Releases: neuron7xLab/bsff
Releases · neuron7xLab/bsff
Release list
v0.4.0
BSFF v0.3.0 — conjunction gate + SLSA provenance
BSFF v0.3.0
Falsification engine
- Frequentist-AND-Bayesian conjunction gate.
SURVIVEDnow requires both a rank-order surrogate rejection (p ≤ α) and effect-size corroboration (BF10 ≥ threshold; default 3, strict 10). Fail-closed — it only ever demotes toUNSUPPORTED. Closes a measured false-positive hole: the rank-order p-value is anti-conservative for strongly autocorrelated linear-Gaussian nulls (finite-N IAAFT bias, Kugiumtzis 2002). - Measured operating characteristic. A labelled ground-truth battery (deterministic chaos vs linear-Gaussian / white-noise nulls) is recomputed each CI run. Power held at 1.000 on Hénon/logistic; null false-positive rate restored to ≤ α on every class. Reproducible via
tools/calibrate_operating_characteristic.py; full result inartifacts/operating_characteristic.json.
Supply chain
- All GitHub Actions hash-pinned to commit SHA;
renovate.json+ Dependabot cooldown/grouping keep them fresh. scorecard.yml/release-artifact.ymltoken permissions moved to job level (Token-Permissions 10; Scorecard publishing unblocked).zizmorstatic workflow audit in CI;persist-credentials: falseon all checkouts.- SLSA Build L3 provenance (
bsff-v0.3.0.intoto.jsonl) attached to this release; subject digests cryptographically bind the published wheel and sdist.
Verification
- 90 tests; coverage ~87%. All contract gates + zizmor green.
- Provenance subjects verified byte-for-byte against the release artifacts.
This kernel reports whether a claim survived the configured falsification attacks. It does not prove BCI claims, is not externally validated against TISEAN, and is not regulatory validation.
BSFF v0.2.0
Falsification-first BCI/EEG claim kernel. Reference covariance validation, MIAAFT benchmark matrix, pipeline.run() alias, deterministic SURVIVED/REFUTED/UNSUPPORTED trichotomy, MOABB example, JOSS scaffold. 38/38 tests, CI green, tokenless PyPI publishing wired (Trusted Publishing).