feat(sol): frontier provider mesh with proof-carrying execution

[ghost]

Purpose

Continue the SOL control plane into a provider-neutral frontier interoperability mesh while keeping Automaton-3 as the sole authority root, and bind the constitutional preconditions required before AGNT-004 can be evaluated for admission.

Exact lineage attestation

BASE = #225 @ cd920379df1a23a61b969f0e52549c04bc3a91bb
CURRENT_HEAD = 31aec51c32caa2431cb94ee742c912059802568b
MERGE_BASE = cd920379df1a23a61b969f0e52549c04bc3a91bb
AHEAD_BY = 98
BEHIND_BY = 0

The current head is a one-commit fast-forward from historical head f467c06735b3b2240832d057a1f7d5e4098b39f9. The only delta is .github/workflows/frontier-provider-mesh.yml, repairing two verification-surface defects discovered by exact-head reproduction:

1. the previous unittest command executed 0 LUT-KAN parity tests because agents/tests/test_lut_kan_parity.py contains pytest-style functions plus a direct __main__ assertion harness;
2. the credential grep had a shell-quoting syntax error and never actually scanned the candidate.

The repaired workflow executes the 3 LUT-KAN assertions directly and uses a Python credential scanner over the same frontier/Sensorium surface.

Exact-head external runtime witness — ESTABLISHED

A separate dependency-bearing GitHub-hosted runner checked out the AEGIS source by exact SHA 31aec51c32caa2431cb94ee742c912059802568b and reproduced the PR verification surface.

RUNNER_REPOSITORY = tarikskalic33/info
RUN_ID = 32215198034
EXTERNAL_ARTIFACT_DIGEST = sha256:6c92f1c70abe8be338f4ea0038c9fd2f2941a28ee90ef08309d5a1e538e52e62

Observed execution:

EXACT_STACKED_LINEAGE = PASS
COGNITIVE_EVIDENCE_BINDING = 6/6 PASS
FRONTIER_PYTHON = 59/59 PASS
LUT_KAN_PARITY_ASSERTIONS = 3/3 PASS
CONSTITUTIONAL_BUNDLE = 35/35 PASS
FRONTIER_SENSORIUM_VITEST = 79/79 PASS
TYPESCRIPT_TYPECHECK = PASS
CREDENTIAL_SCAN = PASS

The 79 Vitest assertions include the exact current Sensorium observation, VCM bridge and degradation surfaces together with frontier runtime, transports, fencing and Automaton-3 verification tests. This promotes an external exact-head AGNT-004/frontier component witness for the current SHA.

This witness is not represented as AEGIS repo-native CI. The AEGIS-hosted Actions account/billing blocker remains tracked separately in #269.

Current ledger

LINEAGE_CONTINUITY = ESTABLISHED
BASE_ALIGNMENT_WITH_CURRENT_225 = ESTABLISHED
REPLAY_REQUIRED = FALSE
FORWARD_MERGE_REQUIRED = FALSE

CONSTITUTIONAL_BUNDLE_OMEGA_1_2_3 = IMPLEMENTED_AND_EXTERNALLY_EXACT_HEAD_TESTED
CONSTITUTIONAL_EXTERNAL_EXACT_HEAD_PASS = ESTABLISHED / 35_OF_35

AGNT_004_DESIGN = SPECIFIED
AGNT_004_IMPLEMENTATION = ESTABLISHED_IN_BRANCH
AGNT_004_OMEGA3_EXPLICIT_BOUNDARY = ESTABLISHED
AGNT_004_EXTERNAL_EXACT_HEAD_COMPONENT_WITNESS = ESTABLISHED
AGNT_004_REPO_NATIVE_EXACT_HEAD_CI_PASS = NOT_ESTABLISHED
AGNT_004_EMPIRICAL_VALIDATION = NOT_ESTABLISHED
AGNT_004_FULL_AUTHORITY_INTEGRATION_PASS = NOT_ESTABLISHED

FRONTIER_EXTERNAL_EXACT_HEAD_TEST_PASS = ESTABLISHED
EXTERNAL_EXACT_HEAD_TEST_PASS = ESTABLISHED
AEGIS_REPO_NATIVE_EXACT_HEAD_CI_PASS = NOT_ESTABLISHED / BLOCKED_BY_ISSUE_269
PRODUCTION_ADMISSION = NOT_ESTABLISHED

Constitutional Sensorium boundary

The observation contract remains explicitly non-authoritative:

authorityEffect = OBSERVATION_ONLY
observationTier = T2
authorityWeight = 0
mayGroundStateTransition = false

Provider/model output remains evidence only and cannot grant AEGIS authority. D3 requires explicit operator approval; D4 remains denied in this slice. frontier-automaton3-verifier.ts delegates authority evaluation to the existing Automaton-3 path rather than creating a second authority root.

Security/dependency debt surfaced by the witness

npm ci completed successfully but reported:

7 vulnerabilities
1 moderate
6 high

This is a separate dependency-security triage item. The successful functional/constitutional witness does not erase or downgrade that debt.

Deliberately not performed

• no merge to main;
• no mutation of PR SOL: governed cross-platform control plane foundation #225 or PR fix(governance): add fail-closed provider containment gate #265;
• no live production provider calls;
• no credit spend;
• no API-key creation or secret provisioning;
• no IAM/OAuth/DNS changes;
• no empirical-validation claim for the VCM/retention engineering model;
• no production/certification claim.

Remaining before admission

1. Resolve infra(actions): hosted jobs fail before repository code executes #269 and obtain repo-native exact-head CI on the then-current feat(sol): frontier provider mesh with proof-carrying execution #264 head.
2. Triage the 7 dependency vulnerabilities without weakening provider/Sensorium boundaries.
3. Evaluate SensoriumDegradationV1 into the Authority Control Plane only as a contractive input after exact state/topology freshness binding; Sensorium must never become an authority root.
4. Continue the approved implementation sequence for EffectObservation → VerifyEffect → EffectReceipt → CompleteVerification without introducing a generic effect producer or bypassing PR feat(automaton3): bind transitions and separate decision/execution/effect receipts #268 SAFE_INCOMPLETENESS.
5. Keep this PR draft and PRODUCTION_ADMISSION = NOT_ESTABLISHED until those independent gates are satisfied.

Historical #264 heads remain evidence only for the exact bytes they bind; they are not aliases for the current head.

[vercel]

Deployment failed for project hub with the following error:

Resource is limited - try again in 1 day (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/tariks-projects-e9198507?upgradeToPro=build-rate-limit

[tarikskalic33]

Exact-head dependency triage for current PR #264 head 31aec51c32caa2431cb94ee742c912059802568b completed on external dependency-bearing runner.

RUN_ID = 32215439198
WITNESS_ROOT = 367ed53ee4f93a981314aea4ac41dc51e021b16fcaad4b19cbbbdb9813ffb6a5
ARTIFACT_DIGEST = sha256:22c95336a07aec35632ac784f76fb92b6f33296be93a8966fe556c1ea2e379aa

FULL_GRAPH = 7 vulnerabilities (1 moderate, 6 high)
PRODUCTION_GRAPH (--omit=dev) = 0 vulnerabilities
FULL_AUDIT_RC = 1
PROD_AUDIT_RC = 0

The 7 findings are confined to the dev/tooling graph. Direct findings: ajv, postcss; transitive findings: brace-expansion, fast-uri, js-yaml, nanoid, undici. No npm audit fix --force was performed. This does not establish repo-native CI or production admission; #269 remains the repo-native Actions blocker.