Workspace.
Every successful call to a/v1/discovery/* endpoint —sweep, feature search, or factor mining — lands a JSON leaderboard on the server. This page lists the most recent runs bystarted_at with the top entry's score and the panel dimensions used. Click a run id for the full leaderboard + originating request.
For the build-time fixtures of ADR-0040 agent runs (curated + deterministic), see /agents/runs. For features and strategies surfaced by these runs, see/features and/strategies.
recent discovery runs
| run id | endpoint | objective | top score | entries | panel | elapsed | started |
|---|---|---|---|---|---|---|---|
| — | — | — | — | — | — | — | — |
| — | — | — | — | — | — | — | — |
| — | — | — | — | — | — | — | — |
how this is wired
The page polls/v1/discovery/runs every 60 seconds with ?limit=50. The server lists files by reverse-lexicographic filename — names embed the run'sstarted_at as a prefix so the sort is O(N · log N) with no per-file stat call.
Each row links to/workspace/run/?id={run_id}, which fetches/v1/discovery/runs/{run_id}and renders the full leaderboard plus the verbatim originating request — useful for reproducing a run from the CLI without re-typing the JSON body.
For the operator:set ALPHAKERNEL_DISCOVERY_RUNS_PATHon the API process to enable persistence. When unset, the endpoint returns enabled: false + empty runs and the ribbon above reads "persistence not configured." The discovery endpoints themselves still work — they just don't write anything to disk.