ScoreBenchagent benchmark console

ScoreBench

vliw · without-indices

Strategy comparison Objectivecycles 0runs

Best score trajectory · without-indices

Show first
Limits each run to candidates submitted during its first selected hours of active work.
4h
Run recency
Highlights runs started within this lookback; older runs remain visible but muted.
All
best so far scored candidate x failed or rejected candidate (on its run's best-so-far line) colour + line style = variant
X axis

Resource expenditure · without-indices

cumulative tokens promote (new best) x failed candidate colour + line style = variant
X axis
Y axis

Strategy summary

Generated from SQLite, immutable bundles, and candidate metadata.
Top: score trajectory per variant on the selected score axis. Bottom: cumulative token or estimated cost expenditure on its own axis. Promote markers show submissions that produced a new best; x marks failed candidates. This dashboard is generated from the harness ledger; middleware-observed timestamps, scores, statuses, artifacts, and hashes remain the source of truth.