Waveframe Labs mark
Waveframe Labs
Deterministic Admissibility Kernel

CRI-CORE

CRI-CORE evaluates canonical proposals against compiled authority and returns a deterministic decision. It does not invoke the production mutation; Waveframe Guard consumes the decision at the mediated execution boundary.

Release status

PyPI release: 0.14.0

pip install cricore==0.14.0

This published version is within Guard 0.19.0’s declared dependency range.

GitHub release: v0.14.0

Published September 9, 2026 with strict-by-default evaluation. Direct PyPI metadata confirms the matching 0.14.0 artifacts are published; the release record’s pending-publication wording is historical.

Inspect v0.14.0 release

Kernel contract

Input

  • Canonical proposal produced at the runtime boundary
  • Compiled authority contract
  • Exact contract identity and declared evaluation context

Output

{
  "commit_allowed": false,
  "reason": "...",
  "stage_results": [...]
}

The caller must treat a blocked or configuration-error result as non-permission to mutate.

v0.14.0 mode behavior

Strict is the default.

Omitting mode selects strict. A context declaration cannot silently downgrade strict evaluation.

Local advisory evaluation must be explicit.

Use exact mode="local" only for advisory/local evaluation. Advisory commit_allowed output must not be treated as production mutation permission.

Configuration errors reject early.

Only exact strict and local strings are accepted. Invalid or conflicting declarations reject before policy evaluation.

Guard remains stricter at the execution boundary.

Guard 0.19.0 execution is never advisory: local versus cloud selects authority resolution/connectivity, not permission strength.

Truthful evidence boundary

CRI-CORE v0.14.0 does not by itself establish cryptographic provenance for the current structured pipeline.

The v0.14.0 release explicitly leaves the stronger integrity/publication evidence work open. Existing integrity and publication stages still accept empty mappings in cases documented by the release. Guard's separate Ledger publication verification remains the authority-verification boundary for the current mediated runtime path.

Where it fits

Before CRI-CORE

Guard receives the actual tool call, resolves and verifies published authority, derives approved runtime facts, and uses Proposal Normalizer to produce the canonical proposal.

After CRI-CORE

Guard enforces the result. A blocked callback never runs. An allowed callback is invoked only through the mediated path, followed by truthful execution/mutation attestation.

View the current runtime architecture

v0.14.0 release hashes

Wheel
sha256:11a22b4fb15137a0e535b8dd71bf927260800559221f0e10512029d59fb18702

Source distribution
sha256:458c5f4742a041a3ed2e334fbe57b4e5bb85bf50b85aa56871537d58d9503dec

Boundary responsibility

  • Does not author or interpret source policy.
  • Does not compile authority contracts.
  • Does not normalize the caller's original tool invocation by itself.
  • Does not execute the production callback.
  • Does not prove that alternate mutation paths are unavailable.