fix: enable explicit decision measurement startup

[seonghobae]

Scope

Stacked on #1107 at de21ffd. Head babea95.

Expose the existing default-off decision receipt option through serve and an explicit CLI switch. Reuse native/store validation. Preserve prior positional arguments and adjacent #911 parser work. No numerical estimator or production default changes.

Evidence

• Ownership audit: 91 open PRs, 16 entrypoint-file overlaps, no serve or CLI invocation hunk overlap; exact heads in doctoring.
• RED: four failures before implementation. Final focused suite: six passing checks.
• Related CLI plus focused suite: 58 passed with warnings-as-errors, latest 2.00 seconds.
• Expanded CLI/telemetry suite is NOT clean: 106 passed, two parent-reproduced failures and HTTPError cleanup at shutdown. Parent-only allocation tracing identifies test_telemetry.py:1223. No warning suppression or unrelated lifecycle changes.
• Offline core wheel built and installed into a fresh isolated environment; six tests passed outside checkout in 3.19 seconds with isolated imports. This is source-candidate core-wheel evidence, not native integration or final release proof.
• Independent diff review completed; documentation count finding repaired. Actual browser screenshots inspected for changed doctoring, AGENTS, CLAUDE and Gap sections at 1265x712 English. No observed clipping or overlap in that scope.

Remaining boundaries

Bounded latest-256 export retains measurement_complete=false and reconciliation_required=true. Complete ingress reconciliation and independently adjudicated correctness are still missing. No accuracy/latency KPI gain, hosted acceptance, protected merge, release or deployment is claimed. Preserve required checks and independent review.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[seonghobae]

Exact-head package follow-up: rebuilt the core wheel offline from clean babea95. SHA-256 a16d800953ecf9f8bf3cf5d4b8e5bd62e67039f56b05825e1065d643d9509d90. Reinstalled that wheel into the isolated environment and ran the six entrypoint tests outside checkout with python -I, --noconftest, --import-mode=importlib and -W error: 6 passed in 1.47 s. This is core-wheel entrypoint evidence only; no native integration or customer KPI gain claim. The PR evidence and limitations sections were directly screenshot-inspected in GitHub at 1265x712 English and remain readable. Parent #1107 Security run 34695611099 completed successfully at de21ffd; that result does not establish this Draft successor CI or independent approval.

[seonghobae]

Verified an additional acceptance boundary at babea95 in an isolated detached checkout: real CLI startup with --decision-receipts --state-db, the built native Rust receipt module, loopback HTTP POST /v1/chat/completions, and SQLite persistence. The request returned HTTP 200 and export contained one acknowledged receipt with a non-null durable acknowledgement time. measurement_complete=false and reconciliation_required=true remained intact. Only repository mock model agents were used; no provider credentials or inference traffic.

Build: PYO3_PYTHON=/Users/seonghobae/contextual-orchestrator/.venv/bin/python cargo build --manifest-path rust/Cargo.toml --locked --offline -p contextual-decision-receipt (Python 3.14.4, PyO3 0.29.2). An initial direct-dylib harness load had linked Python 3.14.5 and crashed under 3.14.4; selecting the exact interpreter at build time resolved that harness mismatch. This is native source-build integration evidence, not an installed-wheel or end-to-end reconciliation claim. No PR source was modified.

[seonghobae]

Installed-wheel acceptance now passes at babea95. Built with locked maturin 1.15.0 for CPython 3.14.4, installed the core/native wheels plus hash-locked runtime requirements into an isolated environment, and executed from /tmp using Python -I with an empty environment. Explicit assertions require both core and native imports to resolve beneath the installed interpreter prefix.

The real CLI, loopback HTTP and SQLite check returned HTTP 200 with one acknowledged persisted receipt and a non-null durable acknowledgement time. measurement_complete=false and reconciliation_required=true remain intact. Models are repository mock fixtures only. A standalone reproduction script and raw stdout are retained locally; this is not a hosted CI, released artifact, provider or complete reconciliation claim.

Native wheel SHA-256: 78aa8b48ac5fefe54b92c238c3c8c27856aeb3193d07731ec00cb6057259dab0. Core wheel SHA-256: 9e8e635ff68cbdef012e0e9b8d465e7491a542b3caf9f6b0104c3a3fb3c9a0c1.