[jonahgabriel]

OMN-19194

market-drain measured (2026-09-22, OMN-17427 integration pass): the shadow
tests+coverage job's scoped slice (-m "unit and not kafka") takes 621-663s
of pytest time plus ~90s of collection -- 711-753s total on a normal runner.
The old --timeout 900 in run_shadow_slice.py left only 150-190s of
margin, and a ~25% slower runner (observed on PR #2777 at heads 544aa5a
and 96c7c3e, and PR #2779 at af89283, plus merged PR #2773) pushed the
same correctly-scoped slice past the wall: exit 124. This step fell through
to the generic failure branch on that code, so reason-graph and
product-readiness / evaluate reported a false PRODUCT_FAILED root for a
runner-speed timeout, not a real product defect. Both contexts are
non-required, so nothing was blocked, but the report was wrong.

Two-part fix, mirroring the OMN-18820 exit-125 precedent already in this
file:

• raise the budget to 1200s -- 300s+ (27%+) margin over the worst-observed
  timed-out case;
• add an exit-124 branch reporting product_conclusion=timeout and
  succeeding the step, so a genuine remaining hang still routes to the
  infra dimension instead of a false PRODUCT_FAILED root.

This is a sibling fix under the same ticket as the onex_change_control
shadow-timeout fix (onex_change_control#10828, already merged) -- that repo's
shadow job was duplicating its full, unscoped test suite and needed
narrowing; this repo's shadow job was already correctly scoped and just
needed a realistic budget plus the same infra-classification the 125 case
already gets.

Evidence-Ticket: OMN-19194
Evidence-Source: OCC#10850

[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).

[onexbot-occ-writer]

OCC autobind did not mint a companion for this PR: no changed-file candidate could be proven RED against the merge base, and emitting a PR-existence probe instead would be non-falsifiable evidence (OMN-15247). Hand-authored evidence is required.

[github-actions]

✅ Hostile Reviewer — PASSED

Blocking findings (critical/error): 0
Total findings: 9
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)