Upstream/docs/NEXT.md
jvanvik 931f59e11e Phase 3: trade/end-game client, bot simulator, first balance pass
Built:
- Client: trade screen, world/end-game screen (landing sites, shipping, node build, final ranking), report sharing + alliance reports tab, alert describers for all notification types, error-code strings (en+nb) for every server CODE.
- tools/bot.js (+ tools/bot/): 50 headless bots against the real server in-process with a fake clock; resource/unit conservation ledger, exit 1 on broken invariants. docs/BALANCE.md has method and every run behind a number.
- Balance changes (evidence in docs/BALANCE.md): secondQueueCoreLevel 10 -> 3, TRANSPORT.capacityPerLink 500 -> 1000 (end game was shipping-limited in bot runs).
- Server fixes: POST /reports/:id/share 500 without a pending reportCleanup; bounced link shipments were silent about cargo overflow.

Tested: npm test 207/207 (new: bot determinism + invariants, client alerts/world/i18n over CODES, two regression tests that failed before their fix); openapi check up to date. Real-browser scratch runs (not in repo): WebSocket push shows an incoming-attack banner ~214 ms after launch, polling fallback ~52 s, 15 scripted flows, 71 screenshots.

NOT tested: Playwright suite (e2e/ still empty), real devices/Safari, nginx proxy websockets, throttled networks, HTTP-level bot load. NOT claimed: balance. Bots are not humans; open balance items (loot waste, escrow refund clipping, no garrisons at landing sites, end game not scaling with world speed) are listed in docs/BALANCE.md.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q2MJaQ5LXAHzGnTDgceT14
2026-10-11 20:24:41 +02:00

27 KiB

NEXT

Honest status, updated at the end of every stage.

Done

  • Stage 0 skeleton: db + migrations, clock, scheduler with catch-up, router, auth/tokens, error envelope, openapi generation, static server.
  • Stage 1 economy (server): lazy resources, buildings, build queue, storage caps, power battery.
  • Stage 2 world (server): seeded tiles, spawn ring, map viewport query.

Stage 5 (partial): social, server only

Built (migration 003_social.sql): alliances (found, invite/accept/decline/cancel, leave, kick, roles, leader transfer, disband, pinned notes, member list with POP count and last online), diplomacy between alliances (war, neutral, NAP, confederation; proposals; 24 h / speed pact cancel through the diploChange event; revoke), messages (DM, alliance broadcast, inbox/outbox/thread/ alliance feed, read, delete own copy, block list, DB-backed rate limit, text sanitising), rankings (operators, alliances, battles), alliance in /me and /map, lookup helpers in server/alliances.js (operatorRelation, atWar, confederated, pactPartners). Tests: test/social-*.test.js (pure rules, alliances, diplomacy with the fake clock and event queue, messages, rankings), every response in them is checked against its declared schema (createTestServer({ contract: true }) in test/helpers.js).

Not done:

  • Marketplace / Internet exchange: DONE on the server, see "Trade, end game, gaps" below.
  • Shared report folder and "see each other's incoming": DONE on the server (below). The pact effect "no attacks inside a NAP/confederation" is enforced by the military stage (pactBlocks).
  • Web client: see the client sections below; WebSocket push (stage 6) exists server-side and in the client.
  • Message retention/cleanup of old rows, alliance rename/description, mass-mail abuse reporting, sitter accounts.
  • Rankings are not cached; battle rankings stay empty until the military stage writes operator_stats.
  • Not run: Playwright, load, or any bot balance run; the social numbers in SOCIAL (limits, TTLs, role table) are first guesses.

Stage 3/4: military and expansion (server only)

Built (migration 002_military.sql: town_units, stationed, town_research, town_power, movements, reports, report_recipients, operator_stats, incumbent_takes): 10 unit types and research in data.js/rules.js; training queue (trainDone per batch, sequential per building); research (researchDone); movements scan/raid/attack/takeover/reinforce/settle/transport with latency, ETA, recall (90 s), return legs with loot; combat resolved at arrival in the event queue (full attack, raid, firewall vs DDoS rigs, exploit kits on a chosen building, loot with Rack hall cache, losses, reinforcements, research, battle points in operator_stats); scans with intel and detection; takeover with loyalty, capital and last-POP protection, footprint gate; founding POPs with wild-slot choice; incumbent sites (static defence growing with world age, one-time Hardware bonus, +10% research speed per site); beginner protection (both directions, re-checked at arrival); power: offline units when the battery is empty, decommission after 24 h (powerCheck); incoming warnings with hidden type; reports (shared row, per-user read flag, 14 day cleanup event); transport movement and arrival hook (movements.hooks.transportArrived) for the trade stage. Routes: see docs/API.md, all in docs/openapi.json. Tests: test/military-*.test.js (rules as pure functions, train queue order, raid end to end, event order and determinism, takeover and its protections, offline units and decommission, restart catch-up from a reopened DB file, protection, settle, incumbent, transport, recall, research, reports/cursor/retention, API smoke over HTTP). Fake clock, in-memory DB, port 0; no Playwright. Not done / not claimed:

  • Nothing is balanced. All unit stats, costs, research curves, incumbent defence and the 50-unit-scale combat numbers are first guesses; there was no bot run when this was written (tools/bot.js exists now: see "Stage 7" below and BALANCE.md; most of these numbers are still untuned). Do not cite any of them as tuned.
  • Training cannot be cancelled; no sitter, no night protection. (Report sharing and allied incoming: done, see below.)
  • Units in transit draw no power from anyone; an operator cannot choose a new capital POP (no command); a captured POP keeps its footprint.
  • Backbone landing sites and the Backbone node: see "Trade, end game, gaps" below.

Client (stages 1-2 web client)

Built: public/ vanilla JS, no build step, strict-CSP friendly (no inline scripts or handlers, no external requests).

  • Shell (js/main.js): session, App.registerScreen({id, titleKey, navKey, icon, hud, render, onShow, onHide, onLang}), bottom nav (left rail in landscape and on wide screens), live resource bar ticked with rules.resourcesAt from the server clock, hash routing.
  • Screens (js/screens/): login/signup (invite field appears after signup_closed), POP (resource fields by type, infrastructure, build queue with countdowns and cancel, power panel, detail sheet with the server's next plan, live affordability, rename), map (js/render.js canvas: pan, pinch, wheel, keyboard, culling; chunked GET /map cache; tile info sheet), alerts (GET /notifications), settings (language, device tokens + revoke, logout).
  • js/api.js: bearer token, serverTime offset, retry/backoff (network, 429 + Retry-After, 502-504), Idempotency-Key per intent, ApiError.code, event emitter with a reserved push event (no WebSocket client yet).
  • js/i18n.js: en + nb, all player-facing names; PWA: manifest + icons (no service worker, so no offline mode).
  • Tests: test/client-i18n.test.js (key parity, names, error codes, formatting), test/client-api.test.js (retry, idempotency, offset).
  • Looked at with headless Chromium screenshots (phone portrait, phone landscape, desktop) and one scripted signup/build/cancel/map-tap flow on a phone profile. Not run: real devices, Safari/iOS, Playwright suite. Gaps: the unread badge is stored in localStorage (no "mark read" route); /me exposes the world seed which the client ignores; no push, no service worker, canvas has no inertia/fling, no light theme.

Client: social (alliance, messages, rankings)

Built (all in the nav "More" group, navGroup: 'more'): screens/alliance.js (found with Peering office gate and translated server errors; received invites accept/decline; overview with size, pinned notes add/remove, leave/disband; members with role, POPs, footprint, last online, a member sheet for message / role / transfer / kick; invite by name, pending invites with cancel; diplomacy with relations, proposals in/out, "new move" via directory search with only the moves rules.diplomacyMove allows, pact-cancel and proposal countdowns, take-back), screens/messages.js (inbox with unread filter / mark all read, sent, alliance feed with broadcast composer, per-operator thread with older-page cursor, compose by name, 2000 char counter, delete own copy, block list and block/unblock, 429 shown as a countdown with the text kept, unread count on the nav via App.setBadge, polled every 60 s), screens/rankings.js (operators, alliances, battles attack/defense/total, your row highlighted plus the server's you card, rank-offset "load more"). Shared bits: screens/social-ui.js (window.SocUI: composer, confirm sheet, alliance profile sheet, error mapping), js/social.js (pure logic: what to show per role via rules.allianceCan/outranks, diplomacy options, counters, cursor merges), js/strings/social.js (en + nb), new sprite icons users / mail / trophy / back, .soc- CSS. All player text goes in with textContent (a test fails on any innerHTML in these files). Tests: test/client-social.test.js (pure logic, string coverage of every static key the screens use, error codes, no HTML sinks). Looked at with headless Chromium (phone portrait/landscape, desktop) against the real server in-process, plus a scripted flow on a phone profile (markup in a message stays text, counter/over-limit, 429 countdown, unknown recipient, leader vs member views, diplomacy move sheet, name-taken error, nb). Not run: real devices, Safari, Playwright suite, a long-list scroll test (>100 ranks), accepting an invite when the alliance is full, alliances at the size cap. Gaps: alliance rename/description and note editing have no UI (server has edit-note); no operator profile page; blocking needs an existing conversation or message (the API takes an operator id and there is no name lookup); no live refresh (push is not wired, screens refetch on open and on manual refresh); the alliance screen does not auto-refresh countdowns' targets after they pass; App.state.operator.alliance is refreshed from /me after alliance changes but is stale if you are kicked while away until the screen is opened.

Client: military, expansion, WebSocket

Built (public/, same rules as before: no innerHTML for player text, no inline handlers):

  • Nav overflow (main.js): 5 slots; registerScreen({ navGroup: 'main' | 'more' }), with any more screen the 5th slot is a "More" sheet; App.setBadge(screenId, n, 'alert'|'info') (More sums its screens, red if any is an alert). Bar now: POP, Map, Units, Moves, More (Reports, Alerts, Settings, the social screens). App.alertKinds[type] lets a screen describe its own notification types in the Alerts feed; App.popSections lets screens add cards to the POP screen (incoming banner, expansion card).
  • Shared state (milstate.js): App.mil.moves (last GET /movements: outgoing, incoming, alliedIncoming) is the one source for the nav badge, POP banner, movements screen and map lines; App.refreshMovements/refreshReportsUnread/refreshTowns (the last refetches /me without the destructive setSession, used after a settle/takeover); [data-cd] countdown elements driven by one ticker; refetch when anything on the wire is overdue.
  • Units (screens/units.js): home units with research-adjusted stats, reinforcements here (send home) and your units elsewhere (call back), training queue with countdowns + train sheet (rules.trainPlan preview, max-affordable, queue-full), research list + sheet (rules.researchPlan, lab level / cost / time / reason), offline warning with the decommission countdown.
  • Movements (screens/movements.js): incoming list on top in red (origin + ETA only, as the API gives it), your outgoing legs with progress and a Recall button for the first 90 s, allied incoming; the POP screen gets a pulsing banner, the nav button a red badge.
  • Send dialog (screens/send.js, App.openSend({x,y,fromTownId,type}), from the map tile sheet, Units, Movements): GET /map/tile, types filtered by milcore.moveOptions (greyed ones say why), unit pickers (slider + stepper + all), cargo for transport, settle with the wild-slot choice, latency/ETA/carry preview from rules.movementMs, confirmEndProtection checkbox, exploit-kit target, an idempotency intent per opened dialog, server errors shown inline (invalid_target reasons translated).
  • Reports (screens/reports.js, in More): cursor-paginated list, unread highlight + badge, mark read on open, mark all read, detail layouts for raid/attack/takeover (role-aware victory/defeat, units sent/lost/left per group, strength, loot, siege, takeover loyalty), scan (+ incumbent intel), scan_detected, incumbent, settle (opens the new POP), reinforce, movement_failed; coordinates jump to the map.
  • Expansion (screens/expand.js, send.js, main.js, render.js): POP switcher in the resource bar, expansion card on the POP screen (crews at home, footprint gate, POP list), settle through the send dialog; map draws your/incoming/allied movement lines with markers, a red ring on your POPs that something is heading for, and a Send button in the tile sheet.
  • WebSocket (ws.js): createSocketClient (injectable WebSocket and timers) + browser wiring. Auth as first message, ping every 25 s with an 8 s pong timeout, jittered backoff 1 s .. 60 s, PROTOCOL close codes (4401/4400/1009 stop, 4429 waits 30 s), refetch after every re-ready. Pushes go to api.emit('push') and refetch the affected thing; slow REST poll (45 s with the socket down, 3 min with it up).
  • Tests: test/client-mil.test.js (type filter, unit roles, previews, perspective, backoff), test/client-ws.test.js (auth, ping/pong, backoff, close codes, no token, throwing constructor); client-i18n parity covers strings/military.js.
  • Looked at: headless Chromium (phone portrait/landscape, desktop; en and nb on the phone) against the real server in-process with a real clock, seeded units/movements/reports, plus scripted flows on a phone profile: real WebSocket open, an incoming push from another operator adding a row and a toast without polling, recall, raid/attack/transport/settle sends through the dialog (protection checkbox gates the button, exploit-kit target reaches the server), train, withdraw, offline warning, report details of every type, POP list after a settle resolved (refreshTowns). Not done / not claimed:
  • Not run: real devices or Safari, throttled networks, a socket behind nginx Proxy Manager, long sessions, the Playwright suite. The settle and takeover resolution was exercised by fast-forwarding movement times in the DB, not by waiting.
  • Takeover and capture flows (town_lost/town_gained pushes, losing the current POP) and the confederation alliedIncoming view have no browser run. The Hostile takeover button stays enabled on a capital / last POP (the client cannot know); the server's refusal is shown.
  • Withdraw and "call back" are one-tap with no confirmation. No unit icons (role icons only), no battle simulator / odds preview, no cancel of training (the server has none), no sound or system notifications, no reports search/filter.
  • The exploit-kit target picker only offers infrastructure types (the server also takes field slots f1..f18). Report group labels for reinforcements show #<townId> because the report carries no names.
  • Close 4401 stops the socket but does not itself send the player to login: that happens on the next REST 401 (api unauthorized).
  • ready.serverTime is not used for the clock offset (no round-trip estimate on the socket); REST responses keep doing that.

Client: trade, end game, alerts (public/)

Built (More group; strings in strings/{trade,world,alerts}.js, en + nb):

  • Exchange (screens/trade.js): tabs Offers / New / Mine / Send. Offers: filters give / want / POP, reach pills, "Take" sheet with a POP chooser from acceptableFrom (distance, transport time, stock, free links, why a POP is greyed out). New: live check with rules.tradeTermsCheck (1:2 band, 1000 per link, distance), stock and free-link checks, the exchange level a taker needs, escrow/expiry rules in words. Mine: status chips, poster/acceptor role, expiry countdown, cancel. Send: direct transport through the send dialog (App.openSend({ type: 'transport' }); milcore.moveOptions got a transportOnly flag so a rival's POP can be the target) plus links in use per exchange POP. No exchange: empty state with the requirement (Core router 4).
  • Backbone race / ended world (screens/world.js, endcore.js, cargo-ui.js): phase banner (countdown to the landing sites, leading alliance, winner, ended), the 7 sites with holder / level / build progress / stockpile, Manage sheet for your alliance's sites (build next level, ship resources, ledger), scan/attack through the send dialog (milcore.moveOptions landing branch), final ranking tabs after the end. Whole-client ended mode: App.worldEnded(), pill in the top bar, card on the POP screen, action buttons (primary/danger/[data-act], not [data-keep]) disabled by a MutationObserver; a stale page learns it from a world_ended error (via App.errMsg), the world_ended push or /me.
  • Map: landing sites drawn as a diamond with a level ring (gold = incumbent, cyan = yours, magenta = rival alliance), legend row, tile sheet (holder, level, Backbone race button, send). Reports: landing type, scans of landing sites, share to alliance, Alliance tab (folder), remove from folder; shared reports are told from nobody's side. Movements: shipments to a site are labelled; alliedIncoming was already there and never counts in the red badge.
  • Alerts (alertkinds.js, pure): every notification type the server writes (found by a test that greps server/) has a describer; includes trade_, backbone_, world_ended, alliance_, diplomacy_, message kinds.
  • Tests: test/client-world.test.js (endcore, moveOptions, string coverage, no HTML sinks, every public script parses), test/client-alerts.test.js, test/client-i18n.test.js now iterates CODES of server/errors.js. 203 tests green.
  • Looked at in headless Chromium against the real server (real clock, in-memory DB; scratch scripts, not in the repo): screenshot sweep of every new screen on phone portrait, phone landscape and desktop (en, plus nb for trade/world) in three world states (day 62, day 100+ with held sites, ended); scripted flows with DB assertions (take an offer, post with live validation, cancel, transport to a rival by coordinates, ship, build, scan/attack dialog, tap a site on the map, share and unshare a report, alerts feed, a stale page hitting world_ended).
  • Push, end to end in real browsers: Alice raids Bob; Bob's page (real login, WebSocket open) showed the incoming banner, red Moves badge and toast 214 ms after the launch with no reload; the banner went away and the Reports badge rose when the raid arrived. With WebSocket undefined, and with a socket that is closed at once (reconnect loop, 7 attempts in the run), the same raid showed up after about 52 s through the REST poll (15 s tick, 45 s gap): worst case about a minute. Confederate Carol gets no push for alliedIncoming (known server gap): it shows when the Moves screen is opened or the poll runs, and never in her badge. Found and fixed in the client: report pushes now also refetch movements (the incoming banner used to linger up to a few seconds after the arrival). Not done / not claimed:
  • Not run: real devices, Safari, a socket behind nginx Proxy Manager, throttled networks, the Playwright suite (e2e/ is still empty), long lists (offers > 25 per filter paged only up to what the sweep seeded).
  • Ended-world locking is a DOM sweep, not per-button wiring: a new primary/danger button that is NOT a game command must carry data-keep. Messages, ranking and settings screens are exempt by selector. Open send/trade sheets that were opened before the end stay usable until the next DOM change or a refused command.
  • Trade: no order book or price history, no notification for new offers in reach, offers are listed per page of 25 with a cursor only. The landing-site manage sheet does not poll while open (it reloads after your own actions and when the site screen refreshes). Ledger shows the top 15 contributors.
  • Site numbers are index + 1 ("Landing site 1..7"), not a name from the server. backbone_open is only pushed on first touch (server E2): the countdown flips by the client clock at opensAt, the sites appear after the next status fetch.
  • A transport to a rival's POP is a gift that warns its owner (incoming, type hidden); the UI says so, but nothing stops a griefing stream of them.
  • Server bug found and fixed (tiny): POST /reports/:id/share threw a 500 when no reportCleanup event was pending (ensureCleanup needs ctx.enqueue); reportShare now wraps its ctx with ctxOf, regression test test/report-share-cleanup.test.js (failed before).

Stage 6 (server half): WebSocket push

Built: /ws on the game's HTTP server (server/push.js transport, server/pushHooks.js routing), contract in docs/PROTOCOL.md. Auth as first message (ready / close 4401), Origin check (net.originAllowed), per-IP cap MAX_CONN_PER_IP (HTTP 429), 1 KB frames, 10 s auth timeout (4408), per-socket token bucket (4429), bad-message budget (4400), ping/pong keepalive that terminates dead sockets and re-checks the token on every tick (revoked/banned: 4401), graceful 1001 close inside srv.close(). Hints: town, incoming, notification, report, message, pong. Every notifications row (via scheduler.notify -> post-commit outbox -> onNotify) pushes a notification hint, plus the typed ones; finished events and committed commands push town to the owners of the POPs they touched. Hints go out only after the transaction that caused them committed (a rolled-back event or command pushes nothing); identical hints of one tick are merged. Small edits outside the new files: scheduler.js (outbox, notify returns the row id), commands/index.js (hooks.onCommand, outbox flush/rollback), index.js (upgrade + shutdown). Tests: test/push-ws.test.js (auth ok/bad/timeout, origin, path, per-IP cap and its release, rate limit, bad budget, oversize frame, dead socket, revoked token, banned user, shutdown, REST with no socket) and test/push-game.test.js (build/train done, incoming only to the defender, report/message/alliance invite hints, recall, coalescing, rollback). Real ws client, fake clock, port 0. npm test: 113 pass (run after the push work, with the other agents' changes in the tree at that moment). Not done / not claimed:

  • The client now exists (public/js/ws.js, see "Client: military, expansion, WebSocket") and was run against this server in a headless browser; not through a proxy.
  • Revocation closes the socket at the next ping tick (default 30 s), not instantly: POST /auth/logout and token deletion do not call push.
  • No load test (many sockets, large catch-ups), no test behind a real nginx Proxy Manager (upgrade headers, idle timeouts, X-Forwarded-For for the per-IP cap); only the code paths of net.clientIp are reused. The proxy's idle timeout (NPM default 60 s) is above nothing we send: the 30 s ping keeps sockets alive, but that is reasoning, not a measurement.
  • Hint gaps: owners of reinforcements standing in an attacked POP get report but no town hint for their home POP; no incoming for confederates; no APNs/FCM sender (the notifications table plus scheduler.onNotify is where it would attach); no resume after a gap.
  • The ws tuning knobs (wsPingMs, wsRate, ...) are createServer options only, not environment variables.

Trade, end game, gaps (server only; migration 004_trade_endgame.sql)

Built and tested (test/trade-*.test.js, test/endgame-*.test.js; fake clock, in-memory db, port 0; contract-checked responses over HTTP):

  • Trade: offers "give X of A for Y of B within Z tiles" with escrow, a link reserved per open offer, 1:2 ratio band, expiry event, cancel, accept = two transports in one transaction (all or nothing), the transportArrived consumer completes the offer, void when the POP is taken over. Conservation is asserted (every unit in POPs + escrow + on the road, before and after; the only loss is the existing Rack hall overflow, asserted too). Routes under /trade/offers.
  • End game: 7 seed-derived landing sites (rules.backboneSites, tileAt with the world age; spawn and settle refuse them from day 0), open at world day 100 / speed, taken with scan/attack against a static heavy defence, node level 1..100 built from a stockpile fed by transports (/backbone/:id/ship, /build), winner at level 100, worldEnd two world days later with a stored ranking snapshot and a frozen world, world_ended for every game command, GET /worlds/:id/status and /results. One end-to-end test runs day 100 -> take -> ship -> build -> win -> ended on a speed-2 world; the ended-world gate is also driven over HTTP.
  • Gaps closed: shared report folder (report_shares, /reports/:id/share, /alliances/:id/reports), alliedIncoming in GET /movements, buildDone is relative to the building's current level (test written first, failed before the fix). Not done / not claimed:
  • Nothing here is balanced. Defence multiplier 12 (+3% per level), node cost 4000 x 1.045^(L-1) (about 7.2 M of each resource in total), node time 2 h x 1.025^(L-1) (about 36 world days), 48 h offer ttl, 10 tiles of reach per exchange level, 500-per-side offers: first guesses, no bot run when written (now there is one: BALANCE.md section 5 says an 18-member alliance needs 55 to 78 days after the sites open with 1000 per link, and does not finish in 175 days with 500).
  • No garrison at landing sites (E1): defence is static, holders cannot station units. If the first world shows the race is decided by who hits first, the next step is a small site_units table with reinforce-to-site and defenders in landing.js.
  • No backbone_open push at the moment the sites appear unless something touches a site first (E2); clients see the phase by polling status.
  • Trade: no multi-link deals (each side <= 500), no price history or order book, no counter-offers, no notification when a new offer appears in reach, offers are not hidden from enemies, the escrow of a taken-over POP goes to the conqueror.
  • Ended worlds: movements still out and queued builds are left as they were (events cancelled, E3); there is no admin "end world now" or "reset" yet (stage 7 world lifecycle: the worldEnd event can be enqueued by hand); several winners are impossible by design.
  • The ranking snapshot is the same numbers as /rankings, computed once; /rankings/* after the end still compute live from the frozen state.
  • Client for all of this: see "Client: trade, end game, alerts" above.

Stage 7 (first part): bot harness and first balance pass

Built: tools/bot.js (+ tools/bot/*): N headless bots (builder, raider, conqueror, turtle, expander, trader, alliancer; a coordinated alliance plays the end game) against the real server in-process (createServer with an in-memory DB and a FakeClock, bots drive srv.commands.run, no HTTP), 50 players x 14 days in ~3 s, x 170 days in ~140 s. Per-day economy, progression, combat, trade, alliance, end game, performance and error tables; --json, --set PATH=VALUE data overrides, --sessions-per-day, --always-on, --speed, --strategy-mix. Exit code 1 when an invariant breaks: exact resource conservation (production measured at every settle, sinks from command results and notifications), exact unit conservation, no negative stock, no failed events, no unexpected error codes. docs/BALANCE.md has the method, every run behind a number, what was tuned and what was not. Tuned (each with before/after runs in BALANCE.md): second build slot at Core router 3 (was 10), transport capacity 1000 per link (was 500). Server fix found by the harness: bounced shipments reported no overflow (events/return.js). Tests: test/bot-run.test.js (12 players x 6 days twice, identical stats, invariants, the checks catch tampering, a fast world reaches the end game), test/bot-server-fixes.test.js. Not done / not claimed: the game is not balanced; the bots are not humans (no reinforcement, recall, messages, sitters; fixed sittings). Open items from the pass: a 3x world's end game is as long as a 1x world's (shipping is per sitting; capacityPerLink x speed needs the client to stop reading the constant); trade escrow refunds are clipped at the Rack hall cap silently; raid loot is split evenly over resources and destroyed on overflow; the capital runs out of things to build ~50 days before the backbone opens at 4 sittings a day; a built-up landing site is out of reach of any single-POP army (no garrisons, E1), so there is no fight for a site after the first days.

Not done

  • Everything else. See CLAUDE.md "Working in stages".