Let a human owner authorize another human in the started room to start a fixed difficulty once within 24 hours. The owner retains the seat, cookie, private hand, recovery setup and takeback control. Consent is explicit even while online; neither host status nor inactivity authorizes a takeover. Activation enters the existing paced private-view AI path without strategic changes. Add additive migration 0006, both exact session fences, owner/delegate-only metadata in the authenticated room snapshot, atomic CAS activation and consumption, expiry and terminal retry receipts. Replacement hides older consent; own AI changes invalidate unused consent. Recovery or handover of either bound seat invalidates the grant. Future-version guards prevent stale computed state from overwriting a racing action. Exact browser retry records retain the original operation/version across refresh and cannot rearm old consent. Shared HTTP continuation preserves coalescing across ordinary and control routes. Independent review found no material remaining issue. Revoke messaging accurately handles activation winning the race. Browser QA exercised grant/revoke, preserved consent after refresh, reciprocal fixed-profile activation, consumed receipts, keyboard takeback and private ownership. The panel now remains expanded after success and action buttons are at least 44px. Responsive screenshot acceptance remains open because the browser viewport override targeted a failed secondary-origin tab; desktop controls and no-overflow width were verified. The QA room is stopped at a genuine human setup decision with both seats taken back. Types, lint, all 4906 offline tests, production build and all 45 HTTP tests pass on the stable source fingerprint. Seven new offline cases cover client consent/retry validation and real SQLite privacy, restart, one-use races, future versions, takeback, expiry and both-seat recovery/handover; two new HTTP cases cover authentication, injection, fixed-profile continuation and takeback. The first broad run exposed 68 legacy VM fixtures missing the new production dependency; their mappings were updated and the real shared continuation remains exercised. The historical 0005 migration test verifies backfill before applying 0006 for current code. All 1174 original rooms, 2456 seats and every prior recovery/entry/handover/migration row remain unchanged. SQLite integrity and foreign keys pass; migration and QA added 88 rooms and 188 seats. No reset, server restart, usage check or automation was performed. The current Sites integration has no supported durable Cron/DO provisioning path. Document this platform blocker rather than adding an inert scheduled export; unattended resumption and broader network-failure acceptance remain unfinished. Full rules, component, AI strength and publication gates remain closed. Verified source tree: 2d86fc611e9ef4f5a10db10f96687a31db1ec238908ecf1e9333c50fac9ccaa8
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.
Full AI implementation and tuning wait until all non-AI game features are complete. The AI development plan then targets approximately 75% higher-tier wins for Medium/Easy, Hard/Medium and Brutal/Hard. Until that stage, AI work focuses on minimal legal participation and critical playability fixes.
Choose Learn to play for the interactive introduction
at /learn. Practice shipping, movement, battle plans, Traitor calls and collection before entering a
room; lesson progress stays in your browser and does not change saved games.
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.
For a planned absence, Let a player start AI for your seat grants one named human player permission to start your chosen difficulty once within 24 hours. They can activate it while you are online, but receive no private seat access. You can revoke unused permission or take back control. Recovering or transferring either seat invalidates the grant. See AI permission for saved retries and limits.
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.
Use Pass your seat to another player for a voluntary one-time transfer. The private offer lasts 24 hours; acceptance preserves progress and revokes the old owner’s sessions and recovery kits. The recipient uses Accept a seat handover on the home page, then creates their own recovery kit. Keep the tab open until acceptance is confirmed; exact private retry details survive refresh in that tab. See seat handover for cancellation, retry behavior and current limits.
The table header offers optional Sound effects, with mute, volume and a test sample. Short cues mark phase changes and automatic action notices. Sound starts off; preferences stay in your browser. See table sounds for activation, background behavior and prototype limits.
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.
Reuse the running development server. Restart only when a code or configuration change, or an observed server condition, makes leaving it running likely to cause issues; use judgment rather than an hourly schedule. Preserve database state, warn before interrupting human play, and defer until a safe decision point. Verify server health and restoration after a necessary restart. 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.