station-master/TODO.md

To do

Things worth coming back to. Anything here either gets done or gets an explicit decision not to — the point is that nothing quietly evaporates. A declined item stays, with the reason.

How this file works

Work sections come first, in priority order, and each holds every open item for one area of work. An item is a few lines: what it is, what it blocks, what it would cost. The measurements, rulings and rejected approaches behind it live in Reference at the back, keyed by the same number — so a section can be read at a glance and the argument is still there when it is needed.

The numbers are permanent ids, not positions. An item keeps the number it was raised under for life, wherever it later moves, because commit messages, Gitea comments and other items refer to it by number. Nothing is renumbered. New items take the next free number.

Restructured 2026-08-30, from eight subject sections plus a 43-entry chronological queue that had come to duplicate them. The queue was where the priority lived and the sections were where the reasoning lived, so every live item existed twice and drifted between the two — eight items were closed in one place and left open in the other, which reads as live work and is worse than no entry at all. One item per place now.

Every time — the process this project runs on

Not items. Things that are true of every change, and that have gone wrong when skipped.

• Ask Jesse what the version bump should be. Third digit is a bug fix, second is a new set of features, 1.0 is the first release worth the name — but which one a batch deserves is a judgment about how finished it feels, and it is his. The number lives only in package.json; scripts/build-web.ts stamps it into every page, so it is visible on the deployed site.
• Commits and tags are GPG-signed and I cannot make them. Stage the work, write the message to a file, hand over a ! command. Tags must be git tag -s — a bare git tag makes a lightweight tag and git push --follow-tags skips it without any error.
• Deploy from the right line. npm run deploy:web defaults to the same destination on both lines, so deploying from station-master/ silently replaces the playtesters' build with main's.
• A failing test written before the fix is the only thing that proves a fix. Three releases (v0.7.5 through v0.7.8) each reported the same bug fixed, and each fixed something real that was not the reported fault, because every verification read what the SERVER served rather than exercising the path with the state a returning player actually has. When a report repeats, reproduce the user's state first, and treat "I verified it" as unearned.
• Verifying over the HTTP API is not verifying the game. It renders no dialog. Extended play was recorded as checked on phoenix.local and was unusable in a browser (#35).
• Keep CHANGELOG.md current in the same change. Commit messages stay high level; the reasoning, the measurements and the things that turned out to be wrong live there.

30. Close a Gitea issue with Closes #<n> in the commit — not by hand (Jesse, 2026-08-30). Gitea records the linking commit itself when it auto-closes, so a manual close adds nothing but a step, and the same rule is what the workspace AGENTS.md already prescribes.

    What auto-closing does NOT carry is a RULING, and that still has to be written by hand. When the fix was not what the report implied, comment on the issue before it closes, saying what was decided and why. That is the half worth keeping: Gitea#15 was REVERSED on review — the placement is legal, and what was actually confirmed is that no train crosses the gap — and Gitea#14's comment lists the ten unbuilt cards so they do not vanish along with the issue. A bare Closes #<n> would have lost both.

    This item said the opposite until 2026-08-30 ("close each issue by hand… auto-closing leaves no record"), which was written from how the issues closed in v0.7.1 and v0.7.2 rather than from a decision. The token and the API calls, for reading issues and for leaving a ruling, are in the workspace's AGENTS.local.md.

31. The wrapper release sequence, unchanged since 0.7.2 and worth following exactly. Tag the app, fetch the tag into the wrapper's submodule, bump current.ts in place (the outgoing up has been empty every time, versions.md's common case — no new version file, no migration), rewrite the release notes in all five locales, update README.md and instructions.md, then npm run check / prettier / make x86 / make install. UPDATING.md in the wrapper is the authority and has not needed changing.

    Say which way games in progress go, in every locale, every time. 0.7.2 broke them (206 cards to 121, so a card id recorded under 0.7.1 refers to a different card or to none) and led with it; 0.7.3 and 0.7.4 carried them and led with that. It fails safe either way — src/server/index.ts refuses a save the rules reject, names the move it stopped at, and leaves the file untouched, so an operator can put the old version back to finish a game that matters.

Sections

1. Play it at a table — #39 #35 #42a #40
2. The common board, and watching play happen — Gitea#20 — #13 #15 #18 #75
3. Multiplayer, sessions and operations — #8 #7 #76 #77 #79
4. The screen — #44 #81 #33 #36
5. Replays and saved games — #14 #47 #48 #49 #50 #51 #52
6. Rules — #12 #80 #82 #83 #85
7. Play balance — #61 #62 #63 #64 #67 #68 #69 #70 #71 #72 #73 #66 #65 #74
8. The bot — #41 #57 #59 #53 #54 #58 #55 #56 #60
9. Code health and housekeeping — #46 #45 #84 #87
10. Documentation and assets — #15a #86 #88

Then, at the back: Reference (the measurements, rulings and rejected approaches behind the items above) and Done (everything closed, kept because several of them are the only record of a ruling or a lesson).

---

Play it at a table

The largest gap in the project, and none of it is a coding gap. Features are shipped, packed, running on phoenix.local — and the items below name the ones no person has met at a board. Everything else in this file waits behind a release; this waits behind an afternoon.

Test runs WERE made across 0.7.4 through 0.7.9 (Jesse, 2026-09-07) and produced no change requests — the two bugs that did come out of them are Gitea#21 and #22, fixed in v0.7.9.1. So this section is not "nobody has touched it since 0.7.4"; it is the narrower and still-true claim that the specific paths below have not been exercised at a table. More testing is planned at the end of the 0.7.9 series, before 0.8.0 starts — that is the moment to close these, not a separate errand.

• #39 — None of v0.7.4 has been played by a human. The Yard Office offer, the Red Flag hold and its out-of-phase prompt, and the loaded-Extra make-up rules are tested end to end, packed, and running on phoenix.local — and nobody has met any of them at a board. Two are interruptions that stop the Mainline Phase and put a question in front of somebody mid-thought, which is exactly the kind of thing only play reveals. See Reference · #39.

• #35 — Extended play has never been played at a real table. It was verified over the HTTP API, which renders no dialog — and when a human first reached it in a browser it was unusable (fixed in v0.7.9). The multiplayer vote has still never been driven through two browsers: what a second player sees while waiting, and whether "waiting on Carol" reads once Carol has closed her laptop, are unanswered. See Reference · #35.

• #42a — Nobody has clicked through the solitaire setup screen's own fields and confirmed the dealt game matches what was chosen. It took three attempts to become reachable at all — reachable is not the same as correct. See Reference · #42a.

• #40 — An older save may not replay, and players are not told so anywhere they will see it. Not a v0.7.4 fact and not a bug: a save is re-played through the current rules, so any narrowing of what is legal can stop one. The rule is written down now (README.md § Design notes); what is still owed is a line wherever a build is announced. See Reference · #40.

---

The common board, and watching play happen — Gitea#20

This is v0.8.0. One shared, seatless display of the public game, usable on a TV or in OBS on its own and publishable into the table's Jitsi meeting. The plan is docs/plans/jitsi-common-board.md, seven steps, of which 1-4 are the useful release and 5-7 are the Jitsi publisher. Items that look like screen polish live here because they need step 4's ordered presentation mechanism and nothing cheaper.

• #13 — I cannot see what the other players did — bots included. Settled 2026-08-29 as the harder reading: not log legibility but the ordered, per-action presentation of everyone else's turns. Jesse: "It's not fun to do my turn and have magic happen in the background." This is Gitea#20 step 4 pointed at a player's own screen. See Reference · #13.

• #15 — A "most recent action" line under the status block. The text already exists and is already correct — this is placement, not content. Decide with #13: in solitaire "most recent" is the right unit; in multiplayer what you missed is everything that happened while you were WAITING. See Reference · #15.

• #18 — Give every phase a visible beat. Not a timing problem — pump runs every automatic phase before the page renders once, so they are never drawn at all. A sleep fixes nothing; it needs the async stepped pump that Gitea#20 step 4 specifies, which is why it lives here rather than under The screen. See Reference · #18.

• #75 — Let the game join a call and talk to the table — the chat, audio and nudge half of the idea Gitea#20 took the visual half of. Long-term. See Reference · #75.

---

Multiplayer, sessions and operations

Running a game with other people in it — getting back into one, seeing what the server thinks is happening, and the turn structure that is still solitaire-shaped.

• #8 — A lost session token locks a player out of a running game permanently. The lobby half was fixed 2026-08-23. Needs Jesse's call on whether a token in a URL is acceptable. See Reference · #8.

• #7 — The StartOS "Games in Progress" action is one long unreadable run-on per game. ON HOLD, 2026-08-29 (Jesse) — StartOS 0.4.0.2 is expected to improve how action results are displayed. Re-open against it and re-read the output before designing anything. See Reference · #7.

• #76 — Multiplayer train make-up is a round, not one player's job. Unbuilt rather than wrong: the engine has no per-player turn within the New Train phase. See Reference · #76.

• #77 — Three things deliberately deferred while planning the server — bots covering for an absent player, and two others. See Reference · #77.

• #79 — D19's switching instrumentation still needs writing, once real people are playing. See Reference · #79.

---

The screen

What is drawn and where, for a player at the board. The three items settled on 2026-08-30 shipped in v0.7.9; what is left is the history panel and the end-of-game statistics.

• #44 — How much history the panel holds should be configurable. The cap is slice(-60) with no recorded reason anywhere. Where the setting lives is an open question and the item exists to ask it — a StartOS action, per game, or per browser. The replay viewer already answers the same question differently and in a different unit. See Reference · #44.

• #81 — The log's start marker only works while the whole log fits, and since #23 reversed the panel it marks the bottom rather than the top. Unblocked now that #23 is settled. See Reference · #81.

• #33 — The second pass on the results screen — badges, and the brainstorm Gitea#16 asks for. Needs Jesse and a conversation, not code, to start. The raw material is already kept, 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. See Reference · #33.

• #36 — There is no per-Stage "this train did not move" signal, so "longest an engine sat on a siding" cannot be answered. The comment on Gitea#16 said trainStoodStill would supply it; that is wrong, and was found by reading advance.ts. Settle with #33, its only consumer. See Reference · #36.

---

Replays and saved games

The save format, the two replay viewers, and how a game gets shared.

• #14 — Stamp the history with wall-clock time, so "how long did that turn take" can be answered afterwards. Half of it already exists server-side and is read by nothing, so the first job is to look at a real game's timings file rather than to build. The open question is granularity, and it is Jesse's: per intent or per turn span. See Reference · #14.

• #47 — How would a player publish a replay so other people can watch it? Today "Save replay" downloads a file and the only route to the site is sending it to Jesse. A whole 5-Day game compresses to ~1 KB, so it fits in a URL — which makes share-by-link the cheap first move whatever else is built. See Reference · #47.

• #48 — Decide what the standalone replay viewer is for. node src/sim/replay.ts writes a self-contained HTML file nothing links to; the site reads JSON saves instead. Keep it as a developer tool or fold its two extra panels into the JSON viewer and delete it. Blocks #49 and settles ten of #46's unused declarations. See Reference · #48.

• #49 — The standalone replay viewer shows no yards. The play page and the site viewer both do. A rendering job, not a modelling one — but do #48 first, since deleting the tool settles it. See Reference · #49.

• #50 — Undo is unlimited step-back, and that is a decision to revisit. You cannot undo your way to a better die roll, but you can see a train's departure Stage and then spend the turn differently. Deliberately left open until it has been played with. See Reference · #50.

• #51 — The 5 MB replay size limit is invented, not a browser constraint. It has earned its place — it caught a 5.2 MB payload — but the number wants a reason. See Reference · #51.

• #52 — Save/restore is not version-aware. An old save stops replaying rather than failing loudly, which is the safe direction but says nothing about what changed. This has bitten once, and #40 says it may bite again. See Reference · #52.

---

Rules

Where the implementation and the rules disagree, or where the rules do not say. Several of these need RAR or Jesse rather than code.

• #12 — Partly reproduced: cars left behind when backing up over them. Half was Gitea#17 and is fixed; the "I can later drive right through them" half is still unexplained and needs a board from whoever filed it. See Reference · #12.

• #80 — The 22 opponent-directed cards — 10 Action, 12 Space-use — are out of every deck until they are built. See Reference · #80.

• #82 — An unload does not check the facility's commodity. See Reference · #82.

• #83 — The deck is docs/Deck cards5.xlsx exactly, bar ten cards that are not built. Gitea#14 closed with those ten listed on the issue so they do not vanish with it. See Reference · #83.

• #85 — The 0.4.9 playtest line is behind on a rules ruling, and that was checked rather than assumed. See Reference · #85.

---

Play balance

The economies, the densities and the ceilings. Deliberately deferred as a body of work — the rules want settling first, and every figure quoted here from a full-length solitaire run predates v0.7.9's collision-floor change (#61).

• #61 — A solitaire game can now end on the collision floor (v0.7.9). Small — 1 game in 200 — but every full-length solitaire figure quoted in this file predates it. See Reference · #61.

• #62 — The crew tray count is due a re-examination, and the district-widening change is what triggers it. See Reference · #62.

• #63 — A district can now only widen as far as its Main reaches (v0.4.8) — worth watching in the numbers. See Reference · #63.

• #64 — The rebalance itself, deliberately deferred until the rules are right. Card counts, industry counts and the economies all want moving together rather than one at a time. See Reference · #64.

• #67 — The marginal Local Operations action is worth ~0, and that is the real ceiling. Three separate experiments agree. See Reference · #67.

• #68 — Freight is stuck at ~2.7 loads a game and three fixes have not moved it. See Reference · #68.

• #69 — Gitea#2 — passenger operations starve themselves of coaches. Jesse ruled the shortage stays ("it is possible to run out, that's part of the strategy"), so the three balance options here are declined rather than deferred; only the silence was a bug, and that is fixed. See Reference · #69.

• #70 — The rolling stock supply is a guess, marked provisional in content.ts. See Reference · #70.

• #71 — Office card density (Depot 4→8, Station 2→4, Terminal 1→2) was a blunt fix for an unwinnable-opening rate. See Reference · #71.

• #72 — Industry density (9 → 27, Gap 12) restored roughly the prototype ratio. See Reference · #72.

• #73 — Train density left alone by decision, but noted. See Reference · #73.

• #66 — Two balance items the victory-condition redesign did NOT supersede, and why they stand. See Reference · #66.

• #65 — Two balance items superseded by the 2026-08-20 victory-condition redesign, kept for the reasoning rather than deleted. See Reference · #65.

• #74 — A third item superseded by the victory-condition redesign, kept for the reasoning. See Reference · #74.

---

The bot

The developer bot exists to measure the game, not to be a good opponent — so a bot weakness matters when it stops a measurement being trustworthy. Read #57 before tuning any weights.

• #41 — The bot never plays Red Flags — zero in 200 games since Gitea#19, and that is deck luck rather than unwillingness. It takes the danger prompt unconditionally; what it never does is plant a flag ON PURPOSE to buy a Stage for switching, which needs it to know it wants time. See Reference · #41.

• #57 — The bot's priorities are not the problem — measured across ten heuristic variations. Read this before tuning weights; it is the argument that the ceiling is elsewhere. See Reference · #57.

• #59 — The run-around is out of reach of any bot, and the deck is why — measured five ways. See Reference · #59.

• #53 — The bot does not know to bring an expedited train back to the station. See Reference · #53.

• #54 — The bot cannot spot a car at a stub industry, and the cut-ordering rules made that visible. See Reference · #54.

• #58 — The bot cannot get a crew next to an industry, so Flying Switch never fires. See Reference · #58.

• #55 — Bot drift across this release — four measurements taken for the rebalance pass, kept so the next change has a baseline. See Reference · #55.

• #56 — The bot was partly living off an illegal placement, and barring curves from the Running Track took it away. See Reference · #56.

• #60 — Re-run the three "worth ~0" action-mix experiments against the new Revenue floor. The old figures were taken against a target, which the floor replaced. See Reference · #60.

---

Code health and housekeeping

Dead code, untrustworthy tests, and things carried but not used. Individually small; the reason they are one section is that each one found the next.

• #46 — 29 unused declarations across 14 files, and the build does not run the flag that finds them. The flag matters more than the 29 — two are Gitea#18 leftovers in one file, one found by hand and the other missed. Do #48 first; it settles ten of them. See Reference · #46.

• #45 — The Frame carries two things nothing reads — viewerSeat and overHandLimit. Deferred by Jesse 2026-08-30 ("leave it for now"); the decision when it comes is delete-or-document, not a patch. The two tally fields found by the same audit are fixed. See Reference · #45.

• #84 — Five test fixtures pinned a seed and meant "a game like this". All five broke on Gitea#14 for that reason. See Reference · #84.

• #87 — Regions as the primary model — the other half of §8.2. The Division map draws regions; the engine still does not think in them. See Reference · #87.

---

Documentation and assets

What the project says about itself, and what it ships alongside the code.

• #15a — Build documentation FROM the implementation, starting with a card reference. The prompt was finding train card data spread across five documents of three different vintages, one of them superseded. See Reference · #15a.

• #86 — Real audio, as committed assets. Everything the game plays is synthesised from oscillators today. See Reference · #86.

• #88 — card-reference.md's industry table may still be stale beyond Grocer's Warehouse and the Oil Refinery. See Reference · #88.

---

Reference — measurements, rulings and rejected approaches

The argument behind each open item, kept out of the work sections so those stay scannable. Moved here verbatim on 2026-08-30; nothing was rewritten or trimmed.

Play it at a table

#39 — NONE OF v0.7.4 HAS BEEN PLAYED BY A HUMAN.

NONE OF v0.7.4 HAS BEEN PLAYED BY A HUMAN. The Yard Office offer, the Red Flag hold and its out-of-phase prompt, and the loaded-Extra make-up rules are all tested end to end, packed, and running on phoenix.local — and no person has met any of them at a board. Two are interruptions that stop the Mainline Phase and put a question in front of somebody mid-thought, which is exactly the kind of thing only play reveals.

#35 — Extended play has never been played at a real table.

Extended play has never been played at a real table. Verified live on phoenix.local, 2026-08-29, against the installed v0.7.3:0 rather than in tests: a two-seat competitive game (one human client, one bot) was dealt over the HTTP API with days: 1, played to the end of its timetable, and reached awaitingExtension on Day 2 with official = { win, winner 0, daysElapsed } frozen at Day 1 and votes [null, null]. Voting yes as seat 0 was accepted, the bot followed as designed, and the game returned to active with extraDays: 1 and the official outcome unchanged. Both test games were deleted afterwards.

The save carry-over claim was checked rather than asserted: phoenix held five saves before the update, of which WHISTLE-4086 resumed and three were already refused by the 0.7.2 deck change. After updating to 0.7.3 the log is identical — same game resumed with the same 7 intents, same three refusals at the same move with the same code.

What is still untested is what the item is named for: humans, at a table. The multiplayer vote has never been driven through two browsers — what a second player sees while waiting on a first, and whether "waiting on Carol" is legible once Carol has closed her laptop, are still unanswered.

A HUMAN DID REACH IT ON 2026-08-30, AND IT WAS UNUSABLE — fixed in v0.7.9. Jesse played a solitaire game to the end and was never offered the extra Day. Nothing was wrong with the engine or the Frame: renderEnding wrote the two buttons into #actions and then opened #resultsdlg, which is modal, so the question sat underneath a dialog whose only control was Close. The dialog asks it now. The verification recorded above is exactly why this survived — it was driven over the HTTP API, which renders no dialog, so what was proven was that the SERVER supports extended play, not that a player can reach it. Read that distinction into every "verified on phoenix.local" line in this file.

#42a — Nobody has clicked through the solitaire setup screen's own fields and…

Nobody has clicked through the solitaire setup screen's own fields and confirmed the dealt game matches what was chosen. The screen took three attempts to become reachable at all (item 42 below); reachable is not the same as correct. Early item for the next play session, with #39.

#40 — A save from before v0.7.4 may not replay, and nobody has been told.

An older save may not replay, and players are not told so anywhere they will see it. 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 it there. A deck change is the likeliest breaker — a history naming a card the deck no longer deals has no legal answer at all — but any narrowing does it. This is the design working, not a fault: it fails safe every time, declining the load, naming the move and leaving the file untouched.

So this item is no longer "saves before v0.7.4" and should never be restated per version (Jesse, 2026-09-07). The specific 0.7.4 breakages — the Red Flags intent changing shape, a make-up that was legal being refused, a Yard Office arrival asking a question no older history answers — are recorded here as the ORIGIN of the rule rather than as the rule, and WHISTLE-4086 did survive on phoenix.local, which is why it is "may not" rather than "will not".

What is actually owed: the general rule is written down in README.md § Design notes, and a line belongs wherever a build is announced. Nothing in code.

Versioned, migratable replays are a post-1.0 question, deliberately deferred (Jesse, 2026-09-07): "once we get to a solid 1.0 release we will consider a system to version the replays so they are not as fragile — but not worth any effort right now." The reason it would be wasted effort now is that every migration would be written against rules that change again next release.

The common board, and watching play happen — Gitea#20

#13 — I CANNOT SEE WHAT THE OTHER PLAYERS DID — BOTS INCLUDED.

I CANNOT SEE WHAT THE OTHER PLAYERS DID — BOTS INCLUDED. Raised by Jesse 2026-08-22 from

playing a multiplayer game on StartOS: "on my display I need to see other players' moves, even if they are a bot."

What the code already does, checked rather than assumed: game.log is ONE shared log and linesSince(seat) (server/session.ts) sends every seat everything in it, so a bot's turn is not silently dropped — driveBots plays through submit(), which calls record(game, events, actor), and record prefixes any event carrying a player with "Player ". So the moves are arriving, attributed, in the history panel. Whatever is wrong is not that they were never sent, and that is worth knowing before anything is built.

What is genuinely missing is the BOARD. snapshot(s, …, viewer) builds cells from areaOf(s, viewer) alone, so a Frame contains the viewer's own Office Area and nobody else's. Another player can move a train the length of their district and the only trace on your screen is a line of text. The Division map is the one shared picture, and it shows trains on the Mainline, not switching inside a district.

ANSWERED 2026-08-29, and it is the harder reading. Jesse: "I want to be able to watch other players and bots make their moves. It's not fun to do my turn and have magic happen in the background and then have to figure out what others did." So the complaint is not that the history panel is hard to read — it is that the moves are not WATCHABLE. Marking the log is a consolation prize, not the fix.

This is Gitea#20 step 4, pointed at a player's screen instead of the common board. That issue — the public common-board display published into Jitsi — already specifies the mechanism, and docs/plans/jitsi-common-board.md §"Step 4 — Preserve individual human and bot actions" has the design: a display-step collector inside GameSession that captures a projected frame after EVERY successful submit(), human and bot alike, deltas it, and emits one step per accepted intent (not one per GameEvent — an intent drains automatic work behind it, and the event list is not a complete reducer).

The reason it is not simply free once #20 lands is that the plan deliberately stops short of here: "Keep player pushes unchanged: players still receive the final coalesced result after all immediately due bots finish." Extending the step stream to seated players raises questions the common board never has to answer — a spectator can be a second behind, a player waiting to act cannot; and a player animating three bot turns while their own move is due is a game that feels slower, which is the opposite of the complaint. Jesse, 2026-08-29: "this relates to issue #20 and will require a lot more thinking." Design it with #20; do not start it alone.

The constraint below still binds either way, and hardest here: the common board is seatless and shows only public state, whereas a step stream sent to a SEATED player is a Frame, and Frames are redacted per seat.

The constraint on the second, and it is the one that must not be got wrong: a district's BOARD is public — cards on the table, cars standing on them, trains — and a player's HAND, Revenue detail and drawn cards are not. test/redaction.test.ts exists precisely to catch a Frame that leaks the wrong half, and it works by serialising a seat's whole Frame and asserting no other seat's secrets appear anywhere in it. Any "show me their district" feature has to extend that test in the same commit, not after it.

A cheap first move that is right either way: mark the log where the viewer's own last turn ended, so "what happened while I was waiting" is a readable block rather than a scroll. That needs no new data on the Frame — sentLines already knows the boundary.

#15 — INVESTIGATE: a "most recent action" line under the status block.

INVESTIGATE: a "most recent action" line under the status block. Raised by Jesse

2026-08-22: a line below the status block ("Day, Stage, phase, waiting on") and above the Division map, carrying the same kind of text the history does — "Jesse drew from the Home Office deck", "played right-hand turnout at (−3, 0)". His own note: "I'm not sure that's what's going to make the most sense, but I think it's something that should be investigated."

The text already exists and is already correct. describeIntent and narrate produce exactly those sentences, and record() attributes them with the player's name. Nothing new has to be written to say what happened — this is placement, not content.

The slot is real but crowded. Between #turnchart and <main> in play.html there are already three transient banners: #phasenote (a phase CHANGED, auto-hides), #announce (a one-shot announcement — a train completing its run pays everyone), and #presence (someone is disconnected). A permanent fourth line has to not compete with them, and the palette is already spoken for: violet reports where you are, amber means clickable, green and red mean good and bad (turnchart.ts). A "what just happened" line is none of those.

Two placements, and they are different features. Put it in the page and it is the play screen's. Put it in turnChartHtml and it appears in both replay viewers too, which is probably a feature — a replay stepping frame by frame has exactly this question — but it makes the change three screens wide.

The question behind it is the unit, and it is why this should be decided with item 13. In solitaire "most recent action" is right: you took it, you are looking straight at it. In multiplayer the thing you actually missed is everything that happened while you were WAITING, which is many actions and possibly a whole bot turn — and one line showing only the last of them may be the least useful line on the page. Item 13's cheap first move (mark the log where your own last turn ended) answers that better. They may both be right, and one may make the other pointless; deciding them separately risks building both and needing neither.

#18 — INVESTIGATE: give every phase a visible beat — perhaps one second.

INVESTIGATE: give every phase a visible beat — perhaps one second. Raised by Jesse

2026-08-22, watching a game play: New Train, Mainline and the shift change "look like they are being skipped entirely". His suggestion: move to the phase, take a visible beat so the second row shows it changed, then move on.

They are not too fast. They are never drawn. pump() (advance.ts) loops advance() until something needs input, and drain() renders ONCE after the whole batch. So every automatic phase between one click and the next resolves without the page ever painting it. A minimum dwell time on its own therefore fixes nothing — the page has to step advance() one call at a time and render between, which makes this an async pump with a queue rather than a sleep.

#phasenote already exists for exactly this feeling — it announces that the phase CHANGED, because "the page can change out from under a player between one click and the next" — but with only the final phase ever drawn it can only ever announce the last transition of the batch. Stepping the pump is what would let it announce each one.

This is where it meets item 15. Jesse's own example: during the New Train beat the "most recent action" line would read "no new trains to build out" — which is a sentence nothing currently produces, because a phase that does nothing emits no event to narrate. Some of these beats would need a line written for them, and deciding which is part of the same investigation.

The obvious risk, worth stating before anyone builds it: a second per phase is four seconds of enforced waiting per Stage, forty-eight per Day, and a player who has seen it a hundred times will want it off. Whatever this becomes probably needs a speed control, or to scale with whether anything actually happened in the phase.

#75 — Let the game join a call and talk to the table.

Let the game join a call and talk to the table. Long-term. If the game could join a Zoom,

Teams or Jitsi call and post into its chat, it could carry the whole table's shared state without anyone alt-tabbing: the history of actions as they happen, and a prompt when someone is holding the game up — "Now waiting on player Alice to complete the Cargo phase."

• Further out, audio into the same call: a crash when a collision happens, a bell as the Stage clock turns over.
• Further out still, a nudge on a timer — if a player has not moved within some interval, the game says so, by beep or by spoken line: "Still waiting on Alice to complete the Cargo phase." That turns the turn chart's "waiting on" chip into something a distracted table actually notices.

Multiplayer, sessions and operations

#8 — A LOST SESSION TOKEN LOCKS A PLAYER OUT OF A RUNNING GAME PERMANENTLY.

A LOST SESSION TOKEN LOCKS A PLAYER OUT OF A RUNNING GAME PERMANENTLY. Raised by Jesse

2026-08-22: "if I opened a fresh browser window and wanted to resume HOPPER-4607, how would the server know which player I am and which game I'm trying to get to?"

PARTLY FIXED 2026-08-23, and the fixed half was the more common one. A browser that reloaded while SEATED IN A LOBBY used to orphan its chair outright — the token lived in a closure and was only written to localStorage at Lobby.Start, so the player could not return and nobody could free the seat, on a table that cannot start until every chair is taken. The record is written at create/join now, carries stage, and start() walks the two probes (/api/session, then the lobby stream) to land the browser wherever its seat actually is. A player may also LEAVE now (/api/lobby/leave), and the host may clear a chair, so a stranded seat is no longer permanent for the rest of the table either.

What is left is exactly the case Jesse asked about: a genuinely fresh browser, on a RUNNING game. Everything below still stands, and still needs his call on whether a token in a URL is acceptable.

A new tab or window of the SAME browser is fine — localStorage is per-origin and shared across a profile, so start() finds the token and rejoins automatically. A genuinely fresh browser is not: another browser, a private window, another device, or cleared site data. The token lives only in that one browser, and nothing else will accept an identity claim. /api/lobby/join resolves a code against lobbies, and a started game is removed from lobbies at Lobby.Start, so typing the game code answers no open lobby with that code — the same answer a typo gets.

The server knows exactly who you are and cannot be told. Each game's sessions.json holds { token, gameId, player, displayName } and survives restarts — read off the box: HOPPER-4607: player 0 = Jesse | token df9e04c7…. Everything needed is on disk; there is no door. lobby-and-sessions.md §1 says "presenting the token IS the rejoin", which was a fair assumption when a game lasted an afternoon and is a much worse one now that a game survives an update (v0.6.0) and can sit for weeks.

A second, nearer limit: REMOTE_KEY is a single localStorage key, so a browser remembers exactly one multiplayer game. Join a second and the first token is overwritten and gone, with the same lockout. D13 says one game at a time is expected but "deliberately not enforced" — the client enforces it by forgetting.

Three ways out, and the third is the one that fits what is already built:

1. Show the player their own rejoin link — a URL carrying the token in the fragment, to copy and keep. No new server state, and the fragment never reaches the server. It is still a credential in a link, so it lands in history and in whatever they paste it into.
2. Rejoin by game code + display name + join secret — do not do this. Every player holds the join secret, so any of them could claim another's seat by typing their name.
3. An administrator action, "Get Rejoin Link" — pick a game and a player, get a URL to send them. Gated by the admin secret, so only whoever runs the box can issue one, and no player can impersonate another. Fits the existing admin-action pattern exactly.

(1) and (3) together, most likely: the player keeps their own link, and the administrator can reissue one when they did not. Keying remembered sessions by gameId — with a picker when the browser holds more than one — fixes the single-key limit at the same time. Jesse has not yet decided whether a token in a URL is acceptable; the alternative is a bare token pasted into a field, which is uglier and stays out of history.

#7 — The StartOS "Games in Progress" action is one long unreadable run-on p…

The StartOS "Games in Progress" action is one long unreadable run-on per game.

ON HOLD, 2026-08-29 (Jesse): StartOS 0.4.0.2 should make action displays better. The diagnosis below is that the action-result view collapses newlines — which is exactly the sort of thing a platform release fixes. Re-read the real output on 0.4.0.2 before building anything; the nested-group rewrite may turn out to be unnecessary, and designing around a limitation that has just been lifted is worse than waiting.

Raised by Jesse 2026-08-22 after using it against four games. Lives in the WRAPPER repo (station-master-startos, startos/actions/gamesInProgress.ts), whose AGENTS.md says work belongs in issues on that repo rather than a TODO.md — recorded here because this is where the project's list actually is; move it if that policy is meant to bind.

What he asked for, taking the current output field by field: a separator between the players and the Day/Stage line; the phase in parentheses rather than after an em dash (Day 1, Stage 1 (Local Ops)); a separator before "Waiting on"; one after the waiting-on player and seat, before the start time; and one between the start time and the last-move time.

Why they are all missing at once, most likely. describe() joins its lines with \n, so the intent was one field per line. Every separator Jesse is missing is exactly where a newline is — which says the StartOS action-result view does not render newlines in a single's value, and collapses the lot into one line. Worth confirming in the UI before designing around it, since the whole diagnosis rests on it.

The structural fix, better than adding separators. ActionResultMember can itself be a group (osBindings/ActionResultMember.d.ts — "a new group of nested values, experienced by the user as an accordion dropdown"), so groups nest. Each game can be a collapsible group whose members are individual single rows — Players, Position, Waiting on, Started, Last move — instead of one string. That gives every field its own labelled row, makes the separator question disappear rather than answering it, and collapses cleanly when there are many games. Do this rather than punctuating the run-on.

Sorting, also asked for, and worth having once a box holds more than a handful: by game name, by start time, or by last-move time, ascending or descending. An action's input spec is built at open time, so a Value.select for the field and another for the direction costs almost nothing — and sorting by last move ascending is how you find the game nobody has touched, which is the main reason to open this action at all.

#76 — Multiplayer train make-up is a round, not one player's job.

Multiplayer train make-up is a round, not one player's job. When a new train is built,

players take turns adding cars to the consist; in solitaire one player does all of it. The engine currently has no per-player turn within the New Train phase, so this is unbuilt rather than wrong.

#77 — MULTIPLAYER — three things deliberately deferred while planning the se…

MULTIPLAYER — three things deliberately deferred while planning the server. Decisions and

reasoning are in docs/architecture/multiplayer.md §11; these are the ones left open.

• Bots should take minimally damaging, defensive actions when a player steps away, so a game is not permanently halted. Deliberately NOT automatic today: a turn timer forfeiting is different from a bot competing, and the clearance ruling is the one decision that changes another player's score. Bots fill empty seats at lobby time only (D8).
• Let a player resign and hand their railroad to a bot to finish. Same care needed as above, but it is consented rather than imposed.
• A forcing turn timer — explicitly NOT in the design. lobby-and-sessions.md §5 used to specify one: on expiry the server took "the safest legal action", including denying a clearance. Cut in the review, because it is the same objection as a bot playing for an absent player — the clearance decision changes somebody else's score, so anything that answers it automatically changes the game. Explore later if halted games turn out to be a real problem at a real table; the reasoning worth keeping is that deny is the safe default, since a held train costs a Stage and a wrecked one costs 5 Revenue and feeds the collision floor.
• The opening D12 for the Eastern Division Point (§4.4) decides nothing. Done in v0.4.1, and displayed in v0.5.4. It orders the whole chain, west to east by ascending roll; openingRolls is on the Frame now and the play page prints the chain under the Division map — West to East: Alice (1) → Bot 2 (5) → Bot 1 (11) — so the rolls that formed it are visible rather than only their result (lobby-and-sessions.md §4).
• Revisit the join secret (D14). One server-wide secret, passed out of band, gates create and join. Enough for a private box, probably not enough if stationmaster.<domain> is pointed at the open internet for long. Note that one-game-at-a-time per person is expected usage and deliberately NOT enforced — enforcing it needs cross-game state whose only job is deciding when to release someone, and getting that wrong locks a player out.

#79 — D19's switching-instrumentation still needs writing, once real people…

D19's switching-instrumentation still needs writing, once real people are playing. "13%

for the bot" (multiplayer.md D19) was a one-off measurement, not code — nothing in bot.ts or the sim tools logs it today. It needs live human wait-state data, so it can't usefully land before Phase 2 and realistically not before Phase 4 (real people at a lobby, not bots). A few lines when the time comes: log whether a legal local-only action existed for a waiting player, and whether they took it the moment their turn arrived.

The screen

#44 — The history panel's 60-line cap is hard-coded, and how much history it…

**The history panel's 60-line cap is hard-coded, and how much history it holds should be

configurable.** Raised by Jesse 2026-08-30, alongside the reversal (#23) that made the panel worth reading in the first place. The questions below are his, and are deliberately left open — this item exists to ASK them, not to answer them.

Where the number lives, and what justifies it. main.ts renders session.lines().slice(-60). There is no comment saying why 60, and nothing in CHANGELOG.md records a reason — so the figure is unexplained rather than chosen, the same shape as the 5 MB replay limit under Replay / Save Games.

THERE ARE ALREADY TWO ANSWERS ON TWO SCREENS, IN TWO DIFFERENT UNITS, which is worth knowing before a third is added. The play page keeps the last 60 lines. The site replay viewer keeps the last 19 steps — for (let k = Math.max(0, at - 18); k <= at; k++) (replays.ts) — a window of moves rather than of text, which yields wildly different amounts depending on how much each step narrated. Whatever is decided has to say whether those two are the same setting or deliberately different ones.

The questions, in Jesse's order. None of these is settled.

1. At the StartOS wrapper level, via an action? Note what that would and would not reach: a wrapper action configures the SERVER, and solitaire runs entirely in the browser with no server involved at all. So this can only ever be the multiplayer half of an answer, never the whole of one.
2. Per game? It would ride in GameConfig — into the URL, into every save, and identical for everyone at the table. That makes a DISPLAY preference part of the ruleset, which cuts against the principle already recorded under Replay / Save Games: a save must reproduce a game from decisions alone, and two recordings of the same game should not differ because someone liked a longer scrollback.
3. Somewhere else? The obvious candidate is Settings in localStorage, beside districtMode, soundOn, zoom and gameCardOpen — per browser and per person, which is what every other display preference on this page already is. Cheapest, and the only one of the three that works identically in solitaire and multiplayer.

What is not free whichever is chosen. The panel rebuilds its whole innerHTML on every render, so the cap is doing performance work whether or not that is why it was picked — raising it means deciding what the panel does with a thousand lines. And the item below is coupled: "— the game began —" is drawn only while the whole log fits, so this cap decides when that marker can appear at all.

#81 — The log's start marker only works while the whole log fits.

The log's start marker only works while the whole log fits. Added 2026-08-23: a multiplayer

game marks the top of the history with "— the game began —", which is honest only while the panel is showing every line there is. The panel caps at slice(-60), so past sixty lines the marker is suppressed rather than lying about where the top is — and "what happened while I was waiting" (item 13) still has no marker at all. Both want the same mechanism, and item 23's newest-at-the-top question decides what that mechanism draws.

#33 — The second pass on the results screen — badges, and the brainstorm Git…

The second pass on the results screen — badges, and the brainstorm Gitea#16 asks for. Needs Jesse and a conversation, not code, to start. 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. The raw material is already 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.

#36 — There is no per-Stage "this train did not move" signal, so "longest an…

There is no per-Stage "this train did not move" signal, so "longest an engine sat on a siding" cannot be answered. The comment on Gitea#16 said trainStoodStill would supply it. That is wrong, and was found by reading advance.ts: the event fires only 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, so a streak folded from it reads "1 Stage" for ever. What it would take: a new event per Stage per stationary tray (cheap to emit, 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. Settle with #33, its only consumer.

Replays and saved games

#14 — INVESTIGATE: stamp the history with wall-clock time.

INVESTIGATE: stamp the history with wall-clock time. Raised by Jesse 2026-08-22: "store a

date/time stamp with the history information. Even if it's not displayed immediately, someone could tell afterwards, or later if we decide to display it — how long things took between different turns and actions."

Half of this is already built and nothing reads it. server/session.ts closes a TurnTiming { player, phase, day, stage, startedAt, endedAt } every time the acting player, phase, Day or Stage changes, and persistence.appendTiming writes each one to the game's timings file. So every multiplayer game on the box already has per-turn wall-clock on disk. Nothing displays it, and nothing has ever read it back. First job is to look at that file from a real game — the answer to "how long do turns take" may be sitting there already.

Solitaire has none of it. game.log is { text, tone }[] and a save is { seed, config, history }. There is no clock anywhere on that path.

The constraint, and it is deliberate rather than an oversight (stated twice in server/session.ts): wall-clock is kept OUT of history because a replay must reproduce a game from decisions alone. Do not put a timestamp on an Intent. It would ride into every save, change the save format, and make two recordings of the same game unequal for no gain — the engine has no clock and must stay deterministic.

And a timestamp on a log LINE does not survive. game.log is rebuilt by fromSave on undo, on restore and in the replay viewer, so a time recorded on a line is gone the first time the player takes a move back. Whatever is built has to be a SIDECAR — times keyed by intent index, written only on the live path — and every consumer has to handle its absence, because a replayed or imported game legitimately has no clock at all. That absence is the honest answer, not a hole to fill with createdAt (the same call lastMoveAt already makes, and for the same reason).

The open question is granularity, and it is Jesse's to settle: per INTENT (every draw, every Move — finest, biggest sidecar, and the only thing that answers "how long between actions") or per TURN SPAN (what already exists, one row per player-phase, and enough for "how long between turns"). Per span is free today; per intent is new work on both the solitaire and the multiplayer paths.

What it would feed if built: the "Games in Progress" admin view (item 7 above, which already shows lastMoveAt), a post-game "that Day took 40 minutes" summary, and any future pacing question about whether a 12-Stage Day is too long at a real table — which is the sort of thing only a table can tell us and only a clock can record.

#47 — INVESTIGATE: how would a player publish a replay so other people can w…

INVESTIGATE: how would a player publish a replay so other people can watch it? Today

"Save replay" downloads a JSON file to the player's own machine, and the only way it reaches the site is by sending it to Jesse to drop into public/replays/ and redeploy. The question is what a self-service version would look like.

The constraint. The site is fully static — dist/ is uploaded to File Browser and Start9 Pages serves the folder — and the replay list is a build-time manifest.json because static hosting cannot list a directory. So publishing needs something that accepts a write.

The one measurement that matters: a full 5-Day game is 451–1017 bytes compressed (brotli), about 600–1150 characters base64. A save is the seed plus the intents and the engine recomputes the board, so a whole game fits in a URL.

Four shapes, roughly costed:

1. Share by link, no server (~1–2 hours). Put the compressed save in the URL fragment (replays.html#s=…); "Share replay" copies a link and anyone opening it watches the game. The viewer already parses saves and already has a file-open path, so this is compression, a hash reader and a copy button. The fragment never reaches the host. It is a link rather than a gallery: nobody discovers a game they were not sent.
2. A write endpoint (a day or two, and it is a service). Accepts a POST, validates the save by replaying it through the engine — save-replay.ts already does exactly that check — writes the file and regenerates the manifest. The work is the surround: auth or rate limiting, abuse handling for a public write, CORS, and a deploy story. It also ends "static hosting is all this needs", which has been load-bearing.
3. Browser writes to File Browser directly — rejected. It needs FB credentials in a static page, so anyone viewing source gets write access to the whole File Browser, and the manifest would need a read-modify-write from the browser that loses a save when two people publish at once.
4. Curated, manual (zero code). What happens today, and it composes with (1): players send links, Jesse publishes the good ones.

The question behind the question is whether a gallery of strangers' games is wanted on a personal StartOS box at all. If it is, (1) is the piece (2) would need anyway, so it is the right thing to build first either way.

#48 — Review the standalone replay against the site's replay viewer.

Review the standalone replay against the site's replay viewer. `node src/sim/replay.ts

--seed 1234 --out replay.htmlwrites a self-contained HTML file; the site instead reads JSON saves frompublic/replays/`. Nothing links to the standalone one and its output is gitignored, so it is a developer tool that happens to look like a product feature. It carries two panels the site viewer does not — the bot's decision trace ("what it chose, why, and what it passed over") and the timetable — which is debugging material rather than something a player wants. Decide: fold the decision trace into the JSON viewer and delete the standalone, or keep it and accept that it is a tool. No action for now.

#49 — The yards are shown on the play page and in the site replay viewer, bu…

**The yards are shown on the play page and in the site replay viewer, but not in the

standalone one.** Corrected 2026-08-30 — this said "not in EITHER replay viewer" and that has not been true for some time: replays.ts renders vdivyard, vclsyard, vdivtot and vclstot from f.yards, and replays.html has the markup. Only sim/replay.ts, the standalone dev tool, lacks them. The Frame carries them, so it remains a rendering job rather than a modelling one.

Do the decision below first. The standalone viewer is the subject of an open keep-or-delete question three items down; adding panels to something that may be deleted is the wrong order.

#50 — Undo is unlimited step-back, and that is a decision to revisit.

Undo is unlimited step-back, and that is a decision to revisit. The save is the seed plus

the intents, so undo() replays without the last one and can walk all the way to the deal. The RNG advances with the replay, so the same play re-rolls the same 1D12 — you cannot undo your way to a better die. But you CAN see a train's departure Stage and then spend the turn differently, which is an ordinary solitaire take-back and also a real information leak. Options if it starts to feel like cheating: make the Stage boundary a commit point, or cap the depth at the current Stage. Deliberately left open until it has been played with. Multiplayer gets nothing until there is a proposal/agreement flow — undo there is a table decision, not a button.

#51 — The 5 MB replay size limit is arbitrary.

The 5 MB replay size limit is arbitrary. Invented, not a browser constraint. It has earned

its place — it caught a 5.2 MB payload that turned out to be the whole grid re-serialised every frame — but the number itself deserves a reason.

#52 — Save/restore is not version-aware.

Save/restore is not version-aware. A save from an older ruleset stops replaying rather than

failing loudly, which is the safe direction but says little about what changed. This has now bitten once: both published replays were dead — one got 42 intents into 360, the other 4 of 338 — and nothing said so; they simply ended early and looked like short games. A save should carry a ruleset stamp and the page should say "this replay was recorded under an older ruleset and stops at Stage N" rather than presenting a truncated game as a whole one.

THE FULL FIX IS DEFERRED PAST 1.0, ON JESSE'S CALL 2026-09-07: "once we get to a solid 1.0 release we will consider a system to version the replays so they are not as fragile — but not worth any effort right now." The reason is sound and worth keeping: every migration would be written against rules that change again the next release, so the work would be redone rather than reused.

What that ruling does NOT defer is the honesty. A truncated replay presented as a whole game is a wrong answer, not a missing feature, and saying "this stops at Stage N" needs no version stamp at all — fromSave already knows how many intents it replayed of how many it was given (session.ts:454 reports exactly that for multiplayer). That half is small and stands alone. The general rule now lives in README.md § Design notes; #40 is the "tell the players" half.

Rules

#12 — PARTLY REPRODUCED: "when I back up to collect standing cars and, furth…

**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 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 between. carsOn (state.ts) is the SINGLE answer to "what is standing here", and the movement walk, the sweep in carsCoupled and every renderer all ask it — so cars a train can drive through would have to be cars that are on screen and not in carsOn, and there is no such place. (One route to one was closed anyway: flyingSwitch's reducer wrote the cut straight into industryTrack, which for a Passenger Facility is not where carsOn looks. check refuses a non-freight target, so it never fired.)

The two questions that would settle it, for whoever has the board: was there a SECOND route to the caboose — a parallel row, or a turnout — so the move could have gone round the cars? And what did the move button say it would couple? The label names every car (describeIntent), so a button that read "couples caboose" and one that read "couples 2 boxcars, caboose" are different bugs: the first is route selection, the second is the sweep.

#80 — THE 22 OPPONENT-DIRECTED CARDS — 10 Action, 12 Space-use — ARE OUT OF…

**THE 22 OPPONENT-DIRECTED CARDS — 10 Action, 12 Space-use — ARE OUT OF EVERY DECK UNTIL THEY

ARE BUILT.** Jesse's call. They were already cut from solitaire (Q6, no legal target with one player); they are now cut from the competitive deck too, because checkPlay answers both categories NOT_IMPLEMENTED and dealing them would make ~9% of draws reject outright. Flip opponentCardsInDeck in setup.ts when they land. They are played AT another player — Watertower, Derail, Railroad Crossing and so on — so they are genuinely multiplayer work, and three Enhancements are waiting on them: Facing Point Locks, Water Column and Overpass are wired and read, and fire only against these cards. Until then those three are dormant by design rather than broken.

#82 — An unload does not check the facility's commodity.

An unload does not check the facility's commodity. laborer.beginUnload gates on

allows.inbound, a loaded car, an empty of that type in the Division Yard and room in the red box — but never on facilityCarTypes(f), which freightAgent.stockOutbound does check. So a Freight House (boxcars) will unload a hopper. Found reading the code for the v0.4.9e district rule, not from play. Low impact today because the district rule now refuses the only same-Office pairing that made it easy to hit, and because the bot spots matching cars — but it is a rule the engine states in one direction and not the other.

#83 — THE DECK IS docs/Deck cards5.xlsx EXACTLY, BAR TEN CARDS THAT ARE NOT…

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.

#85 — THE 0.4.9 PLAYTEST LINE IS BEHIND ON A RULES RULING, and that was chec…

**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 undo() config fix is inert there because its GameConfig carries no victory dials, so 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 — 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 that build exists to collect. Recorded so the divergence is a decision rather than a surprise — and so the 0.4.9 README is not "corrected" to match main's wording, which would then describe behaviour that build does not have.

Play balance

#61 — A SOLITAIRE GAME CAN NOW END ON THE COLLISION FLOOR (v0.7.9, 2026-08-3…

**A SOLITAIRE GAME CAN NOW END ON THE COLLISION FLOOR (v0.7.9, 2026-08-30) — small, but every

full-length figure in this file predates it.** §3.4's check was gated on mode === 'competitive' || mode === 'coop', so the two collision limits were live settings in the solitaire dialog that did nothing; Jesse's ruling was that they should do what they say, and the gate is gone. Measured over 200 standard developer-bot games: loss/collisionFloor fires in 1 game in 200, Days played 5.00 → 4.98 mean with a minimum of 1, and collisions per game unchanged at 0.14 (max 3).

Nothing here needs acting on — the effect is smaller than the noise on every number in this section — but it is a new way for a run to be short, so a mean taken over games that all ran five Days is no longer quite what is being sampled. Worth remembering when the rebalance pass re-measures, and worth watching if a future change makes collisions more common, because the cost of one stops being "−1 Revenue" and starts being "the game". 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.

#62 — THE CREW TRAY COUNT IS DUE A RE-EXAMINATION, and this is the change th…

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.

#63 — A DISTRICT CAN NOW ONLY WIDEN AS FAR AS ITS MAIN REACHES (v0.4.8) — wo…

**A DISTRICT CAN NOW ONLY WIDEN AS FAR AS ITS MAIN REACHES (v0.4.8) — worth watching in the

rebalance rather than acting on now.** Track stays inside the Limits at every row, so extending the Running Track is the only way to buy room for sidings, and a straight laid on the sign is worth more than it was. The bot barely notices — it built outside its own Limits 5 times in 100 games — but the bot also builds close to its Office; a human building deliberately hits this on the first wide district, which is how it was reported. If territory turns out to be the real constraint on freight, this is one of the two places to look (the other is the track supply, in Bot Performance).

#64 — REBALANCE, once the rules are right — deliberately deferred.

REBALANCE, once the rules are right — deliberately deferred. Card counts, industry counts

and the track mix all need a pass together, and none of them should move until the rules stop moving. Standing distortions to account for when it happens: offices are doubled (Q12) and industries tripled (Gap 12), both tuned when the deck held 139 cards and no track; it now holds 235 of which 96 are track, so every draw is diluted by 41% — precisely the pressure those multipliers exist to relieve. The 8 sharp curves have already been taken out on that argument; offices and industries are the two left. Until then, read no balance conclusion from the revenue numbers; they are a functionality signal only.

#67 — The marginal Local Operations action is worth 0, and that is the real…

The marginal Local Operations action is worth ~0, and that is the real ceiling. Three

separate attempts to spend the 60 actions better — capping the draw, pairing the two halves of a load, restricting Enhancements — each measured within noise of zero over 400 paired seeds. 76% of the time an outbound industry has neither a stocked green box nor a spotted car, and only 5% of Stages have a single workable facility anywhere, yet redirecting actions at that does nothing. Something upstream limits how much work exists to do at all; find out what before spending more effort on the option mix.

#68 — Freight was stuck at 2.7 loads a game and three fixes have not moved it.

Freight was stuck at ~2.7 loads a game and three fixes have not moved it. Sidings,

facility placement, car selection and the discarded-load leak all raised revenue (3.2 → 6.5) without raising loadStarted past 2.7. The chain is not leaking and the cars are arriving correctly (57% of drops land on a facility that wants them, 0% on one that does not). The binding constraint is now upstream of routing: 60 Local Operations actions a game, and a load needs a stocked green box AND a spotted car AND a free Laborer to line up in the same Stage. Measure how many Stages have all three before changing any heuristic — the answer may be that the economy, not the bot, is what caps freight.

#69 — Gitea#2 — passenger operations starve themselves of coaches, and the g…

Gitea#2 — passenger operations starve themselves of coaches, and the game says nothing.

Reported from v0.4.9e play: the Sparrow pulls into the Terminal with two loaded coaches, two passengers wait on the platform, four porters are unused, and only ONE of the four intended actions can be taken. Reproduced from the save (docs/station-master-seed947338225-day5(1).json, Day 5 Stage 11): the Division Yard holds 1 empty coach and 0 loaded, while the Classification Yard holds 4 loaded and 4 empty that cannot come back.

The engine is not deviating from the rules. Checked step by step: §9.2 discards the white coach into the Classification Yard on boarding, draws the white coach from the Division Yard on de-training, and §2.2 returns the Classification Yard only when the Division Yard is empty. All three are implemented exactly. The problem is the interaction — a single global refill condition over a pile holding six commodities with very different demand, where coaches (16 of ~60 cars) are consumed by both halves of every passenger cycle. Traced over the reported game the coach pool goes 8+/8− to 0+/1− by Day 5.

RULED — the shortage stays, and none of the three is being built. Jesse: "it is possible to run out, that's part of the strategy." For the record, the options were (a) refill when the Division Yard is dry of the type-and-state being asked for rather than dry of everything; (b) the same trigger but return only the cars of that type; (c) leave the rules alone and raise the coach count in ROLLING_STOCK_SUPPLY. All three are declined. What shipped instead is the EXPLANATION — the impediments panel now says the Division Yard has no white coach, how many are stranded in Classification, and that Classification returns only when the Division Yard is bare. Running dry is a position to play out of, not a broken game, once the screen says so.

#70 — The rolling stock supply is a guess.

The rolling stock supply is a guess. ROLLING_STOCK_SUPPLY (coach 8+8, boxcar 10+10,

hopper 8+8, reefer 5+5, tank 6+6, caboose 6) is marked provisional in content.ts and was scaled alongside the Gap 12 industry increase. Now that the Classification Yard returns stock only when the Division Yard empties, these numbers set the real supply pressure. Adjust from playtesting rather than theory, and watch whether industry density feels light or heavy at the same time.

#71 — Office card density (Depot 4→8, Station 2→4, Terminal 1→2).

Office card density (Depot 4→8, Station 2→4, Terminal 1→2). Chosen to remove a 25% chance

of an unwinnable opening deal. Blunt: it lifts the whole ladder and dilutes every other category. The better answer may be fewer Terminals, a cheaper first upgrade, or more A/D capacity at the Whistle Post itself.

#72 — Industry density (9 → 27, Gap 12).

Industry density (9 → 27, Gap 12). Restored roughly the prototype ratio. The "freight is

only 13–18% of gross" figure that motivated this was partly a measurement bug (see the stats.ts item in Done) and partly the car-selection bug; freight now runs at 37%. Worth re-deciding whether 27 is still the right number now that the industries are actually served.

#73 — Train density.

Train density. Left alone by decision, but noted: 22 train cards in 140 are drawn less often

than 22 in 115 were, and trains scheduled fell 2.9 → 2.1 as a side effect of the other density changes.

#66 — Not superseded by the 2026-08-20 victory-condition redesign (Multiplay…

**Not superseded by the 2026-08-20 victory-condition redesign (Multiplayer) — these two stay

houseRules dials, separate from the new GameConfig victory dials.** Noted only so the two redesigns aren't conflated. REVIEW THE TWO NEW RULES ONCE THEY HAVE BEEN PLAYED — both went in provisional, and both are now selectable rather than fixed. Jesse's call, both implemented and measured, both flagged in rules-v0.2.md. What follows is what was measured when each was the only option.

The opening deal (3 track + 3 other, from two separately shuffled piles). It did what it was aimed at, modestly: run-arounds 4/60 → 7/60 and districts 17.9 → 20.3 cards, with revenue unmoved on its own (−0.1, inside noise). Still nowhere near the 91/100 of the private-supply era, so the supply question is softened rather than answered. Two things to watch at the table: whether opening with six against a limit of three is a real decision or just bookkeeping, and whether three is the right number of each.

One Revenue for every train that clears your section. Now: one Revenue to EVERY player when a train completes its run. Jesse's revision in v0.4.2. The first version paid the Office a train departed, which on a five-Office railroad paid five separate times for one train and paid most to whoever it passed first. It pays once now, when the train runs off the end of the Division, and it pays the whole table — getting a train the length of the railroad is the shared achievement, and every Office it crossed had to clear it. Solitaire is nearly unmoved (7.0 → 7.3 mean over 200 games) because one player's departures and completions run at almost the same rate; in a multi-player game the shape is completely different and needs measuring once multiplayer exists — N players × 1 per completed run against the old N payments per train. The victory-target question stays live: 20 over 5 Days is still reachable largely on traffic, which is either the intent or an argument for raising it — and at the new default of 0 per transit it is not reachable on traffic at all, which is the first thing a playtest should check.

#65 — Superseded 2026-08-20 by the victory-condition redesign (Multiplayer)…

**Superseded 2026-08-20 by the victory-condition redesign (Multiplayer) — kept for the

measurements.** minCombinedRevenue replaces the fixed target these numbers were read against; re-measure once that lands rather than off this. RE-MEASURE THE BOT AT THE NEW DEFAULTS. Both provisional rules below are now settings on the New Game dialog rather than fixed choices, and the defaults are not what the numbers in this file were measured under: the opening hand defaults to three random cards (the prototype rule) rather than 3+3, and train revenue per transit defaults to 0 rather than 1. That second one is the big move — it was worth ~5.4 of a 7.0 mean, so the bot's revenue should fall to roughly the working freight-and-passenger economy alone, which is the number this game has actually been trying to read all along. Every mean, floor and threshold quoted below and in the tests predates it. The three revenue rates run 0–5, so the useful next step is a sweep rather than a single re-run.

#74 — Superseded 2026-08-20 by the victory-condition redesign (Multiplayer)…

**Superseded 2026-08-20 by the victory-condition redesign (Multiplayer) — kept for the

measurements and the reasoning.** LENGTH_PROFILES.target (20 over 5 Days, standard) is retiring in favour of minCombinedRevenue, defaulting to 3 × players × days (15 for 1-player/5-day, not 20) — a different number, deliberately not tuned to match this table. Whether the Office-ladder bottleneck below still applies at the new default is worth re-measuring once the redesign lands, but the fixed "20" this data argues against no longer exists as a target. The victory target (20 over 5 Days) is out of reach by a factor of about four, and the Office ladder is why. Measured over 800 games with the tuned bot, which no longer throws revenue away on collisions (0.0 a game, down from 0.4):

trains scheduled	games	revenue		Office reached	games	trains	revenue
0	110	0.67		Whistle Post	297	0.81	0.62
1	379	1.69		Depot	272	1.49	2.92
2	234	3.72		Station	176	1.89	4.36
3	68	5.68		Terminal	55	1.93	5.04
4	9	5.78

Revenue is almost exactly linear in trains scheduled — about 1.9 a train — and trains are capped by A/D capacity, which is the Office tier, which is a card you have to draw. So the whole economy hangs off one valve: 37% of games never leave the Whistle Post and earn 0.62; 53% of all games earn nothing at all.

Extrapolating the line, 20 Revenue needs roughly 11 trains and therefore 11 A/D tracks. A Terminal has four. The target is not merely missed, it is structurally unreachable under this deck at this Office ladder — no amount of bot skill closes it, and the best game seen in 800 was 26 against a median of 0.

The three ways out are all yours to choose between, and they are different games:

1. Lower the target to what a 5-Day game can produce (6–8 looks like the honest number).
2. Open the valve — more Office cards, or a cheaper first upgrade, or more A/D capacity at the Whistle Post, so the ladder is climbed rather than drawn.
3. Raise revenue per arrival. It is 0.46 today; each arrival can in principle pay 2 for passengers alone. That is the freight/passenger conversion problem, not the traffic problem.

Nothing here is a bot weakness any more, which is what this measurement was waiting on.

The bot

#41 — THE BOT NEVER PLAYS RED FLAGS — and since Gitea#19 that is deck luck,…

THE BOT NEVER PLAYS RED FLAGS — and since Gitea#19 that is deck luck, not unwillingness.

Re-measured 2026-08-29, after the card was redesigned: redFlagsSet fires ZERO times in 200 solitaire games. The old measurement (600 games: OFFERED 4,212 times, PLAYED 4) described a bot that declined a card it was constantly handed. That bot is gone.

Gitea#19 replaced the rule outright: a flag is planted on one side of your own Limits and holds the next train from that direction, and it can be played out of phase when the engine breaks in with "COLLISION RISK! FLAG AGAINST T2?". The bot takes that prompt unconditionally — the engine only raises it when an arrival is certainly about to collide, so there is nothing left to judge. It still never plays one, because the prompt needs two things to coincide: an arrival that would collide (0.14 collisions per game, about one game in seven) AND the district's owner holding a Red Flags card at that moment, from a three-card hand drawn out of 121.

What is left to fix is the OTHER half of the card, and it is the half a human would use: planting a flag on purpose to buy a Stage for switching. That needs the bot to know it wants time, which it has no notion of today. Until then the anomaly canary in sim.test.ts is measuring deck luck rather than reachability, and its comment now says so.

#57 — THE BOT'S PRIORITIES ARE NOT THE PROBLEM — measured.

THE BOT'S PRIORITIES ARE NOT THE PROBLEM — measured. Ten heuristic variations, each paired

over 400+ seeds. Every reordering of what the bot prefers came out inside the noise; the only thing that moved revenue was refusing to schedule a train the Office cannot hold (+1.09 ± 0.16, t = 6.79 at 1600 seeds, revenue 1.19 → 2.28). Notable failures, all instructive:

• Refusing to bury the engine costs more than it saves (−0.35, t = −2.98). It works — burial falls from 8.4 decisions a game to 0.03 — and freight halves with it, because coupling is mandatory (§A.4): the moves that bury the engine ARE the moves that pick cars up. Burial is the price of collecting, not a mistake.
• Reserving Moves to get home costs 0.55 (t = −2.32), though 62 of 120 trains left on the board at game end were stranded in the district. The switching work is worth more than the departures.
• Granting clearance when the train ahead has one Stage left is −0.98 (t = −5.24). Trains move in numeric order, so a follower can enter the region the leader still occupies before the leader moves. "About to leave" is not "gone".
• Preferring coaches at make-up, stocking the platform first, playing Interlocking earlier, hunting the Depot in the Departments: all within noise, and three of them were exact no-ops — Interlocking sits in hand alongside a train card 0.04 decisions a game. The funnel says why: only 8% of Cargo phases have a stocked green box, and the bot already takes 42% of the turns where stocking is productive. The opportunities are not there to be prioritised better. What is left is the economy itself, which is a deck question.

#59 — THE RUN-AROUND IS OUT OF REACH OF ANY BOT, AND THE DECK IS WHY — measu…

THE RUN-AROUND IS OUT OF REACH OF ANY BOT, AND THE DECK IS WHY — measured, five ways.

"Teach the bot to plan across turns" was tried properly and does not work. Every attempt is neutral or negative, and they fail for one reason that the numbers make plain.

attempt	result
hold ALL track for the siding	−0.70 (t = −3.27)
hold only CURVES, the closing piece	−0.26 (t = −3.22), district 17.9 → 16.7 cards
finish a run before cutting another way down	0.00 — 398/400 games identical
treat a second turnout as the closing piece	0.00 — 400/400 identical
spend a curve only on a square that CLOSES	−0.11, and only 15 games in 400 differ at all

The pieces never meet. Over 12,000 Local Operations turns: a turnout and a curve are in hand together on 0.3% of them, and a turnout with a MATCHING-hand curve on 0.2% — about once every eight games. A run-around needs five specific pieces of the right hands in a usable order; the bot does not get to the two-piece prerequisite.

And it is not hand pressure. The hand is FULL — mean 2.66 cards, at the three-card limit on 78% of turns. The bot plays 11.4 track cards a game and discards 1.5, so it spends the pieces as they arrive because a piece that builds anything outscores holding one that might build more later. Holding is the only counter, and holding measures worse every way it is tried.

This is a consequence of moving track into the deck, not a bot weakness: 91 run-arounds per 100 games when track was a private 26-piece supply the player chose from, 29/100 once it was drawn, 4/60 now. If the run-around is meant to be the central switching puzzle — and the rules present it that way — the supply has to change, not the player. Options: give track its own hand or yard the way the prototype did, raise the hand limit for track specifically, or print a siding as a single card. Nothing else reaches it.

#53 — THE BOT DOES NOT KNOW TO BRING AN EXPEDITED TRAIN BACK TO THE STATION…

THE BOT DOES NOT KNOW TO BRING AN EXPEDITED TRAIN BACK TO THE STATION — new in v0.4.9.

The expediteFault mechanic (§7, Q3) charges 1 Revenue every Mainline Phase an expedited train is left off the Office square, and the bot has no heuristic that accounts for it: measured over 30 fresh games, one left Train 4 (3/4 Express) parked on Secondary Track from Day 3 Stage 10 to the end of the game, drawing the fault 26 times. Not an engine bug — the mechanism fires exactly as designed — but a clear next bot heuristic: prefer ending a switching turn with any expedited crew back on the Office square, at least once it has finished the work it went out for.

#54 — THE BOT CANNOT SPOT A CAR AT A STUB INDUSTRY, and the cut-ordering rul…

THE BOT CANNOT SPOT A CAR AT A STUB INDUSTRY, and the cut-ordering rules made that visible.

Coupling is mandatory on your own square now (v0.4.7), so a crew that sets a car out between itself and the only way out picks it straight back up. At a stub industry that is every set-out the bot makes: its trains run engine-first with all four cars behind, so the tail cut always lands on the exit side. The correct play is §A.5's facing point move — couple the car onto the nose, shove it into the stub, set out off the nose, back away — which is the same cross-turn planning already recorded as out of reach of any bot in "THE RUN-AROUND IS OUT OF REACH OF ANY BOT" below.

Measured over 200 paired seeds: -0.55 revenue (t = -3.63) and freight revenue 1.11 → 0.56. Filtering self-recoupling moves out of the bot's options took recoupling from 625 of 1,029 set-outs in 60 games to 101 of 677, and all 101 that remain are this case. Nothing is broken — the game models the difficulty correctly and the bot cannot yet play it — but every revenue figure in this file measured before v0.4.7 is now low by roughly half a point and the rebalance pass should not read the drop as a deck problem.

#58 — The bot cannot get a crew next to an industry, so Flying Switch never…

The bot cannot get a crew next to an industry, so Flying Switch never fires. Industries are

now stub-only and the bot places 2.23 a game (was 3.84), in districts averaging under two rows deep. flyingSwitch is exempted by name in the reachability sweep in sim.test.ts; deleting that line is the test that this is fixed. Same root cause as the item below.

#55 — BOT DRIFT ACROSS THIS RELEASE — four measurements, all for the rebalan…

BOT DRIFT ACROSS THIS RELEASE — four measurements, all for the rebalance pass. Recorded

together so the pattern is visible rather than four relaxed thresholds nobody adds up:

• Switching work down ~16%, 1.76 → 1.48 productive acts a game (400 games), because an expedited train now stands at the Office for a Stage instead of passing straight through, and a train on the A/D track and the Office square is in the crew's way. That is the change doing its job rather than a fault — but it is drift. Collisions also went 0.05 → 0.06 and the worst game went −3 → −9, same cause: the Office fills up. Separately, the work > 2 floor that caught this had never actually been met — it read 2.16 at 150 games and 1.76 at 400, so it was passing on which seeds the sample happened to include. Now 400 games and a floor of 1.2.
• Track laid badly, 7% → 15% of pieces butting a card that cannot accept them. Forced to shed on turn one, the bot would rather lay a piece than discard it; a player would discard the ones with nowhere good to go. It also means the district-size gain from the new deal is partly padding rather than useful railroad.
• Interlocking placed, 15/60 → 7/60 games. Departure Revenue pulls the bot toward other work and it spends its opening on the track it was dealt.
• Aimless shuttling in 3 games of 16 — thirteen are clean, so this is a minority behaviour rather than the every-game waste the detector was written for. Each floor was moved to match what is measured, with the reasoning written into the test. None is a crisis on its own; together they say the bot spends its openings worse than it did.

#56 — The bot was partly living off an illegal placement.

The bot was partly living off an illegal placement. Barring curves from the Running Track

(they have no east-west road and dead-end the main) cost it districts 28.0 → 19.7 cards and revenue ~2.0 → 0.8. It has no plan for where a curve should go once the easy square is gone. Same root cause as "the bot cannot get a crew next to an industry" and "THE RUN-AROUND IS OUT OF REACH OF ANY BOT" below; fix them together, after the rebalance.

#60 — Re-run the three "worth 0" action-mix experiments against the new floor.

Re-run the three "worth ~0" action-mix experiments against the new floor. Capping the draw,

pairing the two halves of a load, and restricting Enhancements were each measured "within noise of zero" over 400 games — but at 400 games the standard error is ±0.33, so a real +0.5 would have looked like nothing. They are nearly free to re-run now and at least one may have been discarded wrongly.

Code health and housekeeping

#46 — tsc --noUnusedLocals finds 29 unused declarations across 14 files, and…

**tsc --noUnusedLocals finds 29 unused declarations across 14 files, and the build does not

run it.** Swept 2026-08-30 and deliberately NOT fixed in the same pass, at Jesse's call — it is a wide, mechanical change and v0.7.9 had enough in it.

The recurrence is the point, not the 29. SIDE_GAP and CHIP_W in board-svg.ts are both Gitea#18 leftovers — constants for a wrapped layout that no longer exists. SIDE_GAP was removed by hand on 2026-08-30 only because someone happened to read the file while closing the superseded Display items; CHIP_W sat two lines away and was missed, and would still be there. A flag finds both without anybody having to be reading that file.

Where they are: sim/replay.ts 10 (the standalone dev tool, which has its own keep-or-delete question under Replay / Save Games — do that first, since deleting it settles ten of these), engine/apply.ts 3, sim/bot.ts 2, engine/legal.ts 2, engine/intents.ts 2, and one each in engine/setup.ts, sim/board-svg.ts, web/game.ts and five test files.

Two halves, and the second is the one that lasts: delete what is dead, then turn noUnusedLocals (and probably noUnusedParameters) on in tsconfig.json so it cannot come back. Without the flag this list simply regrows — it is regrowing now. Worth checking whether the engine ones (LABORER_ACTIONS_PER_LOAD, spaceOn, mainlineProfile, wouldBuryTheEngine, strandedWantedCars) are dead or are half-built work somebody meant to come back to; tsc proves nothing references them, not that nothing should.

#45 — THE FRAME CARRIES FOUR THINGS NOTHING READS — audited 2026-08-30, at J…

THE FRAME CARRIES FOUR THINGS NOTHING READS — audited 2026-08-30, at Jesse's asking. The

prompt was that v0.7.9 found the same defect twice: actingPlayer existed and the Frame threw it away (#43), and collisionsToday/collisionsTotal rode the Frame for three releases with nothing on the board drawing them (#28). So the question became general — what ELSE is being computed, serialised and sent to no one?

Method, so it can be repeated: every top-level Frame field grepped for a .field read across the seven renderers (web/main.ts, web/panels.ts, web/replays.ts, web/lobby.ts, sim/turnchart.ts, sim/board-svg.ts, sim/replay.ts), then the same for Tally's members. 55 of 59 top-level fields are read. These four are not:

• viewerSeat — on the Frame, and the ONLY reference in the whole repo outside view.ts is one assertion in test/multiplayer.test.ts proving it is the seat and not the player index. Nothing renders it. It may well be right to keep — a remote client arguably needs to know which seat it is looking at — but nothing needs it today, and a field kept for a future caller should say so.
• overHandLimit — the fullest example of the shape. The engine computes it, view.ts puts it on the Frame, web/session.ts declares it on the Session interface AND implements it twice (locally by calling the engine, remotely by reading the Frame) — and the only caller anywhere is its own test. A whole plumbed path, four layers deep, with no consumer.
• tally.unloadsBegun — and this one is a visible asymmetry rather than merely unused. The results screen shows "Loads still in the pipeline" as loadsStarted - loadsCompleted (panels.ts), which is the MEN | AT | WORK loading pipeline's in-flight count. The unloading pipeline has exactly the same pair of counters and gets no such line, so the screen reports half of a symmetric mechanism.
• tally.cardsDiscarded — counted by the engine, and the results screen lists "Cards drawn" and "Cards played" beside it without it. Gitea#9 (v0.7.1) made discarding a Timetabled train a legal and deliberate move, so this is a player CHOICE that the game counts and never reports.

The two tally ones are FIXED — 2026-08-30 in v0.7.9, one line each on the results screen. Fixing them turned up something else worth knowing: resultsHtml draws tallyHtml(report?.tally ?? f.tally), so a finished game reports the tally frozen in f.official, not the live one. A test that overrides f.tally alone changes nothing on screen, which is how the first attempt at pinning this failed for a reason unrelated to the fix.

The two plumbing ones — viewerSeat and overHandLimit — are DEFERRED (Jesse, 2026-08-30): "leave it for now." They cost nothing today and v0.7.9 had enough in it. The decision when it comes is delete-or-document, not a patch: either strip the field and the Session-interface method, or write down what each is being carried ahead of a consumer FOR, so the next audit does not re-flag them.

#84 — FIVE TEST FIXTURES PINNED A SEED AND MEANT "A GAME LIKE THIS".

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.

#87 — Regions as the primary model (the other half of §8.2).

Regions as the primary model (the other half of §8.2). The Division map now DRAWS regions,

deriving position from what the crossing already cost. The engine still models a crossing as a countdown of Stages, so two things printed on the cards remain unimplemented:

• entryPoints is declared on every Mainline profile and read nowhere. The Heavy Grade card has five named Start positions, and playing Brakeman is supposed to move your entry point along the card. The engine gets the same ANSWER by taking a Stage off the crossing, which is why the derived drawing looks right — but the mechanism is not the printed one, so a card whose starts do not correspond to its speed would be drawn wrong.
• implications.md §6 calls this "the single largest mechanical gap" and asks for typed cards with speeds and named entries, with crossing time DERIVED from the region walk. Doing it properly changes movement, so it invalidates every balance figure — revenue 8.7, the freight numbers, all of it — and needs a full paired re-measure over 400 seeds. Needs the source Start-position art for the ten card types before it can begin.

Documentation and assets

#15a — Documentation generated from the implementation, not written alongside…

Documentation generated from the implementation, not written alongside it. Raised by Jesse

2026-08-22, immediately after Gitea#7 changed the coach counts on four train cards and the answer to "where do we keep track of that?" turned out to be five places of three different vintages: src/engine/content.ts (the truth), docs/StationMaster-Home-Deck-v0.4.5.md (a readable per-card table, a version-stamped snapshot), docs/rules/implications.md §5 (the transcription of Trains3.pdf, deliberately frozen at what the design SAYS), docs/Trains3.pdf (the artwork), and docs/rules/card-reference.md (an invented placeholder catalogue, banner-marked SUPERSEDED, whose train table still looks authoritative if you land in the middle of the file). Every hand-maintained one of those drifts the moment a card changes, and this release proved it.

The deliverable, at minimum: a reference document for every card that can be played, generated from content.ts so it cannot disagree with the game. Sections, in order:

1. Mainline cards — the Division's own deck, dealt at setup rather than held in hand.
2. Home Deck Cards — Trains
3. Home Deck Cards — Track
4. Home Deck Cards — Industry
5. Home Deck Cards — Modifiers
6. Home Deck Cards — PVP

Each card wants its name, what it does, where it may be placed, and — the part only the implementation knows — whether its printed effect actually resolves yet. content.ts already carries that last one for Enhancements (EnhancementRule.effect, live / dormantSolo / unbuilt, each row citing the file that reads it); the same honesty is what makes a generated reference worth more than a transcription. enhancementText() and mainlineDescription() are the model: prose composed from the data, so a tooltip cannot drift from the rule it describes.

Deliberately NOT including card counts per category. Jesse's call in the same breath: the counts move with play balance, so a document that prints them is stale on the next retune. The same rule was applied to content.ts's own comments on 2026-08-22 — see the pass recorded in CHANGELOG for what came out and what was kept.

BUILT 2026-09-07 in v0.7.9.2, completed in v0.7.9.3, as the first of the two options — a build step writing Markdown, scripts/build-card-reference.ts → docs/rules/as-built.md, with test/card-reference.test.ts re-running the generator and failing when the checked-in file disagrees. All six sections are there, and so is the honesty column: every Enhancement carries its live / dormantSolo / unbuilt status, and the opponent-directed cards say plainly that none of them is dealt. The counts are omitted as ruled — where a count matters it is rendered as a yes/no "is this dealt at all", which is a fact about the design rather than about the current tuning.

WHAT REMAINS, and it is the second option rather than a gap in the first: a page on the site, beside the replay viewer, rendered from the same view-model the game uses. The argument for it is unchanged — a player cannot read a file in the repo — and it is now cheap, because the projection work is done and only the presentation is missing. Do it with Gitea#20 step 3, which builds a renderer for exactly this kind of read-only public page; building a second one first would be the waste.

#86 — Real audio, as committed assets.

Real audio, as committed assets. Everything the game plays is synthesised from oscillators

(src/web/sound.ts), which was the honest choice for a site that fetches nothing — but it is a placeholder, not the finished sound. Sound therefore defaults to OFF.

• "All aboard" most of all. It currently goes through the browser's speechSynthesis, so it is whatever system voice the player happens to have — a robot, not a conductor. A real clip is the single biggest improvement available here.
• arrive (a train pulling into an Office), depart (a train highballing out of one) and crash (§10 — a collision) are now synthesised too, v0.4.9 — three chuffing/screeching cues built from the same oscillator-and-filtered-noise toolkit as stage, wired to trainArrived, trainHighballed (Office departures only), and trainsDestroyed. Good enough to keep as the real thing rather than a placeholder — no WAV clips needed for these three.
• Find and add the rest as assets: steam whistle, grade-crossing bell, couplers clashing. Every file added needs three things recorded alongside it: the sound file itself, its source (where it was obtained from), and its license. The preferred license is CC0 ("Creative Commons Zero") — a public-domain dedication: the creator waives all copyright and related rights, so the file may be used, modified, and redistributed for any purpose, including commercial, with no attribution required and no restriction. That is the cleanest fit for a file committed straight into the repo, since it needs no attribution to track going forward. Only fall back to an equally-permissive alternative (e.g. a license that explicitly permits redistribution with no ongoing obligation) if CC0 isn't available for a given sound, and record that license's actual terms plainly rather than assuming they match CC0. Files also need to be small enough to commit, and a check that the "fetches nothing external" test still passes — assets must be served from the site's own folder, never hot-linked.
• Keep the synthesised versions as the fallback for anything not sourced, so a missing file is a quieter game rather than a broken one.

#88 — CLOSED — card-reference.md's industry table is no longer anybody's source.

CLOSED 2026-09-07 in v0.7.9.3, by removing the question rather than answering it. This asked

whether card-reference.md's Mine Tipple, Produce Shed and Power Plant rows were stale — it prints 3/3/4 for two of them while content.ts has every industry at base 1 out / 1 in / 1 Laborer — and deliberately did NOT rewrite them, on the grounds that changing a Laborer count is a balance decision rather than a documentation one. That reasoning still stands and nothing was changed in the engine.

What changed is that card-reference.md is no longer where anyone looks. docs/rules/as-built.md is generated from content.ts and carries the industry table the game actually runs; the old file keeps its SUPERSEDED banner, now pointing forward, and its numbers are read as what the v0.4.5 placeholder said. The balance question the entry was really guarding is #70 (the rolling stock supply and the uniform 1/1/1 industry model, both marked provisional in content.ts) — that is where it belongs, and it is still open.

Done

Closed items, kept because several are the only record of a ruling or a lesson. Newest first within each group.

Shipped through v0.7.9.4, from the queue

Closed items, newest first. Kept because several of them are the only record of a ruling or a lesson; the numbers stay so cross-references above and below still resolve.

91. Narration was outside the redaction net, and so was everything else nobody had listed. — done 2026-09-07 in v0.7.9.4. The systematic pass Gitea#20 step 1 asks for: serialise a seat's Frame, the public board and the narration it receives, and search all three for every opponent card id, every card NAME that is unique to one opponent's hand, the seed, and any private decision or menu data — across a fresh game, a blind draw, mid-game, a pending decision, Employee Rotation before and after the seating moves, a reconnect push and a played-out game. And the allow-list, which is the plan's actual acceptance bar: every property of publicSnapshot is written down and compared, so adding a field fails the suite until somebody has said out loud that a spectator may see it. That is the check that would have caught both v0.7.9.2 leaks, since both were fields nobody had asked the question about.

    Worth knowing, and it cost two false failures to learn: a card NAME is a type, not an identity. "right-hand curve" names a dozen cards and one is legitimately drawn on the board as a cell label the moment anybody lays track, so searching for it fails on correct code. A name counts as evidence only when EVERY card bearing it is in the one hand. Ids need no such care. And a one-digit seed makes the seed check meaningless — seed 7 matched "Train 7". The seeds in this file are nine digits deliberately.

    Proved by mutation rather than by passing: restoring the seed line fails 6 tests, restoring the blind-draw name fails 1, adding a private field to the public projection fails 7, and making players[] carry hand contents instead of a count fails 5.

    One item on the plan's list has no test because it has no referent: there is no secret objective in this game. objectiveOf derives from config.minCombinedRevenue and the player's own Revenue, both public. Said here so the next reader does not go looking for the gap.

92. A Red Flag standing at an Office's Limits was drawn nowhere. — done 2026-09-07 in v0.7.9.4. It is a token set out ON the board that holds the next train arriving from that side; it was announced once in the log and then existed only in the engine, so a train would stop short with its only explanation scrolled out of the panel. DivisionView's office node carries redFlag now and the map draws a staff and pennant at the end it guards — west on the left, east on the right — because which approach it covers is the whole of the information; a flag in the middle would say one is out and leave the reader to hover for the half that decides whether to run a train. Worth knowing: the same class as Gitea#21 and #22, and the third in a row — when the engine gains a thing that CHANGES what a train may do, ask where it is drawn before asking whether it works.

93. The public projection helpers — Gitea#20 step 1's foundation. — done 2026-09-07 in v0.7.9.4. projectDistrict(state, seat), projectDivision(state), projectSharedTable(state), publicSnapshot(state) and currentActorOfState(state), with snapshot() rebuilt to compose from the same helpers rather than keeping its own copy — so a player's frame and a spectator's cannot come to disagree about the clock, the phase or the score. Behaviour-neutral, and the existing 897 tests are the proof of that.

    Districts are keyed by SEAT and the player resolved through playerAtSeat, because Employee Rotation moves players between districts; a public board that assumed seat and player index were interchangeable would relabel every district the first time anybody rotated. And the public view is composed upward, never by calling snapshot() once per seat — that shortcut builds every private field and then has to remember to strip it, and snapshot defaults its viewer to player zero, so a careless spectator call would have served seat 0's hand.

    One plan finding struck off rather than fixed: it warns that a public display reading clock.currentActor could highlight the wrong district during a decision. Measured across six seeds and 3,600 decision points, that field and actingPlayer never disagreed — both are only consulted when somebody is genuinely acting. currentActorOfState exists anyway, as one place for the next reader to ask.

94. Tell the 0.4.9 playtesters their saves are dead, before they find out. — done 2026-09-07. PLAYTEST-0.7.4.md was written for exactly this and did its job; Jesse, 2026-09-07: "a temporary document to help some of the playtesters out on making the big jump, but that is no longer needed." The jump is made, so the note is retired rather than committed. No per-version save-compatibility fact is tracked from here, on Jesse's call 2026-09-07: a save replays through the current rules, so an older one breaking is the design working rather than an event to log each time. The general rule lives in README.md § Design notes; #40 is the same shape and has been generalised to match.

95. Two multiplayer information leaks in the shared narration log. — done 2026-09-07 in v0.7.9.2, found while planning Gitea#20 step 1. The seed was announced in the opening line of every multiplayer game and a blind Home Office draw named the card — and linesSince(seat) slices one shared game.log with no per-seat filter, so both went to every player. Ruling: solitaire keeps both, because a one-seat table has nobody to leak to, the seed is what a bug report quotes, and a solo player's history naming their own draw is the record. A Department slot is face up and stays named for the same reason. Worth knowing: these were not found by a test, they were found by reading the plan — every redaction test passes [] for the log, so the whole of narration was unchecked. #91 is what remains.

96. docs/rules/ had no current description of the game, and the code pointed at a superseded one. — done 2026-09-07 in v0.7.9.2. content.ts named card-reference.md as "the place that now carries what the cards say" while that file's own banner said not to use its numbers; it describes the v0.4.5 deck, where 3/4 is a Mail-Express with three coaches against a content.ts whose train 3 is the Express with two freight cars. The fix is not a rewritten table — every file in docs/rules/ is a deliberate historical record and worth more intact than patched. A new as-built.md is GENERATED from the same catalogues the engine instantiates from, by scripts/build-card-reference.ts (npm run build:cards), and test/card-reference.test.ts re-runs the generator and fails if the checked-in file disagrees. Worth knowing: a hand-written replacement would have drifted the same way and for the same reason — nothing fails when a table falls behind a constant. Generate it or check it; do not retype it.

97. The map drew westbound trains in the wrong half of a Mainline card. — done 2026-09-07 in v0.7.9.1, Gitea#22. regionOfTransit counts from the end a train ENTERED, which is what the collision rules want; the map wanted "which printed box, left to right" and used the same number, so every westbound card came out mirrored. It cost a collision: the Superintendent cleared Train 3 to follow T5 and it ran into TX17, which the picture had drawn ahead of T5 instead of behind it. Worth knowing: the engine was right and only the picture lied, so the fix is one mirror in view.ts at the drawing boundary — and every existing region test ran eastbound, where the mirror is the identity, which is exactly why it survived them. The same shape as the mirrored consist row at the Whistle Post (seed 270861860): a number meaning "distance run" used where the screen means "place". Ask of any new Frame field whether it is a distance or a position.

98. The Blocked panel explained a refusal by describing something else entirely. — done 2026-09-07 in v0.7.9.1, Gitea#21. A second tank car would not come off at a refinery; the panel said the refinery's green box was empty. The real answer was Train 3's printed rule — the Express may work one freight car per location — so the refusal was correct and the panel sent the player to spend a Freight Agent action that could not have helped. The ruling: no rule changed, only what the screen says about it. The panel asks the reducer's own freightBudgetLeft through a new exported freightRuleSpentHere, so its words cannot drift from the rule. Worth knowing: a panel that answers the wrong question is worse than one that stays silent, because it looks like an answer — and a rule that only lives on a card tooltip is invisible at the moment a player notices a button is missing.

99. "Waiting on" said nobody while the game was stopped on the Superintendent. — done 2026-08-30 in v0.7.9. Frame.actor carried clock.currentActor, null for the whole Mainline Phase, so all three interruptions (clearance, Yard Office, Red Flag) reported that the Division was running itself. It carries actingPlayer now and a new awaiting field says what the question is and which train it is about. Worth knowing for the next Frame field: the engine had the right answer in actingPlayer since the Gitea#5 refactor, and the Frame simply did not carry it. A view that reads one field of clock directly is a place this can happen again.

100. replay.ts printed a raw outcome enum, exactly as the results screen used to. — done 2026-08-30 in v0.7.9. Its summary line rendered "loss — revenueFloor", the same defect Gitea#16 was filed about, alive in the dev-side viewer a release after the playable page was fixed. panels.ts's reasonSentence is exported and shared rather than reimplemented, so the replay and the results screen cannot explain one ending two different ways; the drift test maps win/loss to won/lost so it still checks agreement rather than spelling.

101. Put the Fedora at the right-hand end of the phase row. — done 2026-08-30 in v0.7.9. Details, and why the name does NOT go inside the Supervisor Shift pill, in Display below.

102. Solitaire must ask before it deals, the same way multiplayer's lobby already does. — done 2026-08-29 in v0.7.5, and then fixed three more times. Jesse: "let the user choose their options like the start of a multiplayer game"; "asking first is the only path." A #solitairesetup screen asks the full shared block before a genuinely fresh visit deals anything; a saved game, an explicit ?seed=, or a URL a Deal already wrote all skip past it. The in-game dialog, the lobby and this screen share one wireGameTypeBlock()/commitNewGame() pair.

     THE LESSON, AND IT IS THE MOST EXPENSIVE ONE IN THIS FILE. The same report came back three times, and each of the first two "fixes" corrected something real that was not the reported fault.

     • v0.7.6 — the splash's "Play solitaire" door landed in a leftover Co-op lobby: start() checked a remembered multiplayer session before looking at solitaire's own state. Real bug. Not the one reported.
     • v0.7.7 — every packaged build published the same cache-bust key (?v=nogit, because the .s9pk build has no .git for git rev-parse) and the server sent no Cache-Control, so neither release ever reached the browser that asked for it. Real bug. Still not the one reported. When a fix appears to have had no effect, check that it ARRIVED before re-diagnosing it — a hard reload would have answered it on the first report.
     • v0.7.8 — the actual fault: v0.7.5 skipped the setup screen whenever a save existed ("a saved game is a game to resume"), so any browser that had ever played solitaire could never reach it again. The private window that seemed to vindicate v0.7.7 simply had no save.

     Each of the three was reported as verified, and each verification read what the SERVER served rather than exercising the path with the state a returning player actually has. The thing that worked was a failing test written before the fix. Next time a report repeats: reproduce the user's state first, and treat "I verified it" as unearned until something failed the way they described.

103. Gitea#13, #5 and #19 — three rules corrections. — done 2026-08-29 in v0.7.4 (5e34c73, 2280276, 19a6a47), wrapper 085b88b as 0.7.4:0, installed on phoenix.local. Each issue carries a comment naming its commit and what was ruled. What they left behind is #39, #40 and #41 above.

104. Bump the StartOS wrapper to 0.7.2. — done 2026-08-26 (1bfea8d). Folded into the standing practice at #37 above, which is where the sequence now lives.

25-27. Three Division-map drawing reports from the v0.7.0 build — no track geometry, the buffer stop pointing the wrong way at two players, and east not always being to the right. All answered by Gitea#18; the last was FIXED OUTRIGHT by it and is the reason a single row won. Collapsed into one entry in Display below.

19-22, 24. Five drawing items against the wrapped Division map — vertical track art, the inter-row connector, overflowing captions, filling the dead centre, and seating the viewer at the bottom. All SUPERSEDED by Gitea#18, which replaced the layout rather than fixing the drawing. Two ideas survive them and are recorded in Display below: "SHARED in the middle, YOURS on the right" (now Gitea#20's territory), and why rotating the map to seat a viewer was refused.

9-11, 12a-12g. The v0.4.9d and v0.4.9e gameplay-testing reports — twelve bugs, all shipped in v0.4.9e / v0.7.1. Two trains in one station answering to one button; a Freight House unloading the boxcar it just loaded; the Grocer's Warehouse shipping and the Refinery receiving; Gitea#4 extra train starts; #7 coach counts on four train cards; #6 then #9 on discarding train cards (#9 superseded #6 three days later — a Timetabled train may be tossed face-up to a Department slot, an Extra may not); #8 the per-diem train and the caboose; and #10 the Day-rollover dialog. Reasoning for each is in docs/rules/implications.md and CHANGELOG.md.

**Gitea#2 carries a RULING worth keeping.** Four porters, two passengers on the platform, and
only one may be worked — RULED AND FIXED, but not the way the report implied. The engine is
faithful to the written rules at every step; what bites is that BOTH directions of porter work
move coaches one-way into a Classification Yard that comes back only when the Division Yard is
bare of all ~60 cars. Sixteen coaches in the game, and the reported save runs dry on Day 5 with
eight stranded. **Jesse's ruling: the shortage stays** — "it is possible to run out, that's part
of the strategy" — so the three balance options in Play Balance below are DECLINED, not deferred.
What was actually wrong is that the game said NOTHING: a Porter action that cannot be taken is
simply absent from the menu, and the "why is nothing moving?" panel covered freight facilities
only. That half is fixed.

1-6. The 2026-08-20 multiplayer planning queue and the first StartOS play session — the New Train phase car-placement round, unified victory conditions, Phase 2 server core, the lobby offering every game parameter, deciding what the four optionalRules are (sisterTrains deleted, employeeRotation implemented, the other two already live), and stopping every release from destroying every game in progress — done by replaying the save rather than comparing version strings. All shipped in v0.6.0; reasoning in Multiplayer below.

Closed items, by the area they were in

• fromSave's replayed narration loses "Player X" attribution — fixed 2026-08-30 in v0.7.9, one argument: record(game, result.events, actor), with actor already computed on the line above.

  **The reason it survived is the lesson.** This entry says nothing ever compares a
  `fromSave`-built log against a LIVE-played one, and that was exactly right — the whole existing
  test suite stayed green with the bug in place, because the one log-comparing test compares two
  `fromSave`-built logs and the gap cancelled out on both sides. The new test plays a game,
  saves it, restores it, and asserts the two logs are identical: **the missing direction, not a
  new requirement.** Confirmed to go red with the fix reverted.
  
  Original note below.
  
  **`fromSave`'s replayed narration lost "Player X" attribution — found 2026-08-20 building
  multiplayer Phase 3, not fixed there.** `fromSave`'s loop (`game.ts`) calls `record(game,
  result.events)` without the `actor` argument `submit()` always passes it (`game.ts`'s own
  `record(game, events, actor)` — `actor` is what turns "Chose to draw a card" into "Player X
  chose to draw a card"). So a restored save, an undone game (`undo` rebuilds via `fromSave`
  internally), or a replayed one all lose attribution on every line — invisible in solitaire
  because nothing ever compares a `fromSave`-built log against a live-played one (the one test
  that compares logs, `test/web.test.ts`'s "leaves nothing in the log describing a move that was
  taken back", compares `undo`'s `fromSave`-built log against ANOTHER `fromSave`-built log, so
  the missing attribution cancels out both sides), but it would read as broken the moment more
  than one seat's history is on screen at once — exactly what the replay viewer and any
  multiplayer post-game replay (D20) need to get right. Fixed in `fromMultiplayerSave`
  (multiplayer's version of this function, added for Phase 3) by passing `actor` through; not
  touched in `fromSave` itself since it's used far more widely (undo, save/restore, the replay
  viewer) and deserves its own careful look rather than a fix bundled into an unrelated change.
  

• A blocked PASSENGER facility produces no impediment at all — FIXED. impediments() (src/sim/narrate.ts) opened with if (!f || f.kind !== 'freight') continue, so the "why nothing is moving" panel had never had anything to say about a platform. That was the second half of Gitea#2 and the half that was unambiguously a bug: the player above was not merely blocked, he was given no reason — the button simply was not there. A platform now reports passengers with no train, a train the card bars Porters from working, full coaches, full red slots, the same-district rule, and the coach shortage itself — the last naming how many coaches are stranded in Classification and what brings them back. The reason comes from passengerRefusal, the engine's own predicate, so the panel cannot drift from the rule that actually refused. Fixing the label found a second defect: a Passenger Facility rides on the office card, so every passenger row would have read facility 0,0 next to mineTipple 1,-3; it is named by its tier now.

• The log lowercases the first letter of every narration it attributes to a player — fixed 2026-08-30 in v0.7.9. EXTRA X18 started… rendered as Player Solitaire eXTRA X18 started…, and the same happened to TRAIN 1 MADE UP and COLLISION. A new uncapitalise folds the opening word only when it is sentence-cased — the test is ^[A-Z][a-z], a capital followed by a lower-case letter, which is an ordinary word capitalised because it began a sentence and nothing else is. That also leaves X22 Pee-Dee alone, which a naive "is it uppercase?" check gets wrong because '2'.toUpperCase() === '2'.

  **Misfiled here, and worth saying so.** This sat under Play Balance and is a text bug with no
  bearing on balance at all — which is why it survived a session that had explicitly ruled
  balance work out of scope. Found again 2026-08-30 only by reading the section it did not
  belong to.
  

• The lobby, the setup form and the start of a game — done 2026-08-23 (Jesse's cleanup pass). Kept for the reasoning, since several of these were decisions rather than fixes.

  **The game types.** Co-op, Competitive, Cutthroat, Solitaire and Custom (`src/web/presets.ts`),
  replacing the old two-mode radio. A type is a set of DEFAULTS, not a ruleset: every rule stays
  editable, and editing one selects Custom, which keeps the scoring of the type it came from.
  Jesse's numbers — Co-op pays 1 per transit and asks 3 combined Revenue per player per Day;
  Competitive asks 2 and pays nothing for transits; Cutthroat asks nothing at all and lifts the
  whole-game collision cap, leaving three-in-one-Day as the only shared way to lose; every type
  deals six cards. **The Revenue floor is a formula**, so table size and Day count re-derive it
  rather than making a game Custom — which is why those two, and the seed, sit above the type
  radios as parameters. The type is DERIVED from the numbers, never stored, so no saved game
  carries a label that can disagree with itself (`test/presets.test.ts`).
  
  **Where an Extra may start was missing from the lobby entirely**, so every multiplayer game
  ever played used the most permissive setting (`anyOffice` — an Extra may be planted in another
  player's district) and no host was ever asked. It is a Cutthroat-only default now. The reverse
  hole existed too: the solitaire dialog had none of the three optional rules. Both screens ask
  the same eleven questions through one shared module (`settings-form.ts`), and a test asserts
  the markup carries every field on both — the drift is what motivated the shared block.
  
  **The dead PvP checkbox is gone from both screens.** `buildDeck` ANDs `pvpCardsAllowed` with a
  hard-coded `cardsImplemented = false`, so the control could not do anything whatever it was
  set to. The 22 opponent-directed cards (and the 7 defences held out with them) are now a
  property of the game type, and the fact is stated in words where the checkbox was.
  
  **The lobby itself:** joining is a door of its own rather than a heading below fifteen fields
  the joiner has no use for; a stored join secret collapses to one line and re-opens on a 403; a
  display name is remembered and may not duplicate another at the same table (`NAME_TAKEN`); the
  seating list numbers bots as the game will; the code copies as a code AND as an invite link;
  server codes are translated into sentences in a red block instead of `.dim` grey; the lobby
  stream has an `onerror` that tells a blip from a dead lobby (and finds a game that started
  while the connection was down); and **a player may read the whole rule set before taking a
  seat** (`/api/lobby/preview`, which never carries the seed).
  
  **The start of a game**, which nobody had ever drawn: a handoff curtain with a deliberate beat
  instead of a "connecting" line written into the DISCONNECT banner, an announcement naming the
  game and its type, a marker at the top of the log so the bots' opening turns are visibly after
  the start, the game code and the type in the header for the rest of the game, and a Start
  button that cannot be pressed twice.
  
  **The four transient signals reach a remote client at last.** `createRemoteSession` answered
  all four with empty values, so multiplayer had no sound, no timetable flash, no announcement
  when a completed run paid the table, and no badge on the card you had just drawn. `justDrawn`
  is the redaction-sensitive one — `game.justDrawn` is one field for the whole game and does not
  say whose card it is — so the server remembers who drew and sends it to that seat alone.
  
  **Two things found by RUNNING it rather than reading it.** A bot seat was being reported as a
  disconnected player, which would have put "waiting on Bot 1 — not here yet" on every screen
  for a whole game. And the six-card opening survives a table with bots in it: a three-seat game
  (two humans, one bot) played through Day 1 without stalling on the discard round.
  

• ~WHY DOES A 4-PLAYER COMPETITIVE GAME END AFTER 16 STAGES OF A POSSIBLE 60? Answered 2026-08-20: the collision floor, not the revenue floor. Traced checkVictory (advance.ts:1086-1133): in competitive mode the revenue floor can only fire at the exact Day-5 boundary (Stage 60), so it structurally cannot explain a 16-Stage ending. Only the collision floor can (advance.ts:1076-1080, 3 collisions in one Day, checked at every Stage boundary). collisionsToday is one counter every seat feeds, so a 4-player table burns a fixed shared budget roughly 4x faster than one player would. Also: multiplayer.md §3's 8-game sample predates DEFAULT_HOUSE_RULES (v0.4.2) and most likely ran under what is now LEGACY_HOUSE_RULES — that sizing data is stale on top of the collision-floor explanation. Jesse's call, 2026-08-20: keep the collision caps flat rather than player-scaled (below), so 16-Stage games under default settings are an accepted, deliberate outcome, not something to re-tune away — re-measure multiplayer.md §3's sizing table once the redesign lands, but expect similar early endings by design.

• Victory conditions unified across solitaire, competitive and coop — designed and implemented 2026-08-20. One shared, fully-configurable set of GameConfig dials replaces LENGTH_PROFILES.target, VictoryCondition: 'firstToTarget' (confirmed dead — grepped, never selected anywhere in the codebase today) and the flat COLLISION_FLOOR_PER_DAY constant:

  | dial | meaning | default |
  | --- | --- | --- |
  | `days` | how many Days the game runs | 5, all modes |
  | `minCombinedRevenue` | everyone loses if the table's total Revenue is below this when Days run out | `3 × players × days` — reuses `collectiveRevenueFloor()` (`content.ts:1020`), now also applied to solitaire (1 player) rather than competitive-only |
  | `maxCollisionsPerDay` | everyone loses immediately, mid-game, once collisions in one Day reach this | 3, **flat — not scaled by players.** Jesse's call: more players means more independent chances to collide, not a bigger shared budget, so multiplayer is deliberately riskier than solitaire at the same default |
  | `maxCollisionsTotal` | same, summed across the whole game | 5, flat, same reasoning |
  | `pvpCardsAllowed` | whether the 22 opponent-directed cards (still unbuilt, see below) are in the deck | forced off in solitaire and coop — no valid target for them in either — on by default in competitive |
  
  `0` means "off" for every dial. Win/lose shape is otherwise unchanged from what solitaire
  already does: most Revenue when Days run out wins, unless `minCombinedRevenue` was missed, in
  which case everyone loses — just made configurable per game instead of a fixed `length`
  lookup. Coop keeps its existing "score is the table's total" model, now against a
  configurable floor instead of `profile.target * players.length`.
  
  **New Game dialog:** one shared dialog for all three modes, per Jesse — a mode radio button
  at the top, the same field set underneath for all three, greyed out wherever a mode forces a
  value (the PvP checkbox in solitaire/coop). Solitaire gains the four new dials alongside the
  starting-hand and revenue-rate fields it already has; picking a mode only changes the
  defaults, never the field set. **Deal stays disabled for Competitive/Co-op** with a "needs a
  server" note, since Phase 2 didn't yet expose a way to actually start one from the browser
  (see below) — only Solitaire's Deal path is wired to a real game today.
  

• New Train phase car-placement is one player's job even in competitive mode — fixed 2026-08-20. §7 (rules-v0.2.md:346-363) is explicit: "starting with the Superintendent and working left, each player may place ONE car... the round repeats... until the consist is full," with a worked 2-player example. newTrainPhase (advance.ts:187-282) never implemented the round: enterPhase resets actorOffset = 0 on entering the phase (advance.ts:172) and newTrainPhase never incremented it the way playerPhase does for Local Ops (advance.ts:140), so the actor was always the Superintendent alone, for every car of every train made up that Stage. Fixed by reading the round position off tray.consist.length instead — it already counts placements toward that tray and resets per train with no new state needed. Test in multiplayer.test.ts, "the New Train phase car-placement round rotates."

  **Found in the process, not fixed, logged separately:** `newTrain.passCar`'s `check()`
  (`apply.ts:959-967`) tests whether the *entire* Division Yard is empty rather than whether a
  car suitable for *this* tray exists, and `reduce()` has no case for `carPassed` at all
  (`apply.ts:2246-2247`, falls to `default: break` — applying a pass currently mutates nothing).
  Unreachable in practice today: `trainNeedingCars` only ever flags a tray that already has a
  suitable car waiting, so a legal `passCar` for the flagged tray can't occur. Only matters if a
  future change lets the New Train phase address more than one tray at a time. Not fixed here —
  nothing to verify against an intent that can't legally fire.
  

• The lobby's seat controls could not express "nobody in this chair" — done in v0.5.3. Raised by Jesse 2026-08-21. The seats array grew as people joined, so the four rows on screen were partly fictional: a 2-player game simply started with a 2-long array, and a host who added a bot to a later chair padded the array with a null that silently disabled Start behind a one-line note. The host now picks the table size (2-4) when creating the game and the array is built at that length once, so a gap cannot be expressed rather than merely being rejected. That also removed the need to compact seats at Lobby.Start — which would have shifted the player index every PlayerSession records at join time and that /api/stream and /api/intent route by, quietly handing a player somebody else's railroad. Tested in test/server/lobby.test.ts ("seat index is player index, with no compaction to shift it", "never grows the table, whoever asks", "refuses a chair that is not at the table").

• EVERY RELEASE DESTROYS EVERY GAME IN PROGRESS — fixed in v0.6.0, by option 3. Raised 2026-08-21 after v0.5.2, v0.5.3 and v0.5.4 each killed the games on the StartOS box in turn — v0.5.4's changes were rendering only, and it still refused two saved games.

  **Why it happens, and why the design is right as far as it goes.** A save is a seed plus a
  list of intents (D5), so loading one means replaying those intents through the current engine.
  A move that was legal under the old rules may be rejected under the new ones, and a
  half-replayed game is worse than no game — so `loadGame` refuses on any `engineVersion`
  mismatch and `index.ts` logs it and carries on (D7). Nothing is deleted; rolling the version
  back makes the games loadable again. That is all correct. The problem is only that the test is
  **exact equality against the package version**, which moves for reasons that have nothing to
  do with the rules.
  
  **Why it is getting worse rather than better.** It was harmless while Jesse was the only
  player. It stops being acceptable the moment other people are seated: their game is destroyed
  because somebody shipped a CSS fix. It also interacts badly with the stranded-session bug
  fixed in v0.5.5 — the refusal is precisely what stranded a browser on a blank page.
  
  Three ways out, cheapest first:
  
  1. **A separate rules version, bumped by hand.** `RULES_VERSION` in `content.ts`, stamped into
     the save instead of `package.json`'s version, and raised only when a change can alter
     whether an intent is legal. v0.5.4 would not have touched it and both games would have
     survived. Cheapest and the least clever, but it is a judgement call on every release, and
     getting it wrong silently corrupts a game rather than refusing it — the failure is worse
     than the one it replaces.
  2. **A declared compatibility floor.** The save records the version that wrote it; the engine
     declares the oldest save it will accept. Loading checks `saved >= floor` rather than
     `saved === current`. Same judgement call as (1), but expressed as a range, which makes
     "this release breaks saves" an explicit act rather than the default.
  3. **Verify rather than assume — replay and see.** Load the save, replay it, and refuse only
     if an intent actually rejects. This is the honest test and needs no judgement at all: it
     answers the real question ("does this game still replay?") instead of a proxy for it. It
     costs a full replay per game on boot, which is ~100 ms per finished game (measured
     2026-08-21) and only unfinished games are loaded — so at any realistic table count it is
     free. The work is in reporting a partial failure well: the game is intact up to the
     rejected intent, and a player would probably rather resume there than lose it entirely.
  
  **(3) was done.** `loadGame` no longer looks at the version; `tryResumeSession` replays the
  save and reports the first intent the engine refuses, and `index.ts` resumes or refuses on
  that. A save stamped with a version the server has never run now resumes, provided its moves
  replay — verified against a file hand-stamped `0.4.9-ancient`. A save that genuinely does not
  replay is refused as before, but the log now names the move: *"move 3 of 8
  (localOps.choose) is rejected by the current rules with OPTION_ALREADY_CHOSEN"*.
  
  One thing deliberately NOT done: resuming a partially-replayable game at the last good move.
  The note above suggested a player would rather have that than nothing, and on reflection it
  is worse — the game would silently rewind to a position nobody played to, and the browsers
  holding a later Frame would have no idea. Refusing keeps the file intact, so putting the
  previous version back still recovers the game. Revisit only with a way to tell the table what
  happened.
  

• THE FOUR optionalRules ARE SETTABLE BY NOTHING, AND TWO OF THEM DO NOTHING — resolved in v0.6.0. sisterTrains is deleted: Q9 records that the Second Section card supersedes it, and that card is built. employeeRotation is implemented — the rotation is four lines in advance.ts because the seat/player split (D9) exists precisely for it, so Revenue, hands and the Fedora travel with the player and the district stays with the chair. All three survivors are now settable from the lobby. Original reasoning kept below.

  **Original note:** Split out
  at Jesse's request 2026-08-21, to review on its own rather than as a footnote to the lobby
  item below. `GameConfig.optionalRules` (`state.ts:585-588`) carries `reducedVisibility`,
  `sisterTrains`, `employeeRotation` and `emergencyToolbox`. Neither the solitaire New Game
  dialog nor the lobby exposes any of them, and every construction site in the codebase
  hardcodes all four to `false` (`web/game.ts`, `sim/harness.ts`, `sim/replay.ts`,
  `sim/compare.ts`), so no game has ever been played with one on.
  
  **Check what is real before building a form for it.** Only two are wired:
  
  | rule | status |
  | --- | --- |
  | `reducedVisibility` | **live** — read at `advance.ts:53`, gates on `NIGHT_STAGES` |
  | `emergencyToolbox` | **live** — read at `setup.ts:374`, seeds each player's Red Flags |
  | `sisterTrains` | **nothing reads it.** Declared, defaulted, never consulted — and §9a Q9 records that the Second Section card *supersedes* the Sister Trains optional rule, so this flag is most likely dead rather than unbuilt. Decide whether to implement or delete it |
  | `employeeRotation` | **nothing reads it.** Declared, defaulted, never consulted. Note the seat/player split (Phase 0, D9) was built specifically so this rule *could* exist — the groundwork is there, the rule is not |
  
  So a dialog listing all four would offer two working toggles beside two that silently do
  nothing — the exact failure `checkPlay`'s `NOT_IMPLEMENTED` and `enhancementText`'s
  live/dormant/unbuilt table exist to prevent. Either implement the two dead ones, delete
  them, or label them on screen the way an unbuilt Enhancement already labels itself. Doing
  that is what decides whether this is a UI job or a rules job.
  

• THE LOBBY OFFERS NO GAME PARAMETERS AT ALL — done in v0.6.0. A "Game settings" block on the create form carries the same dials the solitaire dialog has — seed, starting hand, the three revenue rates, days, the combined-Revenue floor, both collision caps, the PvP toggle — plus the three surviving optional rules. Mode and table size set the defaults and every field stays editable, matching the solitaire dialog's own behaviour. Original note below.

  **Original note:** Raised by
  Jesse 2026-08-21 after playing the StartOS build. Creating a multiplayer game asks for a
  display name and a mode, and nothing else — every other dial comes from
  `defaultMultiplayerConfig(mode)` (`web/game.ts`), hardcoded, with no way to change it.
  Solitaire's New Game dialog (`play.html`, `#ng-*`) asks for all of it: seed, starting hand
  (`ng-hand` — three random / six random / three track + three other), the three revenue rates
  (`ng-passenger` / `ng-freight` / `ng-transit`), `days`, `minCombinedRevenue`,
  `maxCollisionsPerDay`, `maxCollisionsTotal` and `pvpCardsAllowed`. Multiplayer should ask for
  the same set. Note that `GameConfig.optionalRules` (reduced visibility, sister trains,
  employee rotation, emergency toolbox) is exposed by NEITHER dialog and is hardcoded false in
  both — worth deciding on separately rather than folding in silently.
  
  **~~The bug this hid~~ — fixed in v0.5.3.** `defaultMultiplayerConfig` defaults to
  `players = 4` and `lobby.ts` called it without the argument, so `minCombinedRevenue` was
  always `collectiveRevenueFloor(4, 5)` = 60 whatever the table's real size — a 2-player game
  played against a floor meant for four (60 rather than 3x2x5 = 30), and missing that floor
  means *everyone loses*. It fell out of the seat-control change: the host now picks the table
  size when creating the game, so the real count reaches `defaultMultiplayerConfig` and the
  ordering problem that caused this (config fixed at CREATE, seat count unknown until START)
  no longer exists. **The form itself is still missing** — that is what this item is now.
  

• Multiplayer proper — Phases 0-4 done (v0.4.0 through v0.5.1), Phases 5-6 to go. The full plan is docs/architecture/multiplayer.md §12. Phase 2 (server core) landed in one pass:

  - `src/sim/frame-delta.ts` — the live per-seat board delta (`deltaFrame`/`applyDelta`), a
    smaller, purpose-written replacement for reusing `replay.ts`'s `compress()` — that function
    interns strings across a whole recorded array, which a live single-frame push has nothing to
    intern against; only its one-step-back "null if unchanged" idea carried over.
  - **Found and fixed a real bug tracing this**: `actionMenu(game, seat)` only used `seat` for the
    `hand` field — everything else came from `currentActor(game)` regardless of who asked, so a
    server computing every connected seat's Menu would have handed the acting player's legal
    moves to a waiting seat, paired with the wrong seat's cards. Fixed in `game.ts` with a guard;
    tested in `multiplayer.test.ts`.
  - The redaction test (§7) is built — `test/redaction.test.ts` — and passed on the first run
    against the existing `snapshot()`, confirming it was already correct, not just apparently so.
  - `src/server/session.ts` — the game session host (pure logic, no sockets, reuses `game.ts`'s
    `Game`/`submit`/`currentActor`/`actionMenu` wholesale rather than re-deriving intent
    application/narration). **Found while building it**: `submit()` derives the acting player from
    `currentActor(game)` itself and does not check who is calling it — safe for `LocalSession`
    (one possible caller) but not for a server, so the session host verifies `seat ===
    currentActor(game)` itself before ever calling `submit`, rejecting with `NOT_YOUR_TURN`
    otherwise. Idempotent resend (a repeat `seq`) and the illegal-intent path (checked via
    `check()` directly, so a rejection never pollutes the shared narration log with "not allowed"
    text meant only for the submitter) are both handled here too.
  - `src/server/http.ts` / `src/server/index.ts` — plain `node:http`, no framework (confirmed
    nothing to reuse and nothing else warranted — zero runtime dependencies anywhere else in the
    project). `POST /api/game`, `GET /api/stream` (SSE, per-seat, with a 20s heartbeat and an
    `id:` line per push), `POST /api/intent`, and static serving of `dist/` so the server can be
    same-origin with itself (D16). No `gameId`/multi-game concept yet — one game per process,
    matching "Phase 2 has no lobby."
  - `src/web/session.ts` gained `createRemoteSession`; `main.ts`'s `start()` switches on `?seat=`
    presence (D4 — one bundle, unchanged). Every `LocalSession`-only call site in `main.ts`
    (`seed()`, `save()`, `undo()`, the New Game dialog) now goes through an `isLocal()` type guard
    rather than assuming, since `session` can now be either.
  - **Found and fixed a real infrastructure bug**: adding `test/server/` broke `npm test`'s glob.
    `"test": "node --test test/**/*.test.ts"` relied on bash's non-globstar behaviour of passing
    the *literal, unexpanded* pattern through to Node (which then globs it correctly itself) —
    that only happens when the pattern matches *no* files at the shell level. The moment a
    subdirectory existed, bash expanded it to just that one file, and `npm test` silently ran only
    the new suite. Fixed by listing both depths explicitly:
    `"test": "node --test test/*.test.ts test/**/*.test.ts"`.
  - Verified two ways: `test/server/session.test.ts` exercises the session host directly (no
    sockets); a live end-to-end curl smoke test (server started, a 2-player game created, two SSE
    streams opened, an intent rejected from the non-acting seat, accepted from the acting seat and
    broadcast to both, a resent `seq` producing no second push, and the board correctly nulled on
    the second push) — see the session transcript. **Not verified**: an actual browser — no
    browser binary exists in this environment, so `RemoteSession`'s DOM-facing code
    (`EventSource`/`fetch` wiring) compiled and typechecks but was not clicked through visually.
  
  **Phase 3 (persistence/resumption) done, same session, 2026-08-21.** Per §12 steps 14-16 and
  `lobby-and-sessions.md` §5-6 (unusually concrete — the exact storage shape was specified, not
  designed here):
  
  - `src/server/persistence.ts` — `game.json` (`{engineVersion, seed, config, playerNames,
    history, status, createdAt}`) and `turn-timings.json`, both atomic-rewrite-then-rename, no
    `gameId`/index yet (one game per process, same deferral as Phase 2's `gameId`).
  - `game.ts` gained `fromMultiplayerSave` — `fromSave`'s multi-player sibling, built on
    `newMultiplayerGame`. **Found while testing it**: `fromSave`'s replay loop calls
    `record(game, result.events)` without the `actor` argument `submit()` itself always passes,
    so every replayed line loses its "Player X" attribution — invisible for solitaire (nothing
    ever compares a `fromSave` replay against a live-played log; `undo`'s rebuilt game is itself
    `fromSave`-built, so the one test that compares logs only ever compares two unattributed
    replays against each other) but immediately visible for multiplayer, where anonymous "Chose
    to..." lines are unreadable the moment there is more than one seat. Fixed in the new function;
    **`fromSave` itself still has the gap** — not touched here, since it is used far more widely
    (undo, save/restore, the replay viewer) and deserves its own careful pass rather than a
    touch-in-passing. Worth its own TODO item if picked up.
  - `session.ts` gained `exportSave()`, `resumeSession()`, and turn-timing tracking — a `TurnTiming`
    span (player, phase, day, stage, start/end wall-clock) closes and reopens whenever the acting
    player, phase, Day or Stage changes; recorded entirely in the session host, never touching the
    engine (which must stay clock-free and deterministic) and never stored inside `history` (a
    replay must reproduce a game from decisions alone). No reporting/aggregation/UI on this data
    yet — §5 calls that "optional... if unobtrusive," and the Phase 3 deliverable is the data
    being recorded, not a view of it.
  - `index.ts` loads `game.json` on boot before starting the HTTP listener: version match →
    `resumeSession`, replayed straight through; mismatch → refused explicitly and loudly (the
    file is left untouched, so rolling the running version back recovers it), server starts with
    no active game rather than replaying under the wrong rules.
  - Verified live, matching this phase's own "done when": server started against a fresh data
    directory, a 2-player game created, intents submitted from both seats, **the server process
    killed and restarted**, both `?seat=` streams reconnected and picked up exactly where they
    left off — same Day/Stage/phase, correct whose-turn-it-is, correct narration attribution.
    Separately confirmed the version-mismatch path: hand-edited `engineVersion` to a bogus value,
    restarted, server logged the refusal and started with no active game (confirmed via `POST
    /api/game` succeeding rather than 409ing).
  
  **Phase 4 (lobby, sessions, reconnection) done, 2026-08-21 — v0.5.1.** Per §12 steps 17-20 and
  `lobby-and-sessions.md` in full:
  
  - `src/server/lobby.ts` — pure logic, no sockets, no filesystem, same split `session.ts`
    already draws. `createLobby`/`joinLobby`/`setBotSeat`/`reassignHost`/`startLobby`, a
    speakable game code (`RAIL-4471` style, from a small railroad-word list rather than a
    dictionary — read aloud across a table, not typed from memory), and the 2-4 player cap
    (`playerCountAllowed`) — see the doc-fix note below.
  - **The server now holds more than one game.** `persistence.ts` gained one directory per
    `gameId` (`games/<gameId>/`) plus a top-level `index.json` naming every game, so `index.ts`
    can resume all of them on boot rather than the one `game.json` Phase 3 assumed.
    `writeGame`/`loadGame`/`appendTiming` needed no signature change — they already took a
    directory directly.
  - **Session tokens replace `?seat=&secret=` on the running-game routes.**
    `lobby-and-sessions.md` §1: the token alone proves identity, so `/api/stream` and
    `/api/intent` now read `?token=` and the join secret's job ends at the lobby door
    (`/api/lobby/create`/`/api/lobby/join`). `web/session.ts`'s `createRemoteSession` takes
    `(token, seat)` — `seat` still passed in rather than learned from a push, since it has to
    answer before any push necessarily arrives, and the caller already has it from the
    join/create/start response.
  - **Bots fill empty seats at `Lobby.Start` only (D8)**, never mid-game. `session.ts` gained
    `driveBots()`: after any accepted intent (and once at construction, for a resume that lands
    exactly on a bot's turn), it plays `developerBot` forward through every consecutive bot seat
    before the push goes out — reuses `legalActions`/`developerBot` wholesale, no new bot logic.
    `SavedGame` gained `botSeats: PlayerIndex[]` so a bot seat survives a restart.
  - **Host rights pass to the earliest-joined remaining player** if the host's LOBBY connection
    closes before start (`lobby-and-sessions.md` §2) — tracked via `Lobby.joinOrder`, a token
    list rather than seat order, since a bot-filled seat has no join time of its own.
  - **Disconnect/reconnect** (§5): `Push` gained an optional `presence` field — connection news
    about ANOTHER seat, built entirely by `http.ts` (which owns the connection table) and never
    routed through `session.ts` or the engine, since a disconnect is transport news about a
    connection, not a `GameEvent`. The page shows a banner naming who has dropped
    (`renderPresence`, `main.ts`) and clears it the moment they reconnect. Reconnect itself needed
    no new engine-side work: `session.connect(seat)` already sent a full un-delta'd `Frame`.
  - **The client lobby** (`src/web/lobby.ts`, wired from `main.ts`'s `start()`): create-or-join
    forms, a live seating screen (host-only bot toggles and Start button, updated over a new
    `/api/lobby/stream` SSE), and `localStorage` in place of `?seat=` for "was I already in a
    game" — found on load, reconnects straight to `createRemoteSession` and skips the lobby
    entirely. A `Multiplayer` button beside `New game` is the entry point; the New Game dialog
    itself is untouched, still solitaire-only, its old "needs a server" note repointed at the
    new button.
  - **Found and fixed while running the live smoke test, not by typechecking:** `/api/intent`
    read its token from the JSON body, but `web/session.ts`'s `submit()` — unchanged from Phase
    2 — sends it in the query string, same as `/api/stream`. Every request failed `no such
    game`. Both sides independently typecheck fine (an HTTP body is `unknown` on the wire), which
    is exactly why the curl-level smoke test exists rather than stopping at `tsc --noEmit`.
  - **Doc fix:** `multiplayer.md`'s D18 said "player cap 6", citing `lobby-and-sessions.md` §2 —
    which actually specifies 2-4 and gives the reasoning (what `test/multiplayer.test.ts` exercises).
    The two had drifted apart; "6" was never implemented or tested anywhere. D18 now says 2-4.
  - Verified: `test/server/lobby.test.ts` (pure logic — creating, joining, capacity, bot seats,
    host transfer, starting) plus new coverage in `session.test.ts` (bot-driving, including two
    bots in one game) and `web.test.ts`. A live smoke test through `curl`: create a lobby, join a
    second player, start, submit intents from both (including the wrong-actor rejection and an
    idempotent resend), reconnect after a real server kill-and-restart, a bot-filled coop lobby
    starting and never stalling on the bot's seat, and a disconnect/reconnect presence notice
    observed on an open stream. **Not verified: an actual browser** walking through the lobby
    screens — none is available in this environment, the same limitation Phase 2's `RemoteSession`
    shipped under.
  

• THE BOARD STILL DOES NOT SAY WHICH WAY A HEAVY GRADE CLIMBS. — done 2026-08-30 in v0.7.9 (cf018b4). A brown wedge in the card's lower right rising toward the climb, with an arrow lying along its slope. It is Gitea#18 that made this drawable: east is always to the right on a single-row map, so a wedge can be read without a compass — under the wrapped layout the same wedge would have pointed a different way at every seat. Orientation was measured rather than assumed (400 seeds x 4 player counts, 535 grades, 0 without one, 276 east / 259 west). Original note below.

  **The board did 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 did 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.

• INVESTIGATE: three explicit display options for the Office map — always hidden, always on, auto-hide. — done 2026-08-30 in v0.7.9 (Next #16). A segmented control, one button per mode, aria-pressed on the lit one. The investigation below was right about the cause and right about the fix; what it did not foresee is the testing trap.

  **THE BUTTONS ARE ADDRESSED BY ID (`#dm-auto` / `#dm-open` / `#dm-closed`), NOT QUERIED OFF
  THE CONTAINER — and that is worth knowing before the next control like this.** The web suite
  runs against a stub DOM whose `querySelectorAll` reads the element's own `innerHTML`, so it
  only ever sees markup THE PAGE WROTE. This control lives in `play.html`, so a child query
  returns nothing there, the wiring loop does nothing, and the whole control ships green and
  completely unexercised. Addressing by id also puts every button under the "asks the page for
  no element its page lacks" check.
  
  **A second gap closed on the way:** none of the five element factories in `test/web.test.ts`
  had `setAttribute`, so the first render threw. Any control reporting its state through ARIA
  was untestable until they got an attribute bag. Original note below.
  
  **All three modes already exist.** `districtMode` is `'auto' | 'open' | 'closed'`, persisted to
  `localStorage` with the sound and zoom settings (`main.ts`). Nothing needs adding to the model.
  
  **What is wrong is that the button is a CYCLE, and it cannot reach every state.** The handler is
  `districtMode = districtMode === 'auto' ? (open ? 'closed' : 'open') : 'auto'` — so from `auto`
  you land on whichever pin is the OPPOSITE of what auto is doing right now, which depends on the
  phase, and every second press goes back to `auto`. You can never get from `open` to `closed`
  without passing through `auto`, and which of the two you can reach at all changes as the game
  moves between phases. That is why it does not feel like a setting.
  
  **Likely three buttons or a three-way segmented control**, one per mode, showing which is
  current — the label work is already done and is worth keeping: it says what pressing it DOES
  ("always showing — click for auto-hide") rather than what the panel is currently doing, which
  was a deliberate fix and should survive whatever replaces the cycle.
  

• INVESTIGATE: the same three options for the Division map, where auto means something different. — DECLINED 2026-08-30. Jesse: "we are not going to hide the division map anymore."

  **Gitea#18 answered this one too, and it was missed.** It was raised 2026-08-22, four days
  before that issue closed, and it is really a sixth member of the drawing pass settled below —
  it just reads as a control question rather than a drawing one, so it stayed open when the other
  five were closed. **The reason to fold the map away was that it grew.** A horseshoe of three or
  a square of four was tall enough to push the board off the screen, which is what made "hide it
  while I switch" worth a control. A single row is `boardH = PAD * 2 + CH + 30` — **150px, fixed,
  at every seat count** (`board-svg.ts`) — and a 150px strip is not worth a control, three states
  and a persisted preference. It also scrolls and zooms rather than reflowing, so it costs the
  same at one seat as at four.
  
  **What it would have cost, recorded because the design was worked out before it was declined:**
  the Division's `auto` is not a third MODE at all. It is two sticky pins plus a transient — "hide
  for now", expiring when `f.phaseKey` moves — so it needed `divisionMode: 'open' | 'closed'`
  persisted beside `districtMode`, plus a `hiddenDuringPhase` that clears itself. Deliberately not
  shared with `districtMode`, whose `auto` is a standing rule keyed on `FOCUS_PHASES`. **And a
  folded Division needed a summary line written for it** (Jesse's call, 2026-08-30, before the
  decline): `.folded` on `#district` hides `#grid` and `.districtrule` but keeps
  `#districtsummary`, so the house idiom is that a folded panel still says something, and the
  Division has no such line today.
  
  **If it ever comes back, this is what to check first:** whether the map has started growing
  again. That, not the control, is the thing that would justify it.
  
  Original note below.
  
  **Today it cannot be hidden at all.** `#division` is a plain `<div>` in an unnamed `<section>`
  in `play.html` with no toggle and no fold rule — `#district` has `.folded` styling and a button,
  the Division has neither.
  
  **`auto` here is not the Office's `auto`, and that is the point.** The Office folds by PHASE
  (`FOCUS_PHASES` — open during Local Operations and Cargo, folded otherwise). Jesse's Division
  rule is "hide it NOW, and bring it back at the end of this phase": you fold the map away to get
  room while switching or working cargo, and it returns of its own accord when you are done. So
  it is a one-shot with an expiry, not a standing rule — the state has to remember WHICH phase it
  was hidden during, and clear itself when `f.phaseKey` moves off that one. Different enough from
  `districtMode` that sharing an implementation with it would probably be a mistake.
  
  **Hidden and shown stay put** until pressed again, exactly as the Office's pins do.
  

• INVESTIGATE: the game's settings belong in a card, not along the top line. — done 2026-08-30 in v0.7.9 (Next #28). A This Game card at the foot of the right-hand column, folded by default and persisted with the other display preferences.

  **The open questions below were answered by Jesse, 2026-08-30.** Which of the six stay on the
  top line: Revenue, the objective and the game code — *plus the collision counts*, which were
  not on it at all. Whether the card folds: yes, like `#district`, with a summary line that
  survives folding. Where the collision counts belong: the top line, because a limit is a setting
  agreed to once and "2 of 3 today" is a number that changes how you play the next Stage.
  
  **Cheaper than this entry estimated.** It says the renderer already exists; what it misses is
  that `configFromFrame` also already exists, is exported, and `main.ts` already called it three
  times — so `rulesListHtml(configFromFrame(f), f.players.length, f.days)` needed no refactor
  whatsoever. The card is markup plus one call.
  
  **The collision counts were the real find.** The Frame has carried `collisionsToday` and
  `collisionsTotal` since v0.7.0 and NOTHING ON THE BOARD DREW THEM, so the one victory condition
  that ends a game early ran invisibly — the same shape as #43's `actingPlayer`, and the second
  time in one release that the Frame had the answer and the view never asked. Original note
  below.
  
  **The original note.** Raised by Jesse
  2026-08-23, playing the v0.7.0 build: "the game-specific information in the very top line should
  probably be a card like Facilities, timetable or blocked. Off on the side, we can give complete
  information about all the game options and not take up valuable real estate at the top of the
  screen."
  
  **What the top line carries today**, in order: Revenue, the objective (`#objective`), the seed
  or seat (`#seed` — the seed in solitaire, `Seat 2` in a multiplayer game), the game code
  (`#gamecode`, added 2026-08-23), the game type (`#gametype`, e.g. "Custom — scored as
  Competitive"), and an abbreviation of the house rules (`#houserules`, "3 cards · 1/1/0"). The
  last four were each added because the information was missing entirely, and the header is now
  carrying them because it was the only place they had ever been put.
  
  **What a card could say that the header cannot.** Jesse's list, plus what the Frame already
  carries: seed and seat, the game code, the game type and what it is scored as, the opening hand,
  all three revenue rates, the Day count, the combined-Revenue floor, BOTH collision limits (with
  the running counts, which the Frame has as `collisionsToday`/`collisionsTotal`), where an Extra
  may start, and every optional rule that is on. **Nothing new has to be sent** — `Frame` gained
  `mode` and `optionalRules` in v0.7.0, and everything else on that list was already on it. The
  read-only renderer already exists too: `rulesListHtml` (`settings-form.ts`) draws exactly this
  list for the join preview and the seating screen, so the card is largely a matter of calling it.
  
  **His own framing of the value**, worth keeping because it names when it is read: "To go, 'Oh
  wait, what did we set that to?' They should be able to look that up, but it does not need to be
  at the top every moment because it is not something that they're likely to need all the time."
  
  **The open questions.** Which of the six stay on the top line — Revenue and the objective are
  glanced at constantly and clearly belong there, the seed and the code almost never are. Whether
  the card folds like `#district` does or is always open. Whether the collision counts belong in
  it or beside the objective, since they are a live score rather than a setting. And it inherits
  the one principle that outlived the wrapped-map items (settled below): **SHARED in the middle,
  YOURS on the right.** A settings card is not yours — it is the table's — so if the common board
  of Gitea#20 ever comes back onto this page, the settings card belongs beside it rather than in
  the right-hand column.
  

• INVESTIGATE: the Fedora belongs at the right-hand end of the phase row. — done 2026-08-30 in v0.7.9 (Next #29). Jesse's fallback was the one taken: .tc-super moved to the end of the row, after the pills, and wrapping under them rather than squeezing the chips on a narrow screen.

  **The appealing version was tried and rejected.** Putting the name INSIDE the Supervisor Shift
  pill reads as "this phase belongs to that player", which is not what the Fedora means — the
  office holds the clearance ruling and starts every round, in every phase. Worth knowing if
  anyone proposes it again.
  
  **The row was looked at as a whole**, per the note left here at the time, because item #15's
  "most recent action" line still wants space in the same strip. Nothing was taken; the row has
  room for one more block at the widths tested.
  

• Five drawing items against the wrapped Division map — ALL ANSWERED BY Gitea#18 (closed 2026-08-26), which replaced the layout rather than fixing the drawing. They were: the map drawing no track geometry; vertical track art on a card laid down a side lane; the inter-row connector routed round the outside with angled corners; the Division Point captions overflowing and the buffer stops pointing the wrong way once the route wrapped; and seating the viewer at the bottom with the table wrapped around them. All five were the same pass, and Jesse's instruction was to do them together — which is what made replacing the layout the cheaper answer than drawing it five ways.

  **What the single row settled.** The route is one line, west on the left and east on the right,
  at every seat count (`board-svg.ts`). Both buffer stops face outward, so the two "wrong way"
  reports cannot recur. The captions sit under their own Division Point rather than hanging off
  the ends, so nothing clips. And **east is always to the right** — the report that decided it,
  and the reason the horseshoe lost to a row that is 1,580px wide at four players against 842.
  It is also what made the Heavy Grade wedge drawable at all (item above).
  
  **The Division map no longer draws office-area detail**, so there is no Running Track on it to
  draw turnout geometry for. That belongs to the Office map, which reads `cell.links` and already
  draws it properly.
  
  **Two things survive their items and are NOT closed with them:**
  
  - **"SHARED in the middle, YOURS on the right"** — Jesse's principle from the dead-centre item,
    and worth keeping as the rule that decides where anything new on the screen belongs, even
    though a row has no centre to fill. It is now **Gitea#20**'s territory: the common board is
    the shared thing, and it got its own display rather than a hole in the map.
  - **Why seating the viewer at the bottom was refused**, since it will be proposed again: the
    Division is a LINE, and the open gap between the two buffer stops is the only thing that says
    so. Rotate the map to seat a viewer and that gap moves with them — sometimes behind them,
    out of the eye's path, which is exactly where the one feature saying "this is not a circle"
    must not go. Jesse called it definitively superseded, 2026-08-26.
  

• INVESTIGATE: history newest-at-the-top, and timestamps on it. — the ORDER is done 2026-08-30 in v0.7.9 (Next #23); the timestamps half is still open and still #14.

  Flat reversal, `scrollTop = 0`, the start marker last. Jesse ruled directly on the phase
  headings, which the note below calls the thing that is not free: they now trail the lines they
  announce, and "stage changes will be beneath (prior to / older than) the following events.
  That is OK." Reading down the panel is reading backwards in time. Grouping by phase and
  reversing the groups was offered and declined as more machinery than the complaint needs.
  
  **The `slice(-60)` cap is untouched and is the other half of this note** — scrollback still
  ends at sixty lines whichever way the panel runs. Worth reopening on its own if anyone ever
  tries to scroll back and cannot.
  
  `replays.ts` deliberately keeps its oldest-first log: it is paired with a frame stepper, where
  "what just happened" is the step you have this moment clicked, so newest-first would fight the
  stepping. Original note below.
  
  **Raised by Jesse 2026-08-22.**
  
  **Timestamps** are item 14 above — the same question, and it should be answered once. Whether
  the history DISPLAYS a time is downstream of whether one is recorded at all, and of the sidecar
  constraint written up there.
  
  **Order.** Today `main.ts` renders `session.lines().slice(-60)` oldest-first and then sets
  `log.scrollTop = log.scrollHeight`, so the newest line is at the bottom and the panel scrolls
  itself down to it. Reversing is nearly free — reverse the slice, drop the auto-scroll — and it
  does what Jesse wants: a glance at the top is always the most recent thing.
  
  **Two things that are not free.** The cap is `slice(-60)`, so "scroll back through the history"
  reaches sixty lines and stops however it is ordered; a real scrollback means raising or removing
  that cap and deciding what the panel does with a thousand lines. And the log carries PHASE
  HEADINGS (`t-phase`) that read forwards — a heading introduces the lines under it — so reversing
  the list puts each heading below the lines it announces. That has to be handled or the panel
  reads as nonsense at exactly the boundaries it exists to mark.
  
  **Worth deciding together with item 15**, which proposes pulling the single most recent line out
  of this panel entirely. If that lands, the argument for reversing the panel is weaker.
  

• 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, and it was ranked THIRD and predicted not to matter because the Division Yard never runs dry. The aggregate was the wrong measure; having the right commodity at the right moment is what counts.

• Make Enhancements reachable at all. The bot never laid a straight on the Running Track (0.00 in 100 games) because two-arc run-arounds do not need one — so 13 of the 18 Enhancement cards had nowhere to go, including Interlocking, the only cure for the only penalty in the game (no free A/D track, 27% of gross). One scored straight fixed it: enhancements placed 0.64 → 3.01, collision cost 2.70 → 1.91, worst game −47 → −24. Revenue +0.70 ± 0.74 paired over 400 seeds — real but not significant alone; the variance reduction is the clearer win.

• Stop the bot discarding its own freight. canStockProductively did not check the Division Yard while the engine's stockOutbound does, so Freight Agent was chosen when nothing could be stocked and the follow-through fell through to an unjam that threw a waiting load out of the green box — 3.10 a game against 2.71 started. Now 0.00. Revenue 6.0 → 6.5, wins 5 → 8 in 100. Also confirmed routing was never the problem: 0% of drops land on a facility that does not want the car.

• Why switching work did not become Revenue. Answered: it was the freight the crew shuffled, not the shuffling. The chain never leaked — 95% of started loads finished — it was barely entered, because a load needs a matching empty car spotted and half the industries never asked for one. Three fixes later (sidings, facility placement, car selection) revenue is 3.2 → 6.0 and freight 26% → 37% of gross.

• Fix car selection. Three of six industries were invisible to wantedCars — a hand-written industry→car map naming two industries that do not exist and omitting three that do — so tank cars were dropped 0 times in 100 games. Derived from INDUSTRY_PROFILES now, and the second commodity of the two-commodity industries is reachable. Revenue 5.0 → 6.0, freight share 25% → 37%, wins 1 → 5 in 100.

• Put the industries on the run-around. Facility placement was unscored — the first legal square — so 0.00 facilities a game sat on a loop; now 1.08. The instructive part was the second bug: scoring facilities onto the siding row dropped run-arounds 91→36, because the anchor test asked a card's KIND rather than its PORTS and an industry in the line read as a dead end. Revenue 4.1 → 5.0. Freight did not follow, which is the item above.

• Make the bot build sidings that are sidings. 0 run-arounds in 100 games → 91. Three bugs, all scoring on local shape without checking it reached anything; the decisive one was that bestTrackLay never declined a piece, so it spent the track supply on whatever was legal.

• Teach the bot what a siding is for. Nose coupling (§A.3) implemented, so approach direction decides which car is droppable; the bot runs around rather than setting out, when the drop can follow. Switching activity transformed, revenue unchanged.

• Curve geometry. Curves were topologically identical duplicates of turnouts, and nothing reached north, so a district could only be a vertical column. Now two-port rotatable arcs.

• Q10 — when track may be laid. During the "draw a card" option, one piece a turn. Track was a card when §6.2 was written; a 26-piece supply has no hand to bound it.

• Q11 — which way a Heavy Grade climbs. Answered from the card: it prints "(Up)" and "Player sets orientation", so it is a property of the placed card, not a compass constant.

• §6.2's reshuffle. Implemented, and the deckReshuffled event it had already declared and narrated — but never emitted or reduced — is now real. Not yet reached in play: solitaire Campaign ends with 168.8 of 243 in the deck and four-player Campaign with 86.6, and no run of any length has emptied it. It is a safety net rather than a live mechanic today, which is worth knowing before tuning draw rates.

• Q12 — Whistle Post lock-in. Players always start at a Whistle Post; office density doubled instead.

• CLEARING AN INBOUND BOX MINTS A CAR — measured at 1.29 a game against a supply of 80. Re-audited in v0.4.3: rolling stock is EXACTLY CONSERVED, 100 games out of 100, range 0..0. The old audit's premise was right — the two directions were not symmetrical — but the asymmetry has since been closed from the other end. unloadBegan and passengersDetrained now take their replacement empty OUT of the Division Yard rather than conjuring it, so a load is a car that moved rather than a car that appeared: one leaves the yard, one arrives in Classification. Jesse's description of the tabletop procedure confirms this is the intended model — the token you push along the MEN|AT|WORK sign IS the car, fetched from the yard by the Freight Agent and swapped onto the industry track at the end. Both conjuring fallbacks now throw rather than minting, so the leak cannot silently return; neither fired across the suite or 100 audited games. ROLLING_STOCK_SUPPLY is therefore unblocked — it was waiting on this and can now be tuned in the rebalance pass. (The first audit's own arithmetic was off in the same way mine was on the first attempt: cars set out on a card, in card.standing, are easy to leave out of the count and make a conserved game look like a leaking one.)

• state = fold(events) was not true, and the docs said it was. Settled: the INTENTS are canonical. Measured before deciding — advance.ts never calls reduce, so 14 of the 46 event types are never reduced: the clock, and the entire Mainline phase, which is every train movement in the game. Folding the log rebuilds a district and not a railroad. Jesse's call, and the cheap one: the plan never needed fold — persistence is { engineVersion, seed, config, history } (multiplayer.md §10) and the wire carries Frames, not events (D2/D3), so reconnection is a fresh Frame rather than an event tail. Making the phase driver reduce would have been a rewrite of the most rule-dense code in the project to buy something nothing uses. Corrected in the README, four architecture documents and six source comments; test/events.test.ts pins the unreduced set so that closing the gap later is a deliberate act, and asserts the property that does hold. If you ever do make the phase driver reduce, that test fails and tells you which docs now understate the engine.

• Measure with error bars from now on. Built: node src/sim/compare.ts 1600 <tweak>=<n> runs the current bot and one variant over the same deals and reports the paired difference. Pairing drops σ from ~9 on the level to 5.3 on the difference, so 1600 seeds gives ±0.13 in about 1m45s — the noise floor is now ~±0.15 rather than ±1.0. Threshold to keep a heuristic is t ≥ 3, and the report prints the better/worse/identical split beside the mean, because a mean carried by a skewed tail is a different claim from broad improvement.

• Confirm the Classification Yard rule against the source. Confirmed, and the guess was wrong. The rule is: used Rolling Stock to the Classification Yard, used engines and cabooses straight back to the Division Yard, and the Classification Yard empties ONLY when the Division Yard is bare — then all at once. The Day-boundary version I had invented was far more generous and worth +2.42 revenue a game the game does not actually grant. Corrected; revenue 9.67 → 7.25.

• Enhancements are placed but mostly do nothing. Measured: forbidding every Enhancement except Interlocking is worth -0.01 ± 0.41 (t = -0.04) over 400 paired seeds. They neither pay nor cost. Left alone. Unlocking the Running Track straight put nine kinds on the board (telegraph 0.73, waterColumn 0.57 …), but only Interlocking has a measured effect. The Telegraph/Telephone/Radio chain adds to the other train's number when dispatching facing trains, which may be worth nothing in solitaire; Water Column removes a Watertower; Facing Point Locks prevents Derail, which is multiplayer-only. Worth measuring what each is actually worth before the bot spends actions on them.

• stats.ts undercounts freight. Fixed: both halves counted, freight share 39% → 49%. Worth revisiting the industry density decision in Play Balance, which was taken on the old number.

• LEFT AND RIGHT ARE ON THE WRONG DIAGONAL — for turnouts and for curves, the same way. The engine's left turnout is {stem:'w', through:'e', diverge:'s'}: a train entering at the points from the west heads east and the diverging route leaves to its right. The engine's left curve is arc sw, which turns an eastbound train right as well. One consistent sign error in the hand↔diagonal mapping, and it mislabels every track card a player ever holds. The artwork is right and does not change — board-svg.ts:462 draws rails from connectionsFor() geometry alone, so only words are wrong. Decision: flip the hand value on the TRACK_CARDS rows AND the two variant tables in the same commit, so the code keeps speaking left/right like the physical supply and now means it. Keep the row order in TRACK_CARDS untouched: setup.ts:95 builds the deck by iterating that array, so flipping only the labels leaves pre-shuffle slot 32 holding a ne_sw curve before and after, and variantsFor(…)[0] still 'sw' — same seed, same board, and every published replay still plays. Also: track.ts:126 arc fallback, track.ts:191-212 doc block, view.ts:1041 and view.ts:1208 diagonal phrases, bot.ts:837-852 arcInHand, seven test files, and the prose plus ~14 data-tip="Turnout · left" strings in docs/design/track-geometry.html — which has no generator and must be hand-edited. Verify by fingerprinting a fixed seed's board before and after: identical geometry, different words.

• A turnout should be playable as an UPGRADE, on top of a card already down. On a straight, or on a curve whose arc matches the turnout's diverging leg. Nothing like this exists — the only "upgrade" in the game is the Office tier change, which is explicitly not a card swap (apply.ts:1290), and canPlaceAt hard-stops at if (existing && !isMovableSign) return false (track.ts:518). Decision: cars AND enhancements both block it — standing.length > 0 → UPGRADE_OCCUPIED, enhancements.length > 0 → UPGRADE_ENHANCED, so an Interlocked straight stays a straight. Express the curve rule on arcs, not hands, so it survives the flip above. No extra connection requirement: all four turnout orientations are port supersets of a straight and of any same-arc curve, so an upgrade can never sever an existing join, and the new leg is allowed to dangle — that is what it is for. The lifted card leaves play, which is already how board cards behave (apply.ts:1249 salvages only cards that were not placed). Reuse card.play with a placement on an occupied square; legal.ts:252 must offer those squares for turnouts, and the attachments set at legal.ts:137 is already exactly that list. The other half of this report needs no work: a turnout carries the through route (carriesThroughTrack), so it is already legal at a Limit, along the Running Track and on Secondary Track. Confirmed, not re-investigated.

• A Depot shows three MEN AT WORK boxes it can never work. Offices are Passenger Facilities — no freight — but setup.ts:176 gives every one a three-slot menAtWork array, and the three renderers loop it with no guard while the green and red rows beside them are guarded (board-svg.ts:548, panels.ts:218, replay.ts:522). Decision: all three tiers — Depot, Station and Terminal are all Passenger Facilities and all get laborers: 0, so the boxes are inert on every one. Fix it in the model, not the renderers, so it cannot reappear in a fourth place: make menAtWork nullable and null for passenger facilities, which is what the "Freight only" comment at state.ts:153 has claimed all along. Freight handling must be neither allowed nor displayed there. Two more artifacts of the same "an Office is a Facility" modelling go with it: board-svg.ts:562 calls a Depot "SHIPS + RECEIVES", an industry's flow word, and board-svg.ts:599's Math.max(1, trackCap) draws it a siding slot although its industry track has length 0. FacilityView needs a kind field; it has no freight/passenger flag today.

• Two Ice Houses can be built in one district. Industries already ban duplicates per Office Area — isLockedOut (apply.ts:1774) covers the Mine Tipple half of the report — but Ice House is a Modifier, not an industry, and checkPlay's modifier branch (apply.ts:662) has no duplicate check at all. Decision: extend the ban to modifier kinds, leave enhancements alone. An Interlocking is a plant at one junction, so a second on another straight is a different installation, and the Telegraph → Telephone → Radio chain is already gated per card. Expect modifiers with copies > 1 to go partly dead in solitaire, where there is one Office Area — correct, since the spare copies exist for other players' districts.

• A drawn card lands at the far end of the hand with nothing to mark it. apply.ts:1219 pushes, the hand renders in state order (game.ts:355), and with flex-wrap the newest card is exactly where the eye is least likely to be. Decision: reverse in the display layer, not the engine — view.ts:898-899 (reversing hand and handWhat identically, or they desync) covers the replay viewer and the standalone replay together, and game.ts:355 covers the play page. Keeping hand.push means the bot's option-iteration order does not move, so every revenue figure in this file stays comparable; an engine unshift would invalidate the lot. The marker is play-page only — a Frame carries card names, not ids, so a replay cannot say which card arrived that step. Follow the game.scheduled precedent for a justDrawn field, but make the badge persist rather than flash: it says which card is new, not something just happened. A static ::before badge as .handcard.target does it (panels.ts:262), no keyframe — the innerHTML rebuild would restart an animation on every render. Clear it in renderUndo() and after fromSave(), or a fresh page load badges last session's draw.

• The inbound boxes render green in the side panel. Green is outbound and red is inbound everywhere the colour carries direction — board-svg.ts:528-545, replay.ts:452, rules §9.1 — except panels.ts, whose shared boxes() helper (panels.ts:162) emits class f for every filled box regardless of direction, so the red row at panels.ts:222 comes out green. The board SVG on the same page draws it correctly, which makes the panel actively contradict the board. Give boxes() its class from the caller as replay.ts already does, add .box.r in board-svg.ts:820's palette, and take the siding off green in both panels and replay.ts:525.

• An Interlocking on the board is a bare label, and it does nothing. The enhancement text is drawn with no tooltip (board-svg.ts:651) while the copy already exists as data in ENHANCEMENT_CARDS (content.ts:589). Done: every card prints its effect, and the one nothing reads says so. CORRECTED — the first pass had five of the ten statuses wrong. It claimed only Telegraph/Telephone/Radio were live, because the survey grepped for four helper function names and read "no match" as "no implementation". In fact seven are live: those three plus Interlocking (advance.ts:770), Yard Office (advance.ts:750), Small Yard (apply.ts:405) and ABS Signals (advance.ts:599, stored on the Mainline node). Facing Point Locks and Water Column are wired but dormant in solitaire; Overpass alone has no code path at all. The shipped tooltip briefly told players four working cards did nothing, which is worse than the bare label it replaced — enhancements.test.ts had passing tests for all four the whole time.

• A modifier's grant can land on a direction its host cannot use, and nothing says so — corrected in v0.4.7, and the correction was itself half wrong. (v0.4.9e: the ORIGINAL verdict was right about the Grocer's. A Grocer's Warehouse is inbound-only — gameplay testing said so and Jesse confirmed it — so the Ice House's outbound grant beside one is genuinely dead, the way the Truck Dock's inbound grant beside the outbound-only Packing Sheds is. What survives from v0.4.7 is the machinery and the decision behind it: an industry's printed flow is absolute, the grant is dropped rather than the direction opened, and suppressedGrants says so on the page. What does not survive is opening the two facilities up. Original v0.4.7 note follows.)* The earlier "no bug" verdict below was wrong. The reasoning had been that a Grocer's Warehouse is inbound-only. It is not — the card reference says "Both" — so the grant was being dropped on a direction the facility should have had. Reported again in play as "grocer's warehouse didn't get extra outbound slot for truck dock". The suppression machinery itself was right and is kept: it still fires for a passenger Modifier beside a Whistle Post, which is not a Passenger Facility — and v0.4.7 gives that grant BACK when the Office is upgraded, which it never used to. Original note follows. Reported as "Ice House added the laborer but not the outbound slot" — checked, and there is no bug: Ice House prints +1 outbound, a Grocer's Warehouse is flow: 'inbound' so allows.outbound is false, and the capacity was raised on a direction that can never render or be stocked. The laborer arrived because laborers have no direction gate. Decision: an industry's printed flow is absolute — drop the grant on hosts that cannot use it rather than opening the direction up. So applyModifier (apply.ts:1788) gates each capacity grant on allows and grows industryTrack.length only by what was actually applied. Audit all 17 profiles for the same trap: iceHouse, truckDock and forklifts all print addOut: 1 and list grocersWarehouse; waitingArea, restaurant and hotel print addOut: 1 for hosts: ['office'], which includes a Whistle Post. Then say it in both directions — the hand tooltip naming which printed hosts cannot use which half (computable from the profiles, no host on the board needed), and the panel's "prints N, Modifiers add M" line showing a suppressed grant as suppressed instead of quietly omitting it. That delta display already cites the Ice House as the bug that motivated it.

• Q13 — rear-end collisions on a Mainline card. Answered: collide on catching up. Implemented, and not on cards that print "trains may pass". Invisible to a bot that always denies clearance; a bot that always allows drops from 7.34 revenue to -5.13.

• Nine of the twelve special-train rules are declared and read by nothing. Done — all nine enforced, and one of them deleted instead. copiesNextScheduled was never carried by any train card: a Second Section is a Maneuver with its own working intent, so the flag was an unreachable second description of an existing mechanic. Cost 0.8 revenue and half the wins (8.0 → 7.2, 14/200 → 6/200), which is what enforcing restrictions does.

• Carry links forward in replay frames. Done, and the premise was wrong in an instructive way: measured, links was 5% of the cells payload. What actually cost was the what prose (32%), the facility object stored a second time inside its own cell (24%) and the rest of the static identity (26%). All three are interned now — 3415 KB → 1877 KB, and a round-trip test runs the page's own unpacking function.

• Every published replay was dead. All three replayed 2 intents of roughly 400 and presented as short games, exactly as the item below it (now in Replay / Save Games) predicted. Re-recorded from bot games with node src/sim/save-replay.ts, which verifies each save round-trips before writing it, and harness.test.ts now fails if a published replay stops short. The version-stamp item in Replay / Save Games is still worth doing — this catches the breakage, it does not explain it to a player.

• Coordinate labels read Y,X on the board — v0.4.9. Now X,Y everywhere a coordinate is shown to a player: the on-card label (board-svg.ts), the switching-crew tooltip and button (main.ts), the rejected-option and switching-group text (game.ts, view.ts), and the blocked-move text (narrate.ts). Display-only — GridCoord{row,col} already had the right geometry (row increases north, col increases east), and internal Map keys are untouched.

• "No switching" blocked moving a train clear of the mainline — v0.4.9. The six no-switching cards (both expresses, Light Engine, Campaign, Circus, Military) mean may not add or drop cars, not may never be touched. switch.move now refuses only a move that would couple a fresh car — the same way dropOnly was already handled — so these trains can still be shunted onto Secondary Track to clear the mainline. switch.dropCars and switch.sortConsist stay blocked.

• Q3 corrected: Expedite governs WHERE a train may stand, not WHEN it leaves — v0.4.9. The forced same-Stage departure (departsThisStage, the shiftChange expedite pass) is gone; an expedited train now arrives and is released like any other train, switchable in between. New fault instead: left off the Office square when a Mainline Phase begins, it costs 1 Revenue (expediteFault, EXPEDITE_FAULT_PENALTY), every Phase it is still caught there. Resolved "3/4 EXPRESS PRINTS A RULE IT CAN NEVER USE" as a side effect — it can now reach the Local Operations turn its printed freight rule needs. Revealed a bot gap instead: logged above under Bot Performance.

• evaluateClearance checked only the first occupant it found — v0.4.9. Found while explaining a playtest report: the Superintendent was asked to rule on a same-direction train instead of being held against an opposite-direction one also on the card, because the loop returned on whichever occupant it examined first rather than checking all of them — invisible until the Telegraph/Telephone/Radio exception made it possible for a card to hold two trains at once. Now checks every occupant for an absolute bar before offering any judgment call. Pinned with a test that fails against the old single-pass code.

• The splash page now shows the box art — v0.4.9. docs/StationMasterSplashScreen.png (3.0 MB) resized to a 145 KB JPEG (public/images/, copied into the build by build-web.ts) and placed beside the title, tagline, blurb and both buttons in a side-by-side hero, stacking to image-above-text on mobile.

• A coach may now be set out at the Office — v0.5.0, Jesse's call. §A.4's blanket "no Rolling Stock may be left at the Office" now carries one exception: any train (not just 7/8) may drop one or more coaches there; freight and cabooses stay banned. Unlocks the ENGINE boxcar coach arrangement that used to lock completely — measured at 1,181 refused set-outs over 60 games, every one of them this case. canDropCarsAt (track.ts) takes a coachesOnly flag instead of refusing the Office outright; trains 7/8's coachStaysOnStationTrack rule now forbids the coach everywhere EXCEPT the Office, rather than everywhere. The Office's "cars fouling the Running Track" collision (advance.ts) is exempted for coach-only standing cars, so a legally parked coach is not a hazard to the next arrival.

• 3/4 EXPRESS PRINTS A RULE IT CAN NEVER USE — closed, v0.5.0. Confirmed already resolved as a side effect of the v0.4.9 Expedite fix (see that entry above); removed from Rules Questions.

• THE INDUSTRY TABLE VS. THE CARD REFERENCE — v0.5.0, Jesse's call: the engine was right, the docs were stale. card-reference.md printed Grocer's Warehouse and Oil Refinery at 2/2 with 2–3 Laborers, and claimed "'Freight House' is not a card"; the engine already had both industries at 1 out/1 in/1 Laborer and already dealt Freight House as a real sixth industry, 6 copies. Corrected the docs (card-reference.md, glossary.md, rules-v0.2.md, open-questions.md) to match the engine; no engine change. Turned up that Mine Tipple, Produce Shed and Power Plant may be similarly stale — flagged as a new item above rather than assumed.

• Poling — closed, v0.5.0. Confirmed already at 0 copies, the same treatment as Sharp Curves, pinned by mainline-cards.test.ts. No code change; removed from Rules Questions.

• Heavy Grade orientation — asked again 2026-08-23, and the answer did not change. Raised as "did we ever fix Heavy Grade to allow user placement of direction?", with the option of giving the choice to the Superintendent considered and rejected. Jesse's call: v0.5.0 stands. Three things came out of the re-examination and are recorded in implications.md §10 Q11 so it is not asked a third time — the advantage is permanent while the office rotates, so a rotating chooser moves the fairness problem rather than solving it; there is no setup phase to ask in (createGame is pure, and pendingDecision is typed for clearance alone across 19 readers); and since v0.6.2 deals the Mainline deck without replacement, only 20% of solitaire games contain a Heavy Grade at all. The docs were the real defect — README.md listed it among three open rules questions, all three of which v0.5.0 had closed, and the Mainline deck reference said the implementation "needs a player-selection step". Both corrected; no code change, and none wanted.

• Heavy Grade orientation stays rolled, permanently — v0.5.0, Jesse's call. The card prints "Player sets orientation", but a Heavy Grade sits on the shared Division chain between two players (or beyond an end Division Point, next to one) — never inside one player's own district — so there is no single player with a fair claim to the choice. Settled as random from the seed, identically for solitaire and multiplayer, overriding the card's print. No code change (the roll in setup.ts was already doing this); the comments and implications.md §10 Q11 previously framed it as a placeholder awaiting an interactive setup phase — corrected.

• Engines are not a separate supply from Crew Trays — confirmed, v0.5.0. rules-v0.2.md:339 (Gap 4b) ties Crew Trays and engine pieces together as one combined resource, player count + 3 — not two independently-tracked supplies. The engine already enforces exactly that via crewTrayCount/freeTrays; NO_FREE_TRAY already fires exactly when engine supply would run out too. No code change; corrected the comments that called this provisional or unsourced.

• Player settings, saved — v0.5.0. A localStorage settings object (SETTINGS_KEY, separate from the game save) now persists district auto-focus mode, sound on/off, and board zoom level across reloads — all three previously reset every time. Falls back to today's defaults on a missing, corrupt, or disabled localStorage, the same guard the save already had.

• The test suite's flakiness under npm test — fixed, v0.5.0. Two changes: (1) a pretest npm script now builds the shared dist/ once, before node --test runs, so every test reading dist/ no longer depends on another test in the file having built it first; (2) the one test that actually exercises the build COMMAND now builds into its own dist-test/ directory (BUILD_DIST_DIR env var, scripts/build-web.ts) instead of rebuilding the shared dist/ out from under the tests reading it. dist/ is now single-writer.

• Curves and turnout diverging legs now draw as smooth curves, not two straight segments meeting at a corner — v0.5.0. curvedRail in board-svg.ts replaces the old hard-cornered "run to the frog, then a straight 45° leg" with a sampled cubic-Bezier easement: tangent to horizontal at the east/west edge (so an abutting straight card's rail still reads as one unbroken line) and tangent to exactly 45° at the north/south edge (so two stacked curves still read as one continuous diagonal). Both plain curve cards and a turnout's diverging leg go through this same code path, so both are fixed by the one change.

• Wide boards can now be zoomed — v0.5.0. Discrete zoom presets (75/100/125/150%) for both the district grid and the Division map, applied by resizing the rendered SVG's own pixel dimensions (not a CSS transform), so the existing overflow-x:auto scrollbars keep doing the panning with no new gesture code. Persisted in the new settings object above.