QuantCheck 0.1.0

Immutable historical baseline of the v0.1 rebuild, frozen before v0.2 development.

Four fault families with manifest-blind detection (look-ahead, unit drift, duplicate observations, revision overwrite), a 124-case held-out benchmark at reserved seeds 1000-1009, public/private artifact separation, a six-command CLI, and a read-only presentation layer.

Held-out result

Identifier	Value
Release candidate	relc_2c6e945a71b85b39
Benchmark	bench_403a85e506ff66ea
Aggregate	agg_571aae0b7c60a4a5

124 configured, 94 successful, 30 failed, 0 incomplete. 130 injected faults, 208 findings, 130 TP, 0 FN, 78 FP. Precision 0.625, recall 1, F1 0.769..., FPR 0.0728.... Research changed 90/90; replay restored 90/90.

The honest part

The 30 failures are no_eligible_targets in three profile/severity cells that have no target on the reviewed fixture. All 78 false positives are cross-detector findings under strict primary-label scoring. Neither was tuned away, and both are visible in the case matrix, the statuses, and the aggregate.

Verification

• CI green on the tagged source, Ubuntu x86_64, every workflow step: runs
  31293937904 and
  31295303799.
  Those runs are on the commits either side of the tag; src/, tests/, pyproject.toml,
  uv.lock, and .github/ are byte-identical across all three, so they cover exactly this
  tag's buildable tree. No run is triggered by the tag itself — the workflow fires on pushes
  to main only.
• CHECKSUMS.md covers 88 committed files; release_freeze.json pins 67 frozen files, both fixture hashes, and the scientific configuration.
• The attached wheel and sdist rebuild byte-identically from the tagged source:
  • ff1fff1880796803cf9454c1199c78e3e71bd41889301eb728e75a798dbbef91 quantcheck-0.1.0-py3-none-any.whl
  • cfc1497d171bc23c01e9558e7f8bb83f5ccfb4ec904019bb05b0b0be43fe75f7 quantcheck-0.1.0.tar.gz
• 1595 tests pass, including under PYTHONHASHSEED 0, 1, and 987654.

Install

This is not a package-index release. pip install quantcheck does not resolve. Install from the attached wheel:

pip install quantcheck-0.1.0-py3-none-any.whl

Limits

Synthetic reviewed fixture, four narrow fault subtypes, controlled comparisons, day-level availability. Not a backtest and not production certification. Produced on macOS arm64; Windows is untested. See docs/LIMITATIONS.md and docs/RELEASE_CHECKLIST.md.