Datasets:

---

perchscan

/

benchmark-results

Like 2

Follow

Perch Semantic Linter 1

Benchmark GPT-6 Luna through the native Decisions API on all 9,506 public and private method versions across bugs, security and semantic lint. Preserve the frozen language questions, supplied method context and alert floors. Include returned probabilities, failures, latency and input-token costs. Publish public predictions and only aggregate private results.

Validation: all dataset hashes and outcome counts verified; confusion counts and question scores checked; runner tests and reproducible tables passed.