Skip to content

Move the non-convergence round cap inside the gate #7

Description

@Blakeolson21

Session 9 measured repeated non-converging review loops (observed round sequences 7/10/9, 6/8/6, 12/11/9, 14/9/10, 1/3/2/2/3/2/3). Reviewers reliably surface new siblings of the same defect class every round, so without a termination contract a gate can consume unlimited rounds and unlimited seats. Today the lead improvised the stop rule by hand: at 5+ rounds, or the same finding class recurring across 3+ rounds, stop funding rounds, park the run, and file a redesign ticket listing the current findings as acceptance seeds (precedent: MO#296, MO#297, MO#298, MO#299, no-slop#6).

Build it into the pipeline: (1) a configurable non-convergence threshold (rounds and recurring-class count) evaluated by the executor; (2) on breach, terminate the run with a distinct outcome (e.g. 'parked-nonconverging') rather than another fix round; (3) emit the accumulated findings in a form suitable for pasting into a redesign ticket; (4) never require a human decision to stop a diverging loop. Pairs with the seat-blocked outcome work on codex/seat-blocked-outcome (no-slop#6).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions