Skip to content

Latest commit

 

History

History
136 lines (111 loc) · 9.84 KB

File metadata and controls

136 lines (111 loc) · 9.84 KB

Evaluation history

Every recorded eval run in evals/runs/, oldest first, on one page: per-suite pass rates, cost, duration, and each prompt-version bump. The run directories themselves are a local, gitignored archive; this rendered page (and the SVG) is the committed artifact. Regenerate with make report (or python -m evals.history).

Read within an instrument, never across. Mock/offline runs are scored by deterministic checks against a mock answer model; live runs call a real answer and judge model. They are different instruments and measure different things, so the chart and tables below group runs by instrument (mode + offline/live) and never plot mock and live scores on the same series. Smoke and full runs differ in sample size too. Only the trajectory within one instrument is a like-for-like comparison.

Pass-rate trajectory per instrument

full · offline (mock) — 6 run(s)

Run (UTC) Overall edge_cases freshness groundedness multilingual refusal conversation cross_agency Est USD Duration
2026-06-12T06:03:12+00:00 11.4% 0.0% 0.0% 0.0% 7.1% 50.0% n/a 0.0s
2026-06-12T06:07:25+00:00 11.4% 0.0% 0.0% 0.0% 7.1% 50.0% n/a 0.0s
↑ prompt bump: answer_user v1→v2, judge_helpfulness v1→v2, system v1→v4
2026-06-13T00:24:22+00:00 10.7% 0.0% 0.0% 0.0% 10.0% 47.4% n/a 0.0s
↑ prompt bump: answer_user v2→v3, system v4→v6
2026-06-30T17:18:26+00:00 9.1% 0.0% 0.0% 0.0% 9.5% 47.4% 0.0% 0.0% $0.0000 0.1s
↑ prompt bump: answer_user v3→v4, system v6→v7
2026-06-30T17:27:28+00:00 8.7% 0.0% 0.0% 0.0% 9.1% 47.4% 0.0% 0.0% $0.0000 0.1s
2026-06-30T17:28:31+00:00 8.7% 0.0% 0.0% 0.0% 9.1% 47.4% 0.0% 0.0% $0.0000 0.1s

smoke · offline (mock) — 2 run(s)

Run (UTC) Overall edge_cases freshness groundedness multilingual refusal Est USD Duration
2026-06-12T06:07:34+00:00 8.0% 0.0% 0.0% 0.0% 0.0% 40.0% n/a 0.0s
2026-06-12T06:14:41+00:00 8.0% 0.0% 0.0% 0.0% 0.0% 40.0% n/a 0.0s

full · live — 30 run(s)

Run (UTC) Overall edge_cases freshness groundedness multilingual refusal conversation cross_agency sensitivity stretch_tagalog Est USD Duration
2026-06-12T06:39:41+00:00 72.9% 83.3% 100.0% 87.5% 14.3% 85.7% n/a 494.7s
↑ prompt bump: system v1→v2
2026-06-12T06:53:08+00:00 84.3% 83.3% 87.5% 93.8% 71.4% 85.7% n/a 488.2s
2026-06-12T07:06:48+00:00 92.9% 88.9% 100.0% 100.0% 92.9% 85.7% n/a 534.6s
↑ prompt bump: system v2→v3
2026-06-12T07:17:04+00:00 92.9% 88.9% 100.0% 100.0% 92.9% 85.7% n/a 482.0s
↑ prompt bump: system v3→v4
2026-06-13T00:01:47+00:00 92.9% 100.0% 87.5% 100.0% 78.6% 92.9% n/a 480.5s
↑ prompt bump: answer_user v1→v2, judge_helpfulness v1→v2
2026-06-13T00:11:12+00:00 95.7% 94.4% 87.5% 100.0% 92.9% 100.0% n/a 486.9s
2026-06-13T00:36:42+00:00 92.2% 92.9% 90.0% 92.3% 90.0% 94.7% n/a 728.7s
2026-06-13T00:50:43+00:00 93.2% 92.9% 90.0% 92.3% 95.0% 94.7% n/a 723.0s
2026-06-17T00:04:04+00:00 94.2% 96.4% 90.0% 92.3% 95.0% 94.7% n/a 797.5s
↑ prompt bump: system v4→v5
2026-06-17T00:46:39+00:00 93.2% 100.0% 100.0% 92.3% 85.0% 89.5% $1.4324 745.8s
↑ prompt bump: system v5→v4
2026-06-17T01:00:41+00:00 95.1% 96.4% 100.0% 92.3% 95.0% 94.7% $1.4538 772.3s
2026-06-17T01:38:37+00:00 94.5% 96.4% 100.0% 92.3% 95.0% 94.7% 83.3% $1.5837 839.3s
2026-06-17T02:22:39+00:00 96.6% 97.0% 100.0% 96.6% 100.0% 94.7% 83.3% $1.6843 898.0s
2026-06-17T02:39:35+00:00 97.5% 100.0% 100.0% 96.6% 100.0% 94.7% 83.3% $1.6869 906.4s
↑ prompt bump: system v4→v5
2026-06-17T03:11:14+00:00 94.9% 100.0% 100.0% 93.1% 95.2% 94.7% 66.7% $1.6856 922.6s
↑ prompt bump: answer_user v2→v3, system v5→v6
2026-06-30T04:04:35+00:00 95.8% 93.9% 100.0% 100.0% 90.5% 94.7% 100.0% $1.6877 838.2s
2026-06-30T04:35:31+00:00 95.8% 100.0% 100.0% 100.0% 85.7% 100.0% 66.7% $1.7021 850.3s
↑ prompt bump: system v6→v7
2026-07-11T21:20:11+00:00 94.5% 97.0% 100.0% 100.0% 95.2% 92.3% 66.7% 66.7% $1.7814 216.0s
2026-07-11T21:24:20+00:00 94.5% 100.0% 90.0% 96.6% 95.2% 96.2% 66.7% 66.7% $1.7730 176.1s
↑ prompt bump: answer_user v3→v4, judge_groundedness v1→v2, judge_helpfulness v2→v3, system v7→v9
2026-07-11T21:52:44+00:00 73.1% 81.2% 70.0% 93.1% 81.8% 85.3% 60.0% 66.7% 63.3% 0.0% $3.1030 368.2s
↑ prompt bump: answer_user v4→v6
2026-07-11T22:03:46+00:00 80.1% 85.4% 80.0% 100.0% 100.0% 94.1% 60.0% 66.7% 70.0% 0.0% $3.1337 292.9s
2026-07-11T22:10:48+00:00 79.6% 85.4% 100.0% 96.6% 90.9% 91.2% 70.0% 66.7% 70.0% 0.0% $3.1468 317.0s
2026-07-11T23:03:23+00:00 84.6% 87.5% 100.0% 100.0% 90.9% 91.2% 70.0% 66.7% 73.3% 46.7% $3.1475 7.1s
2026-07-11T23:05:36+00:00 86.6% 87.5% 100.0% 100.0% 90.9% 91.2% 70.0% 66.7% 73.3% 73.3% $3.1468 15.9s
↑ prompt bump: answer_user v6→v7, system v9→v10
2026-07-11T23:13:18+00:00 90.5% 93.8% 90.0% 93.1% 100.0% 94.1% 60.0% 100.0% 76.7% 100.0% $3.1241 302.6s
↑ prompt bump: system v10→v11
2026-07-12T04:27:28+00:00 94.5% 87.5% 100.0% 100.0% 95.5% 97.1% 70.0% 100.0% 100.0% 100.0% $3.1875 324.8s
↑ prompt bump: system v11→v12
2026-07-12T04:40:07+00:00 94.0% 93.8% 90.0% 100.0% 95.5% 97.1% 80.0% 66.7% 96.7% 86.7% $3.2106 312.3s
↑ prompt bump: system v12→v10
2026-07-12T05:00:18+00:00 95.5% 95.8% 90.0% 93.1% 100.0% 100.0% 80.0% 100.0% 93.3% 100.0% $3.1580 10.5s
2026-07-12T05:01:17+00:00 95.5% 95.8% 90.0% 93.1% 100.0% 100.0% 80.0% 100.0% 93.3% 100.0% $3.1580 0.7s
2026-07-12T07:08:10+00:00 90.0% 93.8% 90.0% 89.7% 86.4% 100.0% 70.0% 100.0% 76.7% 100.0% $3.1798 214.6s

suite:freshness · offline (mock) — 1 run(s)

Run (UTC) Overall freshness Est USD Duration
2026-06-17T00:33:30+00:00 0.0% 0.0% $0.0000 0.0s

smoke · live — 1 run(s)

Run (UTC) Overall edge_cases freshness groundedness multilingual refusal Est USD Duration
2026-06-17T01:18:01+00:00 96.0% 100.0% 100.0% 100.0% 83.3% 100.0% $0.3485 185.1s

suite:conversation · offline (mock) — 1 run(s)

Run (UTC) Overall conversation Est USD Duration
2026-06-17T01:24:28+00:00 0.0% 0.0% $0.0000 0.0s

suite:edge_cases · live — 3 run(s)

Run (UTC) Overall edge_cases Est USD Duration
2026-06-30T04:10:13+00:00 93.9% 93.9% $0.5066 260.2s
↑ prompt bump: answer_user v3→v2, system v6→v5
2026-06-30T04:15:38+00:00 100.0% 100.0% $0.5062 263.2s
↑ prompt bump: answer_user v2→v3, system v5→v6
2026-06-30T04:20:57+00:00 100.0% 100.0% $0.5110 260.4s

suite:edge_cases · offline (mock) — 1 run(s)

Run (UTC) Overall edge_cases Est USD Duration
2026-06-30T17:27:47+00:00 0.0% 0.0% $0.0000 0.0s

suite:freshness · live — 2 run(s)

Run (UTC) Overall freshness Est USD Duration
2026-07-11T21:54:41+00:00 100.0% 100.0% $0.1556 19.0s
↑ prompt bump: answer_user v5→v6
2026-07-11T22:05:26+00:00 100.0% 100.0% $0.1588 19.9s

suite:multilingual · live — 3 run(s)

Run (UTC) Overall multilingual Est USD Duration
2026-07-11T21:55:32+00:00 81.8% 81.8% $0.4252 50.2s
2026-07-11T21:57:24+00:00 86.4% 86.4% $0.4016 19.0s
↑ prompt bump: answer_user v5→v6
2026-07-11T21:58:43+00:00 100.0% 100.0% $0.4060 43.7s

suite:stretch_tagalog · live — 2 run(s)

Run (UTC) Overall stretch_tagalog Est USD Duration
2026-07-11T23:07:14+00:00 80.0% 80.0% $0.2331 27.4s
2026-07-11T23:08:04+00:00 100.0% 100.0% $0.2331 0.1s