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.
Research: role->model fit (Gemma 3 dispatch/judge vs Qwen2.5-Coder worker) benchmarked on REAL agent tasks
goal
Benchmark the pulled models on the ACTUAL agent tasks, not generic evals: (a) DISPATCH — given a ticket.queue JSON, emit a valid run/skip/stop plan; (b) WORKER — given a build ticket + repo excerpts, write a patch to a deliverable file; (c) ADVERSARY/JUDGE — given an artifact, emit a pass/fail verdict + findings. MEASURE per model and size (gemma3:4b/12b, qwen2.5-coder:7b/14b): task success rate, valid-structured-output rate, latency, and context-window adequacy for our ticket bodies + repo context. OUTPUT: a role->model->size mapping, and an explicit list of tasks where a local model is too weak and Claude-as-tool-worker is the fallback. Spawns a build ticket.
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