test(kanban): make progress-stall policy tests probe-deterministic (t_5457397a)

[Kyzcreig]

Kanban: t_5457397a

Problem. tests/hermes_cli/test_kanban_progress_stall.py waited for the REAL ps probe to read a freshly-spawned child as 0.0% CPU (6 consecutive samples, 30 s deadline) before every stall/reclaim policy test. On Linux, procps pcpu is lifetime cputime/elapsed, so a child whose startup burned ~30 ms needs ~30 s to round to 0.0 — much longer on a loaded runner. Result: 6/10 failed together (never read idle to the real probe) in merge-group batches 36060149219 and 36064421258, ejecting unrelated PRs #1002 and #994.

Fix.

• kanban_db._worker_cpu_active split into _process_cpu_table() (the only I/O) + _cpu_active_in_table() (pure parse). Behaviour unchanged: unknown state → True (never authorizes a kill); children never veto.
• Policy tests keep a real worker process (reclaim still really SIGTERMs it) but feed the probe a deterministic table through the seam. The table is parsed by production code, so children-never-veto stays covered.
• Real-probe integration kept: busy process → active (load-immune direction) and new test_real_probe_reads_in_flight_worker_as_idle, which SKIPs with the measured reason (probe errors / loadavg1 > cores) and FAILs only on an unloaded host with a working ps.
• Probe-failure tests still drive the real probe into a broken subprocess.run.
• Reclaim/stall/escalation assertions unchanged.

Verified (Mac Studio, repo venv).

• 11/11 pass (19 s).
• 5/5 file runs green under 64 busy loops on 32 cores (loadavg1 up to 161), 54–73 s each, 0 skips.
• 6/6 kanban_db mutants killed: any-child veto, probe-failure-authorizes, no escalation, CPU veto ignored, probe always-active, no reclaim.

Follow-up (separate card): on Linux the production probe's lifetime-average pcpu means a long-lived worker with meaningful accumulated CPU may read "active" long after it goes idle (veto, fail-safe direction) — worth measuring on ACE-AI.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[Kyzcreig]

🤖 merged-by: apollo · lane: claude-bridge-session · gate: BYPASS: FR PAUSED by Ace ruling 2026-09-22 · why: t_5457397a: probe seam extraction, production parse unchanged (_cpu_active_in_table), policy tests fed a deterministic ps table, real-probe integration test kept (SKIP-with-reason on host load); 38 checks green, Apollo self-review; ejected #1002/#994/#988 batches today

[Kyzcreig]

🤖 merged-by: apollo · lane: claude-bridge-session · gate: BYPASS: FR PAUSED by Ace ruling 2026-09-22 · why: re-arm after merge-queue ejection by the test_kanban_progress_stall load flake on the local pool (fix #1040); PR untouched

[Kyzcreig]

🤖 merged-by: apollo · lane: claude-bridge-session · gate: BYPASS: FR PAUSED by Ace ruling 2026-09-22 · why: re-arm (auto-merge disable+enable) after merge-queue ejection by the test_kanban_progress_stall load flake on the local pool; fix is #1040; PR untouched

[ang-prism]

FleetReview

Review: post-merge · head 7fd339db7262 · duration 5m 05s
Profile: light (merit: default light: lines 164<800, files 2<1000000, hunks 12<1000000, no hot path) · policy: below-size-and-path-gates
Roster: B-state → gpt-6-sol (openai), C-assert-xhigh → claude-code-opus-5-5 (anthropic), F → gpt-6-sol (openai), G → grok-4.6 (xai)

Post-merge review (fleetreview:post-merge override): this reviewed the merge commit against its first parent — the bytes that already shipped. It is not a pre-merge gate pass.

profile: light (rule: default light: lines 164<800, files 2<1000000, hunks 12<1000000, no hot path) · round 0 · members: B-state, C-assert-xhigh, F, G · families: anthropic,openai,xai

Confidence: 4/5

No issues found.

---

FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, D=grok-4.6, F=gpt-6-sol · cost: $0.66 · duration: 5m 01s · rounds: 1 · files examined: 2