Research documentation

Raw Evidence Releases

The original public release files remain the evidence artifacts. HTML documentation provides navigation; it does not replace the manifest, retained observations or verification method. Each manifest records the released file inventory and SHA-256 digests. Hash verification establishes integrity relative to that manifest, not publisher authenticity or independent generation reproduction.

Residency and executors

Manifest and SHA-256 inventory · Release index and edition history

Paired structural comparison

Manifest and SHA-256 inventory · Release index and edition history

IFEval standalone benchmark

Manifest and SHA-256 inventory · Release index and edition history

Performance JSON extracts

Historical editions

The performance release contains curated aggregates; its full raw-run bundle is not included. The paired and IFEval ZIP files have their own retained-response and rescoring scope. Verification boundaries.