Skip to content

v7.7.6 — gauntlet bundling fix: ship the benchmark JSON inside the wheel

Choose a tag to compare

View all tags
@github-actions github-actions released this 27 May 20:08
· 486 commits to main since this release
v7.7.6
fc4ad1e

styxx 7.7.6 — gauntlet bundling fix

Fixed

styxx gauntlet was broken on clean pip install. 7.7.5 shipped the runner code and CLI but the labeled benchmark JSON lived only at papers/consensus-hallucination/ in the source tree — not in the wheel's package data. Users on clean install hit FileNotFoundError. Caught by 7.7.5's own clean-env verification step.

Added

  • styxx/_data/darkcore_benchmark_2026_05_27.json — the benchmark, registered as package data. Verified present in the 7.7.6 wheel (49,436 bytes).
  • Updated load_benchmark() resolution order: (1) explicit path arg, (2) bundled package data, (3) source-tree fallback.
  • Regression test test_load_benchmark_package_data_exists.

Verify

pip install styxx==7.7.6
styxx gauntlet --method styxx.gauntlet:_majority_baseline_predict --task classification --name sanity-check

The majority-baseline correctly fails all three bars (0/3 passed) — confirming the gauntlet runs end-to-end on a clean install.

Why this is in the CHANGELOG

This is the fifth in-session falsification today — 7.7.5 was marked "verified" but the clean-install verification missed the runtime-feature path. Recorded honestly rather than retroactively cleaned.

🤖 Generated with Claude Code

Assets 4
Loading