Upload README.md with huggingface_hub

README.md

@@ -49,25 +49,29 @@ set turned out to be a **methodology finding**, not a clean specificity control
 
 ## Review track — a methodology finding, not a leaderboard
 
-We ran five models on 24 buggy + 22 "post-fix" gameplay methods, expecting to rank them by recall
+We ran seven models on 24 buggy + 22 "post-fix" gameplay methods, expecting to rank them by recall
 and specificity. We are **not** publishing a balance ranking — here's why.
 
 | Model | Recall (real bug flagged) | Flag rate on "post-fix" methods |
 |---|---|---|
 | GPT-5.5 | 100% | 91% |
+| Opus 4.8 | 100% | 100% |
 | Qwen3-Coder (open, no context) | 92% | 77% |
 | Gemini 2.5 Pro | 67% | 82% |
-| Claude Sonnet 4.5 | 46% | 82% |
 | Qwen3-Coder + project context (RAG) | 67% | 18% |
-
-A naive harmonic-mean "balance" would rank GPT-5.5 last — and that is wrong. We hand-read every flag
-GPT-5.5 raised on the "post-fix" methods: they are overwhelmingly **real, reasonable review issues**
-(unchecked return values, missing null guards, non-atomic reward grants, missing idempotency,
-`Resources.Load` null, `OnDestroy` on scene-unload), not false positives — and it returned `CLEAN`
-on the two methods with no residual issue. The ground truth is the problem: a "post-fix" method is
-the version where *one* bug was fixed, not a method with *zero* issues, so "specificity" punishes the
-most thorough reviewer. Binary recall/specificity is the wrong instrument; the right one is a
-per-flag verdict (real / nitpick / false-positive). See the
+| Gemini 3.1 Pro | 62% | 55% |
+| Claude Sonnet 4.5 | 46% | 64% |
+
+A naive harmonic-mean "balance" would rank GPT-5.5 and Opus 4.8 last — and that is wrong. We
+hand-read their flags on the "post-fix" methods, then ran a convention-aware per-flag judge
+(cross-checked by two independent judges) across all models: the thorough frontier models' flags are
+**predominantly real review issues** (unchecked return values, missing null guards, non-atomic reward
+grants, missing idempotency, `Resources.Load` null, `OnDestroy` on scene-unload), not false positives.
+The ground truth is the problem: a "post-fix" method is the version where *one* bug was fixed, not a
+method with *zero* issues, so "specificity" punishes the most thorough reviewer. Binary
+recall/specificity is the wrong instrument; the right one is a per-flag verdict — but two independent
+judges disagree enough (46–88% agreement) that absolute precision needs human gold labels, which is
+why only recall + flag-rate are shown here. See the
 [repo README](https://github.com/leegod/unitybench). Treat this as a case study in why
 bug-detection benchmarks are hard.