath them. Gitea#2 is diagnosed but NOT fixed: it needs a ruling, and the reasoning is in
TODO.md under Play Balance.
GITEA#4 — AN EXTRA STARTS WHERE THE PLAYER PUTS IT. Only one Division Point was ever offered,
chosen by number parity. The number no longer decides an Extra's direction — the start does, which
supersedes the recorded ruling that "the number decides, like everything else on the timetable".
The two cannot both hold: an odd, westbound Extra placed at the WEST end would leave the Division
on its first move having crossed nothing, and be paid for the run. Either end now runs the train
away from itself; at an Interchange or a Control Point the player picks the direction. The
Interchange start is a YARD, off the running line, which is what makes the Superintendent clause
work: placing it can never force a collision, a guaranteed one holds it there for another Stage,
and a potential one is the Superintendent's to rule on — exactly evaluateClearance's `blocked` and
`ask`, so nothing new decides collisions. Where an Extra may start is now a house rule
(divisionPointsOnly / ownOffice / anyOffice, defaulting to what the engine already did). The
legacy `atSeat` intent field still replays as it always meant.
FOUND UNDERNEATH IT: an Extra started away from a Division Point ran empty. isBeingMadeUp tested
position alone, so the Control Point start has been shipping since it was added with a train that
could never be given a consist. Found by playing it, not by the tests, which had only asserted
where the tray landed.
FOUND UNDERNEATH IT: collide left the wrecks on the card. Destroyed trains kept their Transit
entries, and evaluateClearance counts every transit as an occupant, so one rear-end collision
permanently poisoned that Mainline card for every later train.
THE MAINLINE CARDS WERE ROLLED, NOT DEALT — drawn from the nine types with replacement, so a
Division could hold two Interchanges and Plains carried the weight of a card printed once. "An
Extra may start at the Interchange if one is on the board" only reads as a rule if the board holds
at most one. Now dealt from the printed deck without replacement, verified over 1600 deals. This
re-deals every seed: the published replays were re-recorded, and the saved games in docs/ are
retired too — two of those were already dead before this release and nobody had noticed.
GITEA#6 — A TRAIN CARD IS NEVER DISCARDED, Timetabled and Extra alike. The forced play needed no
mechanism: nothing discardable plus a hand over the limit leaves exactly one legal way to end the
turn, and playing a train is unconditionally legal, so the corner cannot trap anyone. The bot
needed no rule either. 400/400 games finished, revenue unmoved, trains scheduled 1.2 -> 1.3. The
player is told on the card and on the button.
GITEA#7 — COACH COUNTS. 1/2 Crack Limited 3 -> 2, 5/6 The Sparrow 2 -> 3. A change to the cards,
so Trains3.pdf and the transcription keep the original numbers with a footnote while content.ts
and the Home Deck reference carry what the game plays.
CONTENT.TS COMMENT PASS — no data changed, only comments. Four were factually wrong, including an
office table naming counts doubled long ago and a pointer to a DEALT_DECK_SIZE that has never
existed. Every Enhancement row cited its implementation by line number and every citation had
rotted; they name functions now. Card counts came out of the comments, since they move with play
balance; source-sheet figures and dated measurements stayed.
TODO.md gains an item for a card reference generated from content.ts, in six sections, so the
documentation cannot disagree with the game.
715 tests pass, tsc clean, site builds.
Gameplay testing on 0.4.9d returned six reports. Five are fixed; the sixth could not be
reproduced and is written up in TODO.md with the two questions that would pin it down.
TWO TRAINS AT ONE PLATFORM ANSWERED TO ONE BUTTON. `porter.board` and `porter.detrain`
carried no tray, so there was one button per platform however many trains stood at it and
the reducer filled the first empty coach on the A/D tracks. `check` and the reducer were not
even asking the same question: `check` skipped a train whose card refuses passenger work and
the reducer did not. Both intents now carry an optional `trayId`, one function resolves the
train and the coach for check/execute/reduce alike, `legal.ts` offers one candidate per train,
and the label names it.
A LOAD COULD BE MADE AND BROKEN WITHOUT GOING ANYWHERE. A Freight House could unload the
boxcar it had just loaded; a platform could detrain the passengers it had just boarded. Full
Revenue at both ends for a movement that never happened. Jesse's rule: a load made anywhere in
an Office Area may not be broken anywhere in that Office Area, ever — it has to be carried to
another district. The load carries the seat that made it (`RollingStock.origin`), stripped by
`pooled` at every yard push. Measured at -0.60 +/- 0.10 Revenue a game (t = -6.1) over 400
paired deals: 78 worse, 3 better, 319 unchanged — free Revenue coming off the board, not a nerf.
THE GROCER'S WAREHOUSE SHIPPED AND THE REFINERY RECEIVED. Both were `flow: 'both'` on the
reading that "Freight House" was a collective term for exactly those two, and therefore what
§9.3 described. The engine has dealt a Freight House CARD since before v0.4.9, so §9.3 names
it and the argument goes. The card set agrees: all three Refinery modifiers grant +1 outbound.
Refinery outbound-only, Grocer's inbound-only, Freight House the one two-way industry — which
leaves exactly the one same-district pairing the rule above refuses.
NOT REPRODUCED: cars left behind when backing up over them. Five layouts tried, including cars
spotted at an industry; every one couples the lot. Three are pinned in `apply.test.ts`. One way
to create such cars was closed anyway — `flyingSwitch` wrote its cut past `carsOn`.
Both published replays that had gone dead were re-recorded; a rules change retires a save, and
`harness.test.ts` is what catches it.
The same change ships as v0.4.9e on the 0.4.9 line, branched from the v0.4.9d commit — the engine
files these fixes touch are identical across the two lines, so the patch applied cleanly both ways.
Also carries the two "Queued 2026-08-22, from playing on StartOS" TODO items that were staged
before this work started (Games in Progress readability, and getting back into a game after
losing a browser). They are notes, and items 9-12 below them are numbered against them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011nbvwWMef8CuEP6t5cgkTv
Both halves came out of playing the StartOS build. The wrapper's health
check and admin actions consume this; they land separately.
The host picks the table size (2-4) when creating a game, and the seats
array is built at that length once. Before, it GREW as people joined, so
the four rows on screen were partly fiction — a 2-player game just started
with a 2-long array, while a host who dropped a bot into a later chair
padded it with a null and silently disabled Start behind a one-line note.
A gap can no longer be written down rather than merely being refused.
That also avoided a trap. Compacting seats at Lobby.Start — the obvious
way to support a "closed" chair — would have shifted the player index that
every PlayerSession stamps at join time and that /api/stream and
/api/intent both route by, handing a player somebody else's railroad with
no error anywhere.
And it fixed a live balance bug: minCombinedRevenue is derived from the
player count, but the config was fixed at CREATE while the count wasn't
known until START, so the lobby guessed 4. Every 2-player game ran against
a floor of 60 instead of 30 — and missing the floor means everyone loses,
so a 2-player competitive game was set up to fail for a UI artifact rather
than a rule.
/api/health gained games:{active,lobby}, read from a new cheap summary()
on GameSession rather than exportSave(), which would copy every intent of
every game to answer a question about none of them. Three admin routes are
new behind an ADMIN_SECRET env var in an x-admin-secret header: GET
/api/games, GET /api/games/<id>/save, DELETE /api/games/<id>. Until now a
started game could not be ended by anyone — no route, no player action, no
resignation — so an abandoned game stayed active in the index and was
faithfully resumed on every boot, forever.
Three deliberate choices there: the admin secret is NOT the join secret,
which every player holds and which would therefore let anyone at the table
destroy anyone else's game; unset means the routes 404 exactly as any
unknown path does, with or without a header, so a server never given an
administrator doesn't advertise that it has one; and a delete returns the
deleted game's save, since the intents are the game (D5) — nothing is
destroyed without being handed to whoever destroyed it.
SavedGame gained an optional lastMoveAt (falling back to createdAt) so
"has this stalled?" survives a restart. Kept out of history for the same
reason the turn timings are: a replay must reproduce a game from decisions
alone, and wall-clock is not a decision.
index.ts logs "Resuming N saved games..." before the loop rather than one
line per game after it. Measured a full 4-player game at 100ms to replay,
and only unfinished games are replayed, so listening before loading would
have bought nothing for the cost of a "still loading" state everywhere.
Verified: 667 tests pass (662 + 5), and the new session tests were checked
against two mutations (lastMoveAt never advancing; resume dropping it) to
confirm they fail without the code. Live against a running server: health
counts tracking through the lobby->game transition, admin auth rejecting a
missing and a wrong secret, list/export/delete, the deleted game's files
and index entry actually gone from disk, a second delete 404ing, the admin
routes invisible when ADMIN_SECRET is unset, and a 3-player table refusing
a 4th player and a size of 5 refused at the door.
Also carries the TODO items raised on 2026-08-21: the lobby offering no
game parameters (the floor bug within it now fixed, the form still
missing), and the four optionalRules — of which only reducedVisibility and
emergencyToolbox are read by anything, while sisterTrains and
employeeRotation are declared, defaulted, and consulted nowhere.
The "Play multiplayer" door on index.html had sat disabled, labelled
"Coming soon", since before the server existed — Phases 2 through 4 built
a working lobby and nothing ever linked to it. Loading the site landed on
the same solitaire splash whether a real multiplayer server was behind it
or not, with no visible way in. Found packaging Phase 6 for StartOS.
The door is now a live link to ./play.html?lobby, and main.ts's start()
routes ?lobby straight to the lobby screen — the same showScreen('lobby');
runLobby(beginRemote) the in-game Multiplayer button already used —
instead of dealing a solitaire game first.
GET /api/health is new, and exists to be failed. The same dist/ ships both
served by src/server/ and uploaded as flat files by deploy-web.ts, and the
bundle is identical either way (D4), so the page cannot know from its own
build which it is; every other route 404s an unknown path exactly as a
static host does, so nothing distinguished them. The splash probes it on
load and closes the door when nothing names itself in reply.
The door starts open and only ever closes, deliberately: a wrong "no
server" is the bug above again — invisible, and it strands a player who
does have one — while a wrong "there is one" costs a click and a lobby
that says it cannot connect. The reply must name itself rather than merely
return 200, or a host answering every path with its index page would pass.
Verified: tsc clean; 659 tests pass (656 + 3); /api/health exercised live
against a running server — 200 with the right body, unauthenticated, while
an unknown path and a wrong method both still 404, which is what makes the
probe discriminate at all.
The probe's own test was vacuous on the first attempt — both its "closes"
cases reached close() through the .catch arm, so deleting the body-naming
check outright still passed. Caught by mutating splash.ts and re-running;
the test now covers all three closing routes and fails without the check.
A real server existed since v0.5.0 but nobody could reach it without a hand-built ?seat=&secret=
URL. This is what makes it a game you can actually create or join.
The server now hosts more than one game: src/server/lobby.ts (new) is pure logic — creating,
joining, bot seats, host transfer, starting — same split session.ts already draws for a running
game. persistence.ts gained one directory per gameId plus a top-level index so index.ts resumes
every saved game on boot. /api/stream and /api/intent now authenticate by session token instead of
?seat=&secret= — the token alone proves identity (lobby-and-sessions.md §1), so the join secret's
job ends at the lobby door.
Bots fill empty seats at Lobby.Start only, never take over a disconnected human (D8): session.ts
gained driveBots(), playing developerBot forward through consecutive bot seats after every accepted
intent. Disconnect keeps the seat and says so — Push gained an optional presence field, built
entirely by http.ts and never routed through the engine, since a disconnect is transport news, not
a GameEvent. Host rights pass to the earliest-joined remaining player if the host drops before
start.
Client: src/web/lobby.ts adds create/join forms and a live seating screen; localStorage replaces
?seat= for reconnecting straight back into a game already joined. A Multiplayer button sits beside
New game; the New Game dialog itself is untouched.
Found only by the live smoke test, not by typechecking: /api/intent read its token from the JSON
body while the client sends it in the query string (matching /api/stream) — every intent failed
"no such game" until caught by curl-level verification.
Doc fix: multiplayer.md's D18 said the player cap was 6; lobby-and-sessions.md §2 says 2-4 with the
reasoning and the test coverage to back it. The two had drifted apart. D18 now reads 2-4.
Not verified: an actual browser walking through the lobby screens — none available in this
environment, same limitation Phase 2's RemoteSession shipped under. 656 tests, 0 failures.
tools/jitsi-harness/ deliberately left untracked — unrelated side-project work, not part of this
release.
Phases 0-1 shipped in v0.4.0 (seat/identity split, per-player turn state, the Session boundary).
This lands Phase 2 (server core, one game, no lobby) and Phase 3 (persistence and resumption) per
docs/architecture/multiplayer.md §12. Phases 4-6 (lobby/reconnection, the 22 opponent-directed
cards, StartOS packaging) are still ahead.
Phase 2: src/server/session.ts hosts a game in pure logic (no sockets) on top of game.ts's existing
Game/submit/currentActor/actionMenu; it verifies seat === currentActor(game) itself before calling
submit, since submit() trusts its caller and a server can't. src/server/http.ts and index.ts add
POST /api/game, GET /api/stream (SSE, per-seat), POST /api/intent, and static serving of dist/.
src/sim/frame-delta.ts is a purpose-built per-seat board delta for one live push at a time. Found
and fixed along the way: actionMenu(game, seat) only used seat for the hand field, so a server
computing every connected seat's Menu would have handed the acting player's legal moves to a
waiting seat. Verified with a live end-to-end smoke test (2-player game, two SSE streams, a
rejected intent from the wrong seat, an idempotent resend) plus test/server/session.test.ts and
test/redaction.test.ts. Not verified: an actual browser (none available in this environment).
Phase 3: src/server/persistence.ts writes game.json and turn-timings.json, atomic-rewrite-then-
rename. game.ts gained fromMultiplayerSave, fixing a narration-attribution bug found while testing
it (fromSave's replay loop drops the actor argument, invisible in solitaire, unreadable the moment
there's more than one seat — fromSave itself still has this gap, deliberately untouched). Verified
live: server killed and restarted mid-game, both seats reconnected exactly where they left off.
Two rules bugs found while building this: the New Train phase never implemented its car-placement
round (every car of every train was placed by the Superintendent alone, in every mode, all along —
now reads the round position off tray.consist.length); and victory conditions are now one shared,
configurable GameConfig set across solitaire/competitive/coop instead of a fixed length lookup and
a dead firstToTarget condition.
Also folds in the three fixes already released on the patch line as v0.4.9b/c/d: a switching
train's crew badge failing to draw once it left the Office square, an unload that always took the
westmost car regardless of which was picked, and a legal decision that could render with zero
buttons.
docs/testing/0.5.0-test-plan.md and three reported-bug save files (docs/station-master-seed*.json)
included for reproducibility. tools/jitsi-harness/ deliberately left untracked — unrelated
side-project work, not part of this release. 635 tests, 0 failures.
A playtest review of seed 58228926 (day 6), plus one long-standing display complaint and the
first real audio beyond a placeholder.
- Coordinate labels read X,Y everywhere shown to a player, not the internal Y,X storage order.
Display-only.
- "No switching" now means may not add or drop cars, not "never touch it" — these trains can
still be moved onto Secondary Track to clear the mainline.
- Q3 corrected: Expedite governs WHERE a train may be left standing, not WHEN it leaves. The
forced same-Stage departure is gone; a new fault costs 1 Revenue if an expedited train is left
off the station when a Mainline Phase begins. Resolves "3/4 Express prints a rule it can never
use" as a side effect.
- evaluateClearance now checks every occupant on a Mainline card before offering a judgment
call, instead of returning on whichever it found first — found while explaining a playtest
report, fixed with a regression test.
- Three new synthesised sounds: arrive, depart, crash.
- The splash page shows the box art.
Full detail, measurements and reasoning in CHANGELOG.md.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FvU99NEakShRMg3nN3fHAZ
Builds docs/plans/switching-paths.md. A passing loop can offer two legal
routes between the same two squares, coupling different cars — the engine
only ever found one, an artifact of search order (Reported by Jesse, undo
379). exploreMoves now enumerates every simple route (per-path visited set,
capped at 4000 frontier nodes) and dedupes on outcome — destination, entry
side, and origin-tagged cars — rather than on reaching the square at all.
switch.move gains an optional `via: GridCoord` naming one intermediate
square on the chosen route; absent, it resolves exactly as before, so
every existing save and bot decision replays identically (575/575, then
579/579 with the new tests). Threaded through the label, the action-list
dedupe, the hover highlight (data-route), and the history (trayMoved.via).
Ruling recorded as Gap 14 in docs/rules/open-questions.md: the player may
choose the path; §A.4's "may not go around" a car does not reach a
different track the player declined to enter.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SAt2YCXgd5qCjBcF2x34aK
Planning only; no source files touched. The BFS visited set in exploreMoves
keys on (coord, entry), so two legal routes that rejoin through the same port
collapse to whichever is shorter — silently discarding the option that couples
cars on the way.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YQAJ4dND7enLj54eLyiF2C
Eight play reports and one design that had been written up and not built. The
through-line is switching: what a card can hold, which end of a train a cut comes
off, which way a train meets cars standing on the line, and what the board and the
log say about all of it.
TRACK ORDER FOR STANDING CARS, AND THE CUT ON YOUR OWN CARD
Two reports turned out to be one root cause. `TrackCard.standing` claimed "in track
order (§A.3)" and had no defined orientation at all, while `CrewTray.consist` does
(nose first, relative to facing) — so every transfer between them was a conversion
nothing performed. §A.3 says what it should be: cars occupy the track "in the same
order they originally held, left-to-right". Left-to-right is west-to-east, and that
is now the defined orientation of `standing` and of an industry track through
`carsOn`. It is the board's orientation, not the train's, so it does not change when
a different train touches the card.
- Setting out is batch-invariant. Four cars at once, four singles and two pairs
parked three different orders, one of them physically impossible. Successive
cuts off the same end stack up towards the engine, so the insertion point is the
train's own place in the row.
- Approaching a cut from either end now mirrors. `couples` is built nearest-first
along the direction of travel and reverses onto the nose, so the farthest car met
ends up nose-most — which is what makes a run-around worth its Move.
- A train no longer drives through its own cut. The walk began at the neighbour of
the start square and never read the start card, so a crew could set cars out and
pull straight away from them. Coupling is mandatory (§A.4) and your own square is
no exception; the cut counts against the four-car limit. Setting out off the end
you are not leaving by still works.
`CrewTray.standingWest` records where a train stands among the cars on its card — a
train may set out off both ends on one square, so which side a cut is on is not
recoverable from the array alone.
On the board, the cut is drawn split at the train — west cars left, east cars right,
engine in the gap — and each car's tooltip says whether it stands ahead of or behind
the engine. The history says which end a cut came off, and a move's button separates
"takes your own boxcar back off this card" from cars found standing on the line.
Decided: taking your own cut back on the square you are standing on is UNDOING the
drop. It is exempt from trains 3/4's per-location freight budget, X13's "drop but not
pick up" and X22's "empties only", and it refunds the budget the drop spent.
Otherwise a legal-looking drop becomes silently one-way.
Measured, 200 paired seeds, developer bot: -0.55 revenue (t = -3.63), freight revenue
1.11 -> 0.56. That cost is the bot's, not the rule's — its trains run engine-first,
so at a stub industry it sets a car out between itself and the only way out, and the
correct play is §A.5's facing-point move, which is the cross-turn planning TODO.md
already records as out of reach of any bot. Filtering self-recoupling moves out of its
options took recoupling from 625 of 1,029 set-outs to 101 of 677, and all 101 that
remain are that case. Read the number as a bot measurement, not a balance one.
THE SUPERINTENDENT'S RULING NAMES THE TRAINS IT IS ABOUT
Reported: the Superintendent could not tell which train he was clearing. The heading
asks the question now — "may Train 6 follow Train 4 onto the same Mainline card?" —
and the trains moved to the FRONT of each button, because the button splits its label
at the first em-dash and showed only the head.
AN INDUSTRY TRACK HOLDS FOUR CARS, LIKE EVERY OTHER CARD
Reported at undo 188: "we wanted to drop two cars, but were only allowed to drop one."
An industry track was built as long as its box count, so a one-box industry had room
for one car. Box count is how much WORK an industry can hold, not how much RAIL it
has. Ordinary track was the other exception, unbounded; both are gone and every card
holds four.
THE FREIGHT AGENT MAY STAGE A LOAD BEFORE THE CAR IS THERE
§6.3 asks nothing of the industry track — the empty car belongs to §9.3's Load the
car, which is the Laborer's action. The gate now lives only there, so cargo can wait
on the dock while the car to ship it in is still being switched in. Nothing can jam:
a load in a green box is waiting, not stuck.
THE TRUCK DOCK UNLOADS, AND BRINGS NOBODY
+1 inbound, no Laborer. It printed +1 outbound and +1 Laborer, which made it a
longer-host-list copy of Forklifts. Beside Packing Sheds it now does nothing at all,
and the hand tooltip says so before it is played.
Also in this release, from the days before: Mainline card tooltips computed from the
crossing rule; an Extra starts from the Division Point its number sends it to; a
modifier's suppressed grant comes back when a Whistle Post is upgraded; the Oil
Refinery and the Grocer's Warehouse ship as well as receive, per the card reference;
and the dormant defences name the attack they answer. `.claude/` is now gitignored —
it holds Claude Code's worktrees, i.e. a second checkout of this repository.
570 tests, typecheck clean. The three published replays were re-recorded twice —
legality changed, so bot play changed. Full detail in CHANGELOG.md.
Adds SVG board rendering shared by the game and the replay, a splash page with a
shareable replay directory, hover tooltips for reference detail, and makes curves
two-port arcs so sidings and run-arounds can finally be built.
