Skip to content

Research: qualify trusted authority-adapter seam RC2 - #78

Draft
camerontjs-dot wants to merge 14 commits into
research/c2-authority-artifact-reconstruction-rc1-20260917from
research/authority-adapter-seam-rc2-20260917
Draft

camerontjs-dot wants to merge 14 commits into
research/c2-authority-artifact-reconstruction-rc1-20260917from
research/authority-adapter-seam-rc2-20260917

Conversation

@camerontjs-dot

Copy link
Copy Markdown
Owner

Classification

Draft Research only. Do not merge as a maintained runtime change, tag, release, or infer production readiness from this PR.

This successor is stacked on Draft Research PR #77 / terminal 97727a7d2f61c46b4e7ef3e11279fb0af7b915cd. It changes no maintained src/**, scripts/**, or tests/** bytes.

Question

Can materially different trusted authority domains converge on one unchanged post-admission Decision seam without making the projection kernel domain-aware or allowing callers to move the trust anchor?

Terminal disposition

SUPPORTED_TRUSTED_AUTHORITY_ADAPTER_SEAM

Decisive execution:

  • executed science head: 360dc46d6b6e510d8ce36fa06f2fb0de5ff47372
  • workflow run: 35232698263
  • job: 105240629694
  • artifact ID: 10501967992
  • artifact ZIP digest: sha256:745346b1dc7530d9f362b89daeea52edfbe6c20dfe44cdf59f6f2187d8c91169
  • packaged evidence TGZ: 74dee690eab4841a5f363d6538dc6d5ecee93051d60b4eaa3bab60a64fa98f49
  • frozen projection-kernel SHA-256: e12d53020b3a6a046f4ee7688db0a5ea5c5e0a63e758b1c1e5711e7b227759e3

Observed result

All four preregistered phases passed:

  1. CAL/C2 shadow seam reproduced the maintained legitimate Decision and exact canonical Contract-D bytes. The five RC1 authority substitutions were rejected before policy.
  2. The frozen Git-bound artifact-manifest authority replay passed unchanged against the exact same projection-kernel identity.
  3. A materially different Ed25519 signed content-addressed authority used the same post-admission seam. A legitimate favorable fact produced CLEAR; a legitimate unfavorable fact was admitted and produced policy HOLD.
  4. Cross-adapter laundering, wrong signer/digest/version/target, caller adapter injection, and caller policy implementation injection were rejected at the trusted runtime boundary.

The deliberately weak controls remained unsafe as expected: direct caller control of the raw kernel policy registry changed a Decision under the same policy ID/version, and caller selection of the signature verification key accepted an attacker signature. This is causal evidence that adapter/policy implementation and trust-root selection must remain trusted machinery.

Supported boundary

raw external authority
        ↓
trusted domain authority adapter
        ↓
trusted domain target resolver
        ↓
trusted domain policy semantics
        ↓
unchanged domain-neutral Decision projection
        ↓
exact Contract D

The common seam begins after authority admission. This PR does not establish a universal raw-ingress schema or plugin system.

Preserved failure

The first hosted run 35232390053 / job 105239555259 is preserved as APPARATUS_LEGACY_DIGEST_PREFIX_ASSERTION_DEFECT. The frozen manifest scientific runner printed PASS, but a wrapper compared prefixed and bare SHA representations and stopped before the new Phase C/D harness. The decisive successor changed only that mechanical representation check via a preserved-harness wrapper/workflow correction.

Remaining boundaries

  • real-runtime CAL Contract-B artifact location remains unresolved;
  • mutable API freshness/revocation/key rotation is untested;
  • trusted adapter and policy implementation binding needs maintained-runtime design and independent conformance before extraction;
  • no arbitrary plugin/caller-installed adapter support is established.

See research/authority-adapter-seam-rc2/RESULT.md for the complete terminal record.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant