Library writes cannot be live-verified; most game writes have no current evidence #6

Closed
opened 2026-10-13 05:49:16 -04:00 by vinny · 1 comment
Owner

docs/STATUS.md lists 66 game writes. Their current status:

  • 3 verified on the current tuple
  • 39 "verified on another runtime only"
  • 24 never verified, including groups.*, loadouts.set_loadout, navigation.fly_to / go_to / patterns, navigation.clear_route, zones.set_no_navigation, engagement.fire_weapon and contacts.set_posture

Controlled mode offers them, which is correct by AGENTS.md: authorization is separate from verification history. But RuntimeVerifier refuses library operations (src/cmonet/execution/runtime_verifier.py, around line 123; docs/TOOL_LIBRARY.md), so there is no way to record evidence for them.

Proposal:

  1. Add a library verification path in which the tester's read-back tools decide.
  2. Run one qualification session on a disposable save.

The STATUS "current" tuple is also AI TEST SCENARIO, while the operator now plays Iran vs WORLD (side Iran, 1,435 own units).

Found in the 2026-10-13 review (PR #1).

`docs/STATUS.md` lists 66 game writes. Their current status: - 3 verified on the current tuple - 39 "verified on another runtime only" - 24 never verified, including `groups.*`, `loadouts.set_loadout`, `navigation.fly_to` / `go_to` / patterns, `navigation.clear_route`, `zones.set_no_navigation`, `engagement.fire_weapon` and `contacts.set_posture` Controlled mode offers them, which is correct by AGENTS.md: authorization is separate from verification history. But `RuntimeVerifier` refuses library operations (`src/cmonet/execution/runtime_verifier.py`, around line 123; `docs/TOOL_LIBRARY.md`), so there is no way to record evidence for them. Proposal: 1. Add a library verification path in which the tester's read-back tools decide. 2. Run one qualification session on a disposable save. The STATUS "current" tuple is also AI TEST SCENARIO, while the operator now plays Iran vs WORLD (side Iran, 1,435 own units). Found in the 2026-10-13 review (PR #1).
Author
Owner

Done in #20 (merged at 503bed4, CMOnet 0.7.8).

  • tools/qualify_library.py verifies the library writes through CMOnet's product path. It uses a scratch controller and a scripted General: each write goes through the coordinator, and the tester's read-back decides the outcome.
  • tools/evidence.py import-library records the results at the library revision each preflight recorded.
  • Live run on 2026-10-13 (AI TEST SCENARIO, build 1900.20, side AI): all 60 library writes completed. 62 of 66 game writes are now Verified normal operation on the current tuple.
  • Report: docs/verification/library/library-writes-2026-10-13.json.
  • STATUS now says verified at an earlier adapter revision only when a library sync moves the revision on. It used to say "another runtime".
  • The four still unverified are CMOnet's own missions.set_refueling and logistics.refuel_from_missions, and the composites forces.mobilize and forces.strike_package.
Done in #20 (merged at 503bed4, CMOnet 0.7.8). - `tools/qualify_library.py` verifies the library writes through CMOnet's product path. It uses a scratch controller and a scripted General: each write goes through the coordinator, and the tester's read-back decides the outcome. - `tools/evidence.py import-library` records the results at the library revision each preflight recorded. - Live run on 2026-10-13 (AI TEST SCENARIO, build 1900.20, side AI): **all 60 library writes completed**. 62 of 66 game writes are now *Verified normal operation* on the current tuple. - Report: `docs/verification/library/library-writes-2026-10-13.json`. - STATUS now says *verified at an earlier adapter revision only* when a library sync moves the revision on. It used to say "another runtime". - The four still unverified are CMOnet's own `missions.set_refueling` and `logistics.refuel_from_missions`, and the composites `forces.mobilize` and `forces.strike_package`.
vinny closed this issue 2026-10-13 20:33:03 -04:00
Sign in to join this conversation.
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
vinny/CMOnet#6
No description provided.