chore(OMN-16815): repin omnibase-infra 0.38.11 and retire the OMN-16767 chain-gate xfail

[jonahgabriel]

Closes OMN-16815.

What this does

Two changes that must land together, because the first makes the second fail.

1. uv.lock: omnibase-infra 0.38.10 -> 0.38.11 (single-package upgrade, uv lock --upgrade-package omnibase-infra; the >=0.38.3,<0.39.0 range in pyproject.toml already admitted it, so nothing else moved — the diff is 3 lines).
2. tests/chains/: retire the broken_by="OMN-16767" strict-xfail marker, and invert the arm-selection assertion it was paired with.

Why now

omnibase_infra v0.38.11 was cut and published today from dev SHA 4529c3486a8bb49a45a623a008e4bbfa2a3a3148. It is the first release carrying the delegation-arm fixes:

• omnibase_infra#2937 — db_io must not steal a typed def-B handler into the projection arm (OMN-16767)
• omnibase_infra#2943 — the type-scoping predicate must agree with the dispatch it scopes (OMN-16767)
• omnibase_infra#2949 — db_io must not suppress result delivery; no consume commits with no effect (OMN-16798)
• omnibase_infra#2951 — a record already on a dead-letter topic must not be re-DLQ'd onto itself (OMN-16798)

The delegation-routing-reducer row in tests/chains/test_event_chain_gate.py was xfail(strict=True) and its own comment named the trigger: "Fix in flight: omnibase_infra#2937. This row turns RED the moment the pin picks that up." It did — verified before the edit:

FAILED tests/chains/test_event_chain_gate.py::test_chain_reaches_terminal_and_never_quarantines[delegation-routing-reducer]
[XPASS(strict)] OMN-16767: chain is dead today. STRICT — this turns RED (XPASS) the moment the fix lands, which is the signal to delete this marker.

That XPASS is the proof the chain is alive: the row drives the real MessageDispatchEngine from the real contract.yaml and requires a ModelRoutingDecision terminal on onex.evt.omnibase-infra.routing-decision.v1 with nothing on quarantine. The marker is gone; the row is now an ordinary passing row with nothing to hide behind if the chain dies again.

The arm assertion is inverted, not deleted

tests/chains/test_projection_arm_constructs.py asserted that wiring selects the projection arm. Under 0.38.10 it did — and that is precisely the OMN-16767 defect (a typed def-B handler handed a raw input_data dict). Post-#2937 the correct arm is the typed def-B arm, so keeping the old assertion would mean asserting the defect. All seven profiles failed on the bump:

AssertionError: [onex-prod] expected the PROJECTION arm (_make_projection_dispatch_callback.<locals>._callback),
got '_make_dispatch_callback.<locals>._callback'

Arm identity is still pinned in both directions: the projection arm is now asserted against by name as the OMN-16767 regression signature, so a silent flip back fails here with the cause spelled out.

The OMN-16794 constructibility claim did not evaporate

Because #2937 clears db_tables for a typed def-B handler, wiring no longer reaches _resolve_projection_database_target for this contract — the call that raised under 0.38.9. Rather than let that coverage disappear as a side effect of an arm change, the claim is now asserted directly against that resolver under all seven topology profiles (test_projection_binding_resolves_under_every_topology_profile), alongside the existing resolver-level pin in test_overlay_table_resolves_to_its_physical_grant_schema. Neither goes through wiring, so neither is hostage to the next arm-selection change.

Non-vacuity proven, not assumed. Monkeypatching physical_grant_schema_for_table back to 0.38.9 behaviour (lambda schema, name: schema) makes the new assertion raise the original error under all seven profiles:

local RAISES: Projection binding 'tenant_projection' principal 'tenant_projection_writer' lacks declared ...
test RAISES: ...
onex-dev RAISES: ...
stability-test RAISES: ...
prod RAISES: ...
judge RAISES: ...
onex-prod RAISES: ...

• Installed version, not just the range: uv run python -c "from importlib.metadata import version; print(version('omnibase-infra'))" -> 0.38.11
• uv run pytest tests/chains/ -q -> 17 passed (was 8 failed / 2 passed immediately after the lock bump)
• uv run ruff format src/ tests/ + uv run ruff check --fix src/ tests/ -> All checks passed
• uv run mypy src/omnimarket/ --strict -> Success: no issues found in 3044 source files
• pre-commit run --all-files -> exit 0, zero failures
• Pre-push: the governed impacted-test selector ran and passed ([prepush-smart-tests] impacted tests passed; allowing push) — no PREPUSH_FULL_SUITE, no -k filter, no bypass.

Release provenance: PyPI omnibase-infra 0.38.11 (info.version confirmed after the ~60s index lag), GitHub release https://github.com/OmniNode-ai/omnibase_infra/releases/tag/v0.38.11.

Evidence-Ticket: OMN-16815
Evidence-Source: OCC#7356

---

Correction — the first push was green for the wrong reason (2nd commit, 60df58a8)

The local 17 passed above was environment-dependent, and CI caught it. The Event Chain Gate went RED on the runner:

ProtocolConfigurationError: [ONEX_CORE_041_INVALID_CONFIGURATION]
No tier has a configured endpoint for task_type='summarization'.

Nothing about the chain differed. My shell carried live cloud credentials (GEMINI_API_KEY, LLM_GLM_API_KEY, OPEN_ROUTER_API_KEY), which made the cheap_cloud rung routable locally; the runner has none, and the local rung (local-reasoner, where summarization routes first) was unroutable on both because its endpoint_url is null in the repo default — site-specific and overlay-supplied by design (OMN-12815 / OMN-15807).

So the gate's verdict was a function of the ambient environment. That is the OMN-16796 defect class, inside the suite written to catch chains dying silently — worth naming plainly rather than patching quietly.

Reproduced locally before fixing, by stripping the environment CI does not have:

env -u GEMINI_API_KEY -u LLM_GLM_API_KEY -u OPEN_ROUTER_API_KEY ... uv run pytest tests/chains/ -q
  -> 1 failed, 16 passed   (same ProtocolConfigurationError)

Fix: a module-level autouse fixture writes a minimal BIFROST overlay giving local-reasoner a complete but deliberately unreachable chat-completions URL (127.0.0.1:9, the discard port). The node under gate is a routing reducer — it emits ModelRoutingDecision and never opens a connection — so an unreachable address costs nothing and keeps the gate server-free and credential-free. tests/conftest.py::_ensure_bifrost_contract_path clears BIFROST_OVERLAY_PATH unconditionally; a module-level autouse fixture runs after it and wins, which is that fixture's own documented override path.

Verified both ways after the fix:

uv run pytest tests/chains/ -q                                  -> 17 passed   (developer env intact)
env -u <every cloud key> -u <every BIFROST_* binding> ...  -> 17 passed   (the CI shape)

Note this does not weaken the gate: the row still requires a real ModelRoutingDecision terminal on onex.evt.omnibase-infra.routing-decision.v1 with nothing on quarantine. It only removes the machine from the answer.

[coderabbitai]

Warning

Review limit reached

Next included review available in 9 minutes.

View limit details

Limit details: You’ve used the included review currently available. Your 131 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: c565ea6d-485a-4cf9-b541-f9e6e100245d

📥 Commits

Reviewing files that changed from the base of the PR and between de13368 and 60df58a.

⛔ Files ignored due to path filters (1)

• uv.lock is excluded by !**/*.lock

📒 Files selected for processing (2)

• tests/chains/test_event_chain_gate.py
• tests/chains/test_projection_arm_constructs.py

---

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

[github-actions]

✅ Hostile Reviewer — PASSED

Blocking findings (critical/error): 0
Total findings: 0
Models succeeded: qwen3-review,qwen3-review-b

---

Gate semantics

Verdict	Meaning	Blocks merge?
passed	>=2 models succeeded, no critical/error findings	No
blocked	CRITICAL or ERROR findings found	Yes
degraded	Fewer than 2 models succeeded (infra unavailable/timeout)	Yes (OMN-15110)

Powered by omniintelligence.review_pairing.cli_review — node-based adversarial review via HandlerLlmCliSubprocess (OMN-8524)

[github-actions]

✅ Architectural Review — PASSED

Errors: 0
Warnings: 0

What this checks

Rule	Description
ARCH-TOPIC-001	No hardcoded Kafka topic strings in handler code
ARCH-DI-001	No event_bus=None bypass
ARCH-DI-002	No direct Handler instantiation outside workflow_runner/adapters
ARCH-DI-003	No reinvented DI containers
ARCH-TOPIC-002	contract.yaml topics follow `onex.{cmd

---

No architectural violations found.

Static architectural lint — no model inference (OMN-14176).