# Owned AI and Intel migration increment

This increment adapts the owner's Escapists AI and Escapists Intel extensions to the independent successor. Original base classes, addon dependencies and resources are not imported. Exact source hashes and destinations are recorded in [migration provenance](migration-provenance.json).

## Player-visible changes

- Administrators can use `/esc intel show` to inspect both probabilities, `/esc intel reroll` to apply the current carrying probability to registered living enemy AI, and `/esc intel reroll show` to inspect progress and failures. Existing carrying/reveal commands remain available.
- Each reroll captures its probability and AI cohort at start. Later settings changes and newly registered enemies do not change the batch. Native inventory work is bounded, concurrent rerolls are rejected, and objective folders, recovered intel and existing location knowledge are preserved.
- Retired AI folders are spent and tracked for round cleanup. They are not deleted through a player's inventory or a delayed loose-item cleanup callback. Failed native operations are reported rather than silently creating replacement documents.
- Team map notifications report rough search areas or identified locations. Several discoveries are coalesced, rough clues do not leak location names, and updates do not replace the existing phase briefing.
- Enemy squads report different currently visible contacts fairly. Per-squad history contains at most 64 identities and advances only when a radio uplink is queued. This changes reporting priority; radio delivery still controls HQ knowledge and physical orders.

## Verification

Focused regressions exercise the actual production policy bodies with native boundaries adapted for the harness. These cover admin dispatch and server authorization ordering, spent-ledger bounds, inventory reroll lifecycle, notification coalescing/round cleanup and fair HQ report selection.

The existing Chat Workbench victory fixture now checks administrator admission/refusal, captured 100 percent probability after a live change to zero, native guard-folder insertion and zero-percent removal, preserved player-held clue identities and discovered radio, and ownership of retired folders for round cleanup. The [combined native run](evidence/native-2026-09-19-225422-owned-ai-intel/report.json) passes two escape/restart rounds with 186 assertions and zero captured script/runtime/engine errors. Each round completes a captured 100-percent batch for 36 AI followed by a zero-percent batch for 40 AI, with no failed or skipped operations. Source remained unchanged through the run and private process cleanup was confirmed. [All repository checks pass](evidence/native-2026-09-19-225422-owned-ai-intel/repository-checks.txt).

The [first native attempt](evidence/native-2026-09-19-225016-intel-removal-refusal/report.json) is retained: one removal refusal ended its batch prematurely. The fix restores the owned original's bounded retry behavior; regressions cover both transient and persistent refusal without consuming rejected documents.

Remote admin requests, multi-client message presentation and simultaneous player/native-inventory races remain multiplayer playtest work. The frozen `.build/playtest-2026-09-19` candidate has not been updated by these source edits.