[echoomegaprime]

Objective

Refs #42.

Echo workspace releases need measured application behavior tied to the exact candidate and running processes. A healthy endpoint, a green source suite, or a local session alone cannot establish production readiness.

This adds a concrete observer for the shipped Echo CLI owner API and Echo ADK workspace. It verifies pinned Git archive identities, source files, installed Python modules, built CLI artifacts, configuration, process arguments and listener ownership before and after executing a bounded journey. It uses the actual native Prime session and owner routes.

The journey creates zero-cost, empty-scope paused bots; checks event idempotency and cancellation before dispatch; edits and restores isolated bot memory; persists one-time/monthly/yearly/cron schedules; verifies department export and reviewed import; exercises real sandboxed Chromium snapshots and screenshot isolation; observes three health/budget samples; and removes only acknowledged resources.

HTTP stays on literal loopback, without proxy or redirect. Request counts, response sizes and elapsed time are bounded. Unknown mutations are not replayed or cleaned up using guessed IDs. Cookies and response bodies are excluded from evidence.

Evidence

Source: a9d613e81f12d21f8637d98b8428b35277fd1b82, based on canonical main 4f7d99e2dd4422f3d81cf5b709581be963f5818a. Four scoped files change; no existing verifier, policy or capability contract changes.

• Full Linux deterministic suite: 661 passed, 1 existing skip, 1 warning; 80.19% coverage, satisfying the unchanged 80% gate. This ran through scripts/verify_p2.py with the repository's normal pytest coverage command.
• Observer suite: 34 passed. Controlled HTTP and real owned-process tests cover missing/ambiguous evidence, source/build/configuration drift, non-listening sockets, redirects, oversized/truncated responses, slow-body deadlines, and unknown delete outcomes.
• P1 acceptance, compileall, pip dependency check, Ruff undefined-name/import checks and diff checks passed.
• Native role adoption and exact-source build preflight: 24 required PASS, zero required failures or blockers; later-stage release evidence is not admitted by this source-only contract. This is source readiness, not release admission.
• The final commit removes one trailing blank line from the tested test file; executable observer bytes remain identical to the live observation.
• Hosted CI completed success at this exact source: certification-forge-ci / 42468104151, including its normal P1/P6/P7 acceptance stages.

Actual candidate observation: on 2026-11-08, CLI 768b364b2a702f9e3ee6ff0b25e1b60e3029d9fb and ADK 3e798d56f34a8657c4db626464a604ed9aa9c85c passed 9/9 engineering checks in an isolated owner-created loopback lane. Chromium and the application APIs actually executed. Acknowledged resources were removed, and the lane owner then stopped both services and verified both listeners closed.

Observation SHA-256: 6897ee86ea668afc96075386f1aeef99ea21a81daacb6ea543f1dd8b0e03e8de. Observer module SHA-256: 30d0e7bed7c7994980b3883c70cce079a93a3ff24d9885b86aa825ed01e84e0b.

Earlier harness attempts and the initial 79.43% coverage failure remain retained as historical failures. Contract/transport corrections and substantive identity/cleanup regressions produced the results above; target guards and the coverage threshold were unchanged.

Required legacy check limitations

The complete P2 wrapper still fails its unchanged scanner on four existing synthetic secret assignments in tests/test_p6_platform_integration.py and tests/test_teacher_generator.py. The complete P3 wrapper fails its unchanged historical closure check for contracts/certforge-capabilities.v1.json (current hash starts ac1014c2). Both affected fixtures and the capability contract were verified byte-identical to the base commit. These gates were not altered, waived or reported green.

Security and release impact

The output is unsigned engineering evidence and always retains NOT_READY, production_e2e:false and signature_verified:false. Real SDK journeys, provider consent/read, bounded autonomous reasoning, external Workspace owner acceptance and trusted collector admission remain outstanding. No runtime model-call counter was instrumented.

The existing production-attestation consumer and scoped Windows-package worker do not provide a generic Echo collector signing entrypoint. This PR neither borrows that transport identity nor provisions new signing trust. It makes the measured Echo journey executable and reviewable for the legitimate collector owner.

• No secrets, private discovery excerpts, customer evidence or personal data are included.
• Existing deterministic verdict and fail-closed behavior is unchanged.
• Public invocation, evidence boundaries and limitations are documented.
• No production deployment, service change, signing or trust mutation occurred; staging and owner admission remain prerequisites.
• Unrelated history and license boundaries are preserved.