Quintine
7454456755
Players can ask total-hand and primary-role card-count comparisons using the existing count controls and compound facts. Component presentation shares the same default-role classification, including canonical Richese cards, Ix alternate uses, and Ecaz slot-only specials. Only the respondent receives the verified aggregate answer; current facts do not impose future custody promises. Eight new tests cover all 60 catalog cards, real engine actions, malformed input, private worlds, all four bots, JSON continuation, and a legal custody transfer. Expanded production-room SQLite cases cover restored total/category/compound facts, suspended auction state, and one winning concurrent answer/discard. Independent review's two copy issues were fixed before broad verification. Validation: types, lint and 4,200 offline tests passed; production build passed; 40 HTTP tests passed. One stable broad run (check 123036 ms, build 6505 ms, integration 23467 ms). Browser room LC3E55GY verified combined role counts, total hand size, invalid-input feedback and refresh at 390px phone width. All 33 pre-integration rooms preserved unchanged; HTTP checks added 30 rooms. No complete-mode, full-game or publication gate opened. Private report: 20260913-hand-inventory-final/report.json Source fingerprint: c6a69a9e69941cdd1e5734d92afedbf10d31fda36c91d26ca7e15f4fb7c59b18 Efficiency pilot milestone 3: aggregate counter 58,695,053 to 59,144,653 at completed verification = 449,600 tokens, including context/coordination overhead. Two bounded workhorse-model helpers, reused UI and SQLite fixtures, one browser room. Counter is not monetary cost and this differently sized batch does not establish a savings percentage. Final Git/push overhead follows this sample.
Dune — Arrakis Table
An unofficial multiplayer adaptation of Gale Force Nine’s classic 2019 Dune, built with React, TypeScript, Vinext, and Cloudflare Workers/D1. Development is ongoing; this is not yet a complete rules implementation. Advanced and unfinished expansion starts remain disabled. Start with current status for implemented systems, remaining requirements and release gates; detailed historical evidence remains in the implementation log.
Run locally
Use Node.js 22.13 or newer and npm. Install the locked dependencies and apply local migrations before starting the server:
npm ci
npm run db:local
npm run dev -- --host 0.0.0.0
Open http://localhost:3000. Create a room, share its eight-character invitation, and add human or AI seats. The host can configure Easy, Medium, Hard and Brutal AI opponents. Current calibration demonstrates a substantially weaker Easy policy; it does not yet demonstrate a consistent ordering among the upper three levels.
Local games persist in .wrangler/state. Keep this directory across restarts. Apply new additive migrations before running code that depends on them. Never delete the local database to resolve a connection or rules problem. .openai/hosting.json belongs to the existing Sites project; do not recreate the hosting project.
Saved seats and reconnecting
The room cookie identifies your seat. Refreshing or reopening the invitation with that cookie restores your private hand and pending decisions. A failed restoration offers an explicit retry. Requests time out after 15 seconds; uncertain game actions are reconciled with a read rather than automatically repeated.
Use Protect your saved seat to generate and privately save a recovery kit before enabling its key. Keep the complete kit separate from the invitation. Recover a saved seat uses that kit to restore the same faction after cookie loss and revokes the previous browser sessions for that seat. Recovery keeps existing forces, resources and sealed commitments. Anyone holding the kit can control that seat and see its private information.
If recovery receives no confirmed response, keep the page open and retry the same recovery request. The server recognizes the receipt and does not rotate twice. Replacing the recovery key invalidates the older kit. Host-initiated takeover remains unfinished. An owning human seat can enable or disable its own AI control during a started game.
New create/join requests save a private, tab-scoped retry record before sending. If the response is uncertain, keep the tab open and use the explicit retry; refresh preserves the same request. Confirmed success removes that record. Clearly rejected first requests return to the form for correction. A retry can recover the original room and seat without creating duplicates, but closing the tab or clearing its data can lose the proof. This record is separate from the long-term seat recovery kit.
An uncertain saved-kit recovery keeps its exact retry details in memory and prevents conflicting navigation. If that recovery has become obsolete, deliberate abandonment releases the controls after explaining that it cannot undo a completed recovery. Keep a valid saved kit before discarding retry details.
Verify changes
Use the development guide for architecture, focused tests and the repeatable verification tools. Tests are discovered automatically; the default suite includes in-memory persistence recovery and needs no server:
npm test -- ecaz-collection ecaz-spice
npm run check
npm run build
Run npm run test:integration against a development server for HTTP/session
changes. npm run test:multiplayer remains available for recovery plus HTTP
checks; npm run test:recovery runs only in-memory persistence tests. All test
commands accept filename fragments and --list after --.
The production build is written to dist. npm start runs that build with Wrangler; apply its database migrations to the intended storage environment before use. The included local migration command targets the development database, not a remote deployment.
Reproduce the current AI study with:
node --import tsx tools/ai-calibration.ts --games 120 --seed 20260907 --out /tmp/dune-ai-calibration.json
See AI calibration, multiplayer audit, component inventory, and visual playtest for scope and limitations. Passing subsystem tests does not certify every rules combination.
Development maintenance
The user-authorized 13 September 2026 local reset established a fresh saved-game baseline; see the reset record. Preserve all games created after that checkpoint.
During active development, preserve database state and perform a controlled server restart approximately hourly. Warn before interrupting human play and defer until a safe decision point. Verify server health and restoration afterward. No recurring automation is installed; the user requested leaving it removed. Fix causes of crashes and lost progress rather than relying on restarts.
The application bundles its artwork, fonts and rules guidance. Source provenance and third-party notices live in developer documentation; the player-facing reference uses internal links. Publication remains gated on completion and integrated verification of the requested rules, expansions and critical player journeys.