CAP-02: the one-round re-arm cap never reaches the trigger that is always blocking #213

New issue

Closed

opened 2026-08-18 10:24:03 +00:00 by john · 0 comments

john commented

2026-08-18 10:24:03 +00:00

Owner

Requirement CAP-02, seeded at the v3.5.3 close. Related: #181, the same scoping question one layer out.

The one-round re-arm cap lives only in references/triage-gate.md, which the risk_surface fire sites never load. task.md and debug.md never mention it, execute.md restates the blocking arm inline UNCAPPED, and every "RE-READ triage-gate.md" instruction in the tree is scoped to the adjudicated arm.

So the bound misses the one trigger that is blocking at every stakes level.

Second half, and it needs a decision rather than a fix: the cap is scoped per correlation id, so a five-plan phase gets ONE round across all five plans. That may be the intended budget. It has never been stated either way.

Fix direction: the cap reaches the blocking fire sites, and its scope is written down deliberately.

Requirement CAP-02, seeded at the v3.5.3 close. Related: #181, the same scoping question one layer out. The one-round re-arm cap lives only in `references/triage-gate.md`, which the `risk_surface` fire sites never load. `task.md` and `debug.md` never mention it, `execute.md` restates the blocking arm inline UNCAPPED, and every "RE-READ triage-gate.md" instruction in the tree is scoped to the `adjudicated` arm. So the bound misses the one trigger that is `blocking` at every stakes level. Second half, and it needs a decision rather than a fix: the cap is scoped per correlation id, so a five-plan phase gets ONE round across all five plans. That may be the intended budget. It has never been stated either way. **Fix direction:** the cap reaches the blocking fire sites, and its scope is written down deliberately.