hermes-blind v0.0.6 — experimental

Initial release. Phase 2 of a 5-phase staged build. Phase 4 empirical variance-reduction test has not been run — the core hypothesis (that the scaffold actually reduces score variance) is structurally framed but not yet measured. Treat this as a research artifact, not a production debiaser, until the ablation study ships.

What ships:

• Five scaffold variants (null, micro, short, v1, full) for ablation studies
• wrap() and extract_disclosure() API
• 23 deterministic unit tests, no LLM calls, green on py3.10/3.11/3.12
• Full OSS polish bundle + CI matrix + sealed audit trail (placeholder signature; grant pending)
• README scored 7.5/10 with 0 hedges by hermes-rubric (audit log under rubric-runs/)

See CHANGELOG.md for what is and isn't validated.