SycoBench-600 v1.0.1: Inspect Evals Register listing

SycoBench-600 is now listed as an external evaluation in the Inspect Evals Register.

• Register entry: https://github.com/UKGovernmentBEIS/inspect_evals/tree/main/register/sycobench-600
• Merged PR: UKGovernmentBEIS/inspect_evals#1901
• Inspect Evals merge commit: 6a1edbccc4fc25b4c22132c9d70f1aac0e72bb2b
• Registered task: sycobench_600
• Pinned SycoBench commit in the register: 5219abd

This release also adds public adoption notes in docs/ADOPTION.md and links the register entry from the project README.

Validation for this release commit:

• python -m ruff format .
• python -m ruff check .
• python -m pytest -q -> 15 passed

SycoBench-600 v1.0.0: ACL Findings camera-ready release

Stable release of SycoBench-600, the ACL 2026 Findings benchmark for measuring sycophancy and correction selectivity in LLM assistants.

This release pins the exact artifacts behind the camera-ready paper.

Included

• Dataset: 600 questions across 8 balanced domains (272 unique stems)
• Raw model logs for 7 models in results/raw_camera_ready/
• Generated tables and figures in build/camera_ready/
• Camera-ready paper source and PDF in paper/
• Reproducibility scripts: scripts/run_eval.py, scripts/validate_paper_full.py
• Test suite covering release artifacts

Validation

• pytest: 10 passed
• Paper consistency check: 0 errors
• All Table 3 and Table 4 numbers match build/camera_ready/tables/main_results.csv
• Headline figures regenerated and verified

Licensing

• Code: MIT
• Dataset, raw logs, generated tables and figures: CC BY 4.0
• Paper and camera-ready PDF: ACL / CC BY 4.0

Paper

OpenReview: https://openreview.net/forum?id=ijR52T6rdD

Citation

See CITATION.cff or the README citation block.