From 7bca588b0f4cbcc3b13d8a4d8bfc845afade269b Mon Sep 17 00:00:00 2001 From: hiro-nikaitou Date: Wed, 23 Sep 2026 03:46:11 -0400 Subject: [PATCH] docs: sync LiveMath held-out test size to 125 Signed-off-by: hiro-nikaitou --- blog/gating-reflection-safe-updates/index.html | 2 +- docs/sleep/RESULTS.md | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/blog/gating-reflection-safe-updates/index.html b/blog/gating-reflection-safe-updates/index.html index 3a752d8c6..41a1f2e7b 100644 --- a/blog/gating-reflection-safe-updates/index.html +++ b/blog/gating-reflection-safe-updates/index.html @@ -1341,7 +1341,7 @@

Part III ยท SkillOpt-Sleep: an overnight optimization plugin

LiveMathematicianBench - 124 items + 125 items Multiple-choice label with choices shuffled per item Small, high-variance reasoning set that tests whether gains survive noise. diff --git a/docs/sleep/RESULTS.md b/docs/sleep/RESULTS.md index 441ffa28f..61f524bd2 100644 --- a/docs/sleep/RESULTS.md +++ b/docs/sleep/RESULTS.md @@ -22,7 +22,7 @@ shared wrapper function. | Benchmark | Held-out test | Scoring | |---|---|---| | SearchQA | 1,400 items | SQuAD exact-match vs gold | -| LiveMathematicianBench | 124 items | multiple-choice label (choices shuffled per item) | +| LiveMathematicianBench | 125 items | multiple-choice label (choices shuffled per item) | | SpreadsheetBench | 280 items | the agent's generated openpyxl code is **executed**, output workbook compared cell-by-cell to a golden file | **Targets:** GPT-5.5, GPT-5.4-mini, GPT-5.4-nano. **Modes:** validation-gated