A repeated SWE Verified critique again points to a 5%–10% invalid-sample rate
xeophon · x · 2026-07-25
This is essentially a repeat pointer to the same SWE Verified audit: the dataset was inspected, and the author argues that a non-trivial share of samples may be bad or underspecified.
The attached excerpt repeats the core findings:
- 85 hard issues were examined.
- 14 invalid samples were identified in the manually reviewed subset.
- The implied invalid-sample rate lands around 5%–10%.
It does not add new facts beyond the earlier post, but it reinforces the benchmark-quality critique.
Related event: SWE-bench Audit Reveals Up to 10% Invalid Samples(2 posts)→
More from Research
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
- Behind the EMNLP keynote: LMs as simulated learners to generate child language acquisition hypotheses — najoungkim · 2026-10-26
- The paper behind the EMNLP 2026 keynote on bigram 'simulated learners' — najoungkim · 2026-10-26