v0.2.0 — Vault Lab ships; release and method versions split

The release record catches up with main: Vault Lab v1 is now part of a published release, not just unreleased main.

Vault Lab

A local, deterministic teaching and falsification environment for the BAION Vault Method. Break a fictional vault five specific ways, watch the real checker catch each one, see the exact byte-level repair — then try to beat the checker.

./lab/start.sh          # guided run (15–25 min)
./lab/start.sh --check  # non-interactive verification (what CI runs)

• Five break/repair cases: missing review, missing reference, candidate byte mismatch, active receipt conflict, review byte mismatch.
• Four judgment scenarios the checker cannot decide for you.
• A checker-escape challenge with a public four-class ruling model and a Hall of Catches.
• Fail-closed: guided mode and --check share one assertion path. Any expectation drift — wrong diagnostic codes, findings on a fixed vault, a substituted checker — exits nonzero and suppresses the final bounded claim. Adversarial regression tests cover the zero-findings-checker and unexpected-code cases.
• No AI model runs inside the lab. No network. All fixtures fictional. No badge, certification, or score.

Governing specification: BAION_VAULT_LAB_V1_REFINED_BUILD_SPEC v0.3.0 (2026-07-16), SHA-256 2afb782d8607e8db0c795b0b57fe0a42ad8182c266a58124efba0362aee7c143 — cited with its required verbatim statements in lab/README.md.

Version model: release ≠ method

This release names two versions on purpose:

• Release version 0.2.0 — what ships from this repository (checker, lab, docs).
• Method version 0.1.0, unchanged — the specification contract vaults declare via method_version. SPEC.md, src/, and schemas/ are byte-identical to v0.1.0 except the SPEC.md status line. The method did not change; a teaching surface was added.

The previous single-string version parity made this release impossible to cut honestly: bumping method_version past the 0.1.x series makes bvm-lint reject its own templates (BVM002). scripts/check_version_parity.py now enforces the two tracks separately, with a regression test for the series-escape case.

Evidence

• Tag commit: 8bfd8f5 (PR #1)
• CI on the tagged commit: verify 3.11 / 3.12 / 3.13 — all green
• REPOSITORY VERIFY PASS: 68 tests, version parity (release 0.2.0, method 0.1.0), diagnostic parity 49 codes, release surface, vault lab check.

Boundaries

PASS means the checker found no declared structural violation in fictional fixtures — not that any claim is true, complete, safe, or decision-grade. Structural conformance only; no factual, scientific, legal, security, or safety certification. Reference methodology, not a standard. v0.1.0 remains the preserved record of the first cut.