REQ-PRODUCTION-005: Phase-5 production/extraction execution and output sale intent #587

Goal

Implement permanent REQ-PRODUCTION-005: deterministic Phase-5 physical production/extraction execution from the already-integrated ProductionPlan and LaborAllocation evidence, followed by the canonical post-production OUTPUT sell intent, with exact input/resource conservation and no phantom stock.

Evidence

• Current master is 06bd1a5082015f0bac08ba30f0b71a7c3f217155; REQ-PRODUCTION-002..004 are integrated, so Phase 5 can consume deterministic current-tick ProductionPlan/LaborAllocation evidence and authoritative ProductionUnit live inventories/capital.
• Canonical Drive Handoff/05 §16 defines realized batches as the non-negative minimum of planned, input, allocated-labor, effective-capacity and finite-resource bounds. Exact recipe inputs and extraction deposits decline once; exact recipe output enters the ProductionUnit OUTPUT inventory first; labor is not refunded when a later physical bottleneck binds; no routine random production multiplier exists.
• Handoff/05 §17 requires the post-production OUTPUT SELL intent before Phase 6: reserve = outputSalesEma × outputCoverageTicks; sellable output = max(0, outputInventory - reserve); actor is the ProductionUnit, purpose INVENTORY_REBALANCE, bucket OUTPUT, and sourcePlanId is the ProductionPlan.
• Handoff/05 §26 requires finite extraction deposits to bind and deplete exactly once; depleted deposits produce zero further extraction.
• The registry acceptance for REQ-PRODUCTION-005 requires realized batches never exceed plan/input/labor/capital/resource bounds, exact stock deltas, output-first inventory flow, canonical market settlement for sales, and no phantom/random stock creation.
• OPEN_QUESTIONS.md has no unresolved blocker for this boundary.

Scope

• Add an authoritative live Region resource-deposit balance keyed by resource/good identity, initialized deterministically from RegionSeed.deposits without changing the immutable scenario seed.
• Implement deterministic one-unit Phase-5 production execution for ACTIVE units only. Validate current-tick plan/unit/recipe provenance and finite non-negative evidence before deriving bottleneck bounds.
• Input bound: minimum live INPUT quantity / recipe input-per-batch coefficient; zero-input recipes are unbounded by inputs.
• Labor bound: current unit allocation total / max(recipe.laborPerBatch, numeric.quantityEpsilon), with +Infinity when laborPerBatch == 0.
• Capital bound: current ProductionPlan effectiveCapacityBatches.
• Extraction bound: for recipes with extractionResourceId, remaining live regional deposit / positive extractedResourcePerBatch; non-extraction recipes are unbounded by resources.
• Realized batches = max(0, min(planned, input, labor, capital, resource)); consume exact input quantities once, deplete exact extraction quantity once, and credit exact recipe output once to the unit OUTPUT inventory.
• Preserve WorldState immutability during planning/execution evidence creation; persist the physical deltas through an explicit state-transition function returning a new WorldState/registries.
• Execute units in stable ProductionUnitId order and reject duplicate/stale/cross-unit/cross-region evidence rather than silently aggregating it.
• Build canonical post-production SELL intents from the post-execution OUTPUT balance and outputCoverageTicks, using OUTPUT inventory bucket and the ProductionPlan as source.
• Add sensitive regressions for plan/input/labor/capital/resource bottlenecks, finite depletion and depleted-zero behavior, exact once-only input/resource/output stock transitions, output-first sale intent, non-ACTIVE zero execution, malformed provenance, insertion-order determinism and input-world immutability.
• Add a truthful REQ-PRODUCTION-005 = IMPLEMENTED ledger row only if this exact branch earns the full boundary; keep MERGE_COMMIT blank until integration.

Non-goals

• No Phase-6/8 market-clearing or settlement algorithm changes; this slice emits the existing canonical MarketIntent and ordinary sale realization remains owned by the market stack.
• No Phase-12 investment conversion/capital formation/depreciation (REQ-PRODUCTION-006).
• No Phase-14 lifecycle transitions (REQ-PRODUCTION-007) or integrated M4 closure (REQ-PRODUCTION-008).
• No new random production multiplier, same-tick revenue financing, labor refund/reallocation, persistent worker-employer matrix, M5 trade/FX, M6 fiscal/Clan dynamics, M7 monetary dynamics, M8 demography/migration, or docs/spec/mirror/** edits.

Acceptance criteria

1. ACTIVE-unit realized batches equal the non-negative minimum of planned, live-input, actual-labor, effective-capacity and finite-resource bounds; no bound may be exceeded and non-ACTIVE units realize zero.
2. Exact recipe input quantities are debited once from INPUT inventory; exact finite extraction quantity is debited once from the owning Region's live deposit; exact output is credited once to OUTPUT inventory. No negative stock, duplicate depletion or phantom output can occur.
3. Extraction recipes with depleted deposits realize zero; non-extraction recipes have no fabricated resource constraint. Invalid extraction provenance/amounts fail fast.
4. Labor evidence comes only from current-tick allocations for the same unit/region/category and is never refunded/reallocated because another Phase-5 bottleneck binds.
5. Phase-5 evidence construction does not mutate input WorldState. The explicit physical transition returns a new world and preserves unrelated wallets/inventories/capital/state.
6. Post-production OUTPUT SELL intent uses the post-execution OUTPUT balance, canonical reserve outputSalesEma × outputCoverageTicks, purpose INVENTORY_REBALANCE, bucket OUTPUT, stable MarketIntent ID and the current ProductionPlan as sourcePlanId; zero sellable output emits no material sale quantity.
7. Shuffled plan/allocation/registry insertion order produces normalized-equivalent execution results and intents in stable unit order.
8. Sensitive tests cover Handoff/05 PCL-T1/T2/T3/T10/T12/T18/T26-shaped cases and fail under controlled removal of an input/resource/output mutation or a bottleneck.
9. Exact-head TypeScript typecheck/tests/build, retained .NET build/test, policy/generated-status/ledger checks and mergeability are green before handoff. The authoring run stops at the producer→judgement boundary for SLOPSTER.

Verification

Run focused production-execution regressions plus repository TypeScript gates, retained .NET gates, implementation-status generation/check, policy checks and exact-head mergeability. Inspect the final diff for no mirror edits, no market-settlement rewrite, no investment/lifecycle/M5+ scope creep, and no second physical-stock authority.