Hardening release focused on reproducible public artifacts, stricter model validation, and a trustworthy distribution path.
Added
- Self-identifying JSON reports with instrument version, schema versions, model SHA-256, horizons, and run configuration
- Public and packaged result schemas
- One deterministic regeneration command for the AI models, parity ledger, complete results bundle, and README figure
- CI freshness enforcement for generated and untracked artifacts
- Narrated ten-node tutorial and Python API quickstart
--version, documented exit codes,py.typed, and Python 3.13 coverage- OIDC-based PyPI release workflow with no long-lived API tokens
Hardened
- Direct self-satisfaction now raises
ModelError - Dependent outcome sources emit
ModelWarning - Explicit horizons are sorted and deduplicated
- OR-gap comparison reuses computed AND survival through one public implementation
- Public and packaged schemas are kept byte-identical by tests
Reproducibility
The committed AI models, reports, parity ledger, and README figure regenerate byte-clean from source. The source and clean-wheel suites both pass 72 tests.
Known boundaries
This remains a single-domain alpha. It does not infer dependencies or activation times, make causal claims, or provide connectivity-only or k-of-n relations. See the method, non-claims, and worked-example parity analysis.