TheTree_mock/docs/plans/original-completion-2026-10-02/execution/p04-01/s5/PENDING.md

REPAIR40 implementation supersession; native evidence remains pending

The implementation-pending statements below are superseded only for S5A/S5B power, S5C command destruction, S5D automatic raid destruction, S5E paid midnight and S5F command/planned exit. The single CASE_TABLE now dispatches these cases. No case is relabelled as an executed proof. Root must establish every outcome from fresh Q raw; an unalignable raid or unsupported condition fails normally.

The S5 order adapter does not infer patience_release or paused_midnight. Fresh S3 can supply patience_release only from its own actually observed raw. The remaining frozen REQUIRED conditions, including shift_completion, same_minute_replay and paused_midnight, are not supplied by historical union. The final checker must remain failed if any required condition is missing.

The earlier text is retained as historical planning context:

S5 — pending implementation, not a native run plan

No S5 executable driver or native receipt is supplied by this partial repair. Do not start the once-only Q native/full batch from this directory.

The four required cases remain unchanged:

1. Real mana depletion, NO_POWER, no charge/progress, reservation retained, recovery and one completion. S4's disconnect_nodes is not depletion. The existing cafeteria fixture has an operating kitchen prerequisite; global depletion can select RELATED_FACILITY_INACTIVE first. A driver must observe that distinction rather than relabel the reason or mutate power availability directly. A separate combined-cause case must prove the approved RELATED-before-NO_POWER reason priority.
2. Actual accident or raid destruction during completion: valid minute-N sale; next-minute failure/release for unfinished affected work; compare command destruction before completion. S3 B4a/B4b use a registered event with SET_FACILITY_STATE and must not simply be renamed an actual raid.
3. 00:00 completion: retain the S2a paid-midnight setup and full typed snapshots, add phase observations proving prior-day settlement/upkeep precede the sale, and independently join the sale to the new accounting day. Retain replay/no-double-settlement proof.
4. Command exit versus AI planned exit at completion: retain actual public commands and guest/service/reservation/ledger joins. Observe the AI DECIDING predicate after completion, without forcing guest state or treating a due exit time as universal active-service cancellation.

All prices, durations, resource rules and thresholds come from the selected Q definitions and current public runtime. No new original values or rules are supplied here. These notes identify implementation work; they are not engine evidence and confer no condition credit.