Merged-state gate + bounded intent reapplication

Parked; reopen only after the merged-state fixture test
Provisional idea; no final shortlist has been approved.Status as of 2026-10-15 09:27 UTC.

01 · PROBLEM

Land the combination safely; when a candidate breaks the combined state, regenerate it on the fresh base instead of hand-resolving.

02 · HYPOTHESIS

Candidate held unchanged; runner tests canonical+candidate; on failure, the original versioned task contract is re-run by an agent in a new fork from the new base; scope drift is diffed against the first attempt; human approves if intent/test policy changed; receipt records parent attempt.

03 · EVIDENCE SO FAR

LATESTNot tested. The agents gathered reports of the pain behind this idea and scored it on paper only; no experiment has been run on it. Cited pain: E-X004, E-X008 (60 branches on stale base), E-C101 (would rather rewrite than resolve), E-C103; fixture research/codex/local-validation.md.website/projects.json

04 · NEXT FALSIFICATION TEST

Not yet scheduled. Smallest demo on record: two agents; A lands; B's clean-merge breaks; platform re-runs B's contract on fresh base, passes, lands with receipt.

StatusParked; reopen only after the merged-state fixture test
We would drop or park it ifOn tasks with hidden acceptance criteria or flaky tests, if reapplication changes scope without detection in >1 of 5 runs, or median loop >15 min, kill as a product (keep as feature).

05 · PUBLIC SOURCES