CHECKING THE SIGNALS

Connecting the dots…

Independent signals for a moving target.

Independent. Unaffiliated with model providers.
Built to make uncertainty visible.

UNLINK

Model performance over time.

Independent evaluations. Original sources. Separate scores.

PERFORMANCE TRACKERS

LIVENERF · CLAUDE CODE

57.7%Latest panel accuracy

45Daily results

2026-11-07Latest result · source date

Daily score 95% confidence intervalHover over a point to inspect

Fixed question panel · pinned Claude Code · subscriptionDownload data ↓

MARGINLAB · CLAUDE CODE

Model period SWE-Bench-Pro · source model boundaries

76.6%Latest coding pass rate

6Daily results in this period

2026-09-29Latest result · source date

Daily score 95% confidence intervalHover over a point to inspect

283 scored cases across 6 daily resultsDownload this period ↓

MARGINLAB · CODEX

Model period SWE-Bench-Pro · source model boundaries

—Latest coding pass rate

—Daily results in this period

UnavailableLatest result · source date

Benchmark source temporarily unavailable.
Check the original source ↗

Source data unavailable

CROWD RADAR

Source data unavailable. View source ↗

Hourly snapshots of a rolling score, not separate hourly evaluations.

MODELREGRESSION

Source data unavailable. View source ↗

Source-defined composite scores. Latest result dates are shown below.

Different tests measure different things. Scores are not combined.Report an experience ↗