A
Aura BenchmarkArchimedes Alternator · daily · unedited
updated 2026-09-16 GitHub ↗

Full-run pass rate apples-to-apples · 320 questions

Run explorer click to overlay

Click a run to add or remove it from the tier comparison below. Full runs are on by default.

Tier comparison every question tier 1–60 · where runs gain or lose

Where runs lose points avg rate per tier bucket · darker = weaker

Difficulty curve avg by tier · full runs

Alternator competence latest snapshot

Verification catch verifier catching errors

Graphify · relation graphs of the benchmark

same data, different geometry · hover anything · generated locally, zero dependencies

Arc — sessions on a line, arcs tie same-model runs · arc height = distance in time · width = run size

Chord — models ↔ tier buckets

ribbon width = questions answered

Flow — full runs → verdicts

correct · partial (est.) · incorrect per run

Radial — every run on record

spoke length = pass rate · ○ = full run

All sessions every recorded run · sortable