Skip to content

Releases: theaiautomators/jev-arena

Recorded video assessments — September 2026

Choose a tag to compare

Two recorded assessments from the Jev Arena video, with interactive offline viewers and supporting evidence.

Assessment Recorded scope Downloads
Arena Full v2 13 profiles; 99,723 static records; timing, task analysis and workflow replays arena-full-v2.html or arena-full-v2.zip
ABCD support decisions Five profiles; 12,720 records; full handbook versus retrieved policy abcd-v1.html or abcd-v1.zip

Download an HTML file and open it in your browser, or extract a ZIP and open report.html. No installation, GPU, credentials or model calls are needed to view the results.

The ZIPs include machine-readable results, sanitized prediction records, relevant methodology, limitations and license notices. Full v2 also includes the original source snapshot. Raw third-party case text and private run databases remain local. Planning files, editorial drafts and superseded run reports are excluded.

Result data, prediction records and original source/freeze evidence match the earlier local exports byte for byte. Scores and reference labels have not changed. These assessments have different tasks and denominators; their results are kept separate. Automated audits are not human semantic validation, and these fixed deployment profiles do not establish a universal winner.

Full v2 findings · ABCD findings · Methods and verification

Checksums for the four report downloads are in SHA256SUMS.txt and the repository's results/manifest.json.

Validation: 82 CPU tests and four navigation tests passed locally. A source-only checkout passed 76 tests with six expected skips for private assessment data, installed locked frontend dependencies, and built successfully. Public-tree/link checks and export content/credential scans passed. No model inference or hosted judging was repeated.