Dune/docs/ECAZ_COLLECTION_IMPLEMENTATION.md

Ecaz shared spice collection

Verified development checkpoint, 9 September 2026. This completes the bounded shared-collection flow; complete Ecaz, Advanced and expansion starts remain gated. The source contract records publisher precedence and sector-capacity interpretation. The earlier runtime readiness review is a design proposal; the implemented commitment order below supersedes its deferred-payment proposal.

Implemented behavior

The engine computes each collecting faction’s eligible sector capacity before extracting spice. When both Ecaz and its reciprocal ally collect from the same desert territory, their combined extraction becomes one shared pool. The board loses that spice once. Ordinary collection and bank income settle automatically; shared spice is held outside both players’ balances until allocated.

Ecaz proposes first as an interface convention. The other ally can accept or counter. Either current decision owner can use the equal split, with odd spice going to the ally. Each settled territory has one persisted receipt. Unaccepted proposals are visible only to the pair; the chronicle records final allocations and their reason. Empty pools create no confirmation.

Advanced co-occupied Arrakeen, Carthag and Tuek’s Sietch pay both allies their normal bank income. A meaningful Karama window precedes that commitment. Cancellation removes only Ecaz’s income from those shared strongholds. Printed Karama and Bene Gesserit Worthless conversion preserve the same source and effect. No available blocker means automatic continuation.

Pending events bind their turn, public collecting forces, original deposits, allocation index and settled receipts. Actions, player views and automatic recovery reject missing or falsely completed collection parents, including controls suspended by Nullentropy Box and Karama conversion. Legitimate card income during allocation is retained: settlement adds to the current balance. Room version checks prevent simultaneous or stale acceptance from paying twice.

Coverage checklist

Area	Verified boundary	Evidence
Implementation	Sector capacities, one pool per desert, ordinary/bank commitment, bilateral allocation, scoped cancellation, recovery integrity	game/board-resolution-quote.ts, game/ecaz-spice-allocation.ts, game/engine.ts
Player controls	Private proposal/counteroffer/acceptance, equal fallback, decision ownership, mobile controls, keyboard completion, refresh	components/ecaz-spice.tsx, components/game-table.tsx
AI	All four profiles use the equal fallback or accept a fair offer without consulting rival balances	game/bots.ts, tests/ecaz-collection-engine.test.ts
Documentation	Internal phase and faction guidance, source precedence, commitment order and explicit release limits	game/reference.ts, this document and the linked source contract
Verification	Pure, engine, interruption, privacy and SQLite concurrency/recovery cases; real browser allocation	Four collection test files listed below

Verification

npm run check passed typecheck, lint and 3,028 offline tests, including the user’s test-discovery and failure-propagation self-tests. npm run build passed. npm run test:integration passed 40 tests against the local development server. These suites have different scopes; historical counts used the previous manual test registration and are not directly comparable.

The focused collection coverage is 60 tests:

• tests/ecaz-collection-quote.test.ts: 20 capacity and income cases.
• tests/ecaz-spice-allocation.test.ts: 11 protocol cases.
• tests/ecaz-collection-engine.test.ts: 26 actual transition, privacy, AI, cancellation and suspended-parent recovery cases.
• tests/ecaz-collection-recovery.test.ts: 3 SQLite scenarios, including racing proposals/acceptance, printed and converted cancellation, reconnect and secrecy.

The browser resumed the isolated three-seat QA room 8S3MRDEK. Actual last movement entered Advanced collection, paid bank income automatically and opened the five-spice Hagga Basin allocation. The Medium AI ally accepted the 2/3 split. Refresh restored the six-spice Wind Pass choice; keyboard navigation and Enter selected its 3/3 equal split. At a 390×844 viewport, controls were readable, 44 pixels tall, visibly focused and free of horizontal overflow. The normal viewport was restored afterward. Room version 126 retained two receipts, one spice on Wind Pass and total balance changes of +7 Ecaz, +8 ally and zero rival. All 2,556 pre-existing non-QA room versions and state hashes were unchanged. This staged position verifies the interaction, not a complete expansion game.

Twenty ordinary production-AI Basic games, spanning two through six players and all four levels, completed 10,206 accepted actions and 465 JSON round trips with zero rejected candidates, stalls or invariant failures. Seed: 20261024. Source fingerprints stayed unchanged during the run. These games cannot exercise Ecaz because expansion starts remain gated, and homogeneous-level games do not measure the relative strength of the levels.

Local logs: /tmp/dune-collection-sep09-check-verified.log, /tmp/dune-collection-sep09-build-final.log, /tmp/dune-collection-sep09-http.log, /tmp/dune-collection-sep09-fullgames.json, and /tmp/dune-collection-sep09-browser-evidence.json.

Remaining release gates

Combined Occupy combat and its source discrepancy, Fremen’s Ecaz endgame exception, Tleilaxu Ambassador runtime, remaining Duke interactions, complete Advanced/expansion/module games and broader presentation acceptance remain unfinished. This checkpoint enables none of those modes. The maintenance automation remains removed.