Every ticket as the root of its provenance graph — goal, patch, approval, runs and (as the ticket-class build lands) mandate, manifestations, guarantees, journey, artifacts and children. Rendered through the <<<TicketGraphList>>> tag.
Implement the harness each rewrite ticket runs. INPUT: a module + its behavioral exemplars (#200/criteria_miner). STEPS: (1) generate >=5 candidates via distinct model/approach configs (route through litellm #184 so the model set is config, not code); (2) GATE each on behavioral equivalence — discard failures, regenerate to keep >=5 passing; (3) PERFORMANCE benchmark each on a shared harness, capturing runtime/mem/allocs via #203 telemetry into a scores table; (4) ALL-AGENT PANEL — N independent judges (diverse models) each score every candidate on {config-completeness, efficiency(from bench), readability}, blind to each other; (5) aggregate (median, flag disagreement), rank, SELECT the winner, list graftable ideas from runners-up; (6) write the per-module deep-dive doc + record scores to a rebuild-scores table (queryable: best/worst by criterion across the whole forge2 rebuild). ACCEPTANCE: run on one real module -> 5 equivalence-passing candidates, a perf table with real numbers, a panel scorecard per criterion, a selected winner with rationale, and the deep-dive doc. Reuse #185 tournament code, #184 litellm, #203 telemetry, #200 exemplars.
patch
none
approval
unapproved
runs
no runs recorded
mandate (clauses)
not yet recorded — lands with the ticket-class build
manifestations
not yet recorded — lands with the ticket-class build
guarantees
not yet recorded — lands with the ticket-class build
journey (blunders & successes)
not yet recorded — lands with the ticket-class build
artifacts (forge)
not yet recorded — lands with the ticket-class build