Repository documentation

Phase 1 alpha release and rollback

Released artifact: v0.1.0a1 at commit a67dfdb5058c73cdb3f65c2517151ef21129a90e. The published wheel literate_ai-0.1.0a1-py3-none-any.whl has SHA-256 03ee1a5216fe5e6435f269f6e18edf78a65d92fc76c50c2632da4be172eb9f58.

Phase 1 introduces literate-ai as a read-only compatibility and shadow dependency. OVA remains the behavioral authority. The alpha must never rewrite an OVA manifest, cache, package, provenance record, publication record, or setting while it is in shadow mode.

Pinning

Consumers pin the immutable Git commit behind the v0.1.0a1 tag or a wheel whose SHA-256 has been verified against the release record. Floating branches and unqualified version ranges are not migration inputs. Source checkouts are indexed independently at their exact revision; the wheel is not treated as a substitute for source intelligence.

Entry gate

Before enabling shadow mode:

1. Verify the framework repository validation matrix on macOS and Linux.
2. Install the exact alpha into an isolated environment and confirm litai spec skills finds all built-in manifests outside the source checkout.
3. Run python -m literate_ai.compatibility.ova --baseline BASELINE --candidate CURRENT.
4. Preserve the report digest with the OVA test evidence.
5. Leave all OVA cache and generation writers on the legacy implementation.

Exit code 0 means semantic parity, 1 means drift or missing artifacts, and 2 means invalid or interrupted input. Only 0 is eligible for an observation run.

Rollback

Rollback is configuration-only during Phase 1:

1. Set OVA's framework mode back to legacy.
2. Restart OVA and verify its health and Settings page.
3. Retain the shadow report and framework cache as diagnostic evidence; do not translate or delete OVA cache state.
4. Remove the optional literate-ai installation only after confirming no process imports it.

Phase 2 rollback is a feature-flag reversal to the last legacy seam. Destructive cache conversion is forbidden, so the legacy reader and cache remain usable throughout the two-release observation period.