MC-4974

Deferred items 3+4 from the 2026-06-11 review session (Elmar-approved). Route via board-manager -> internal Kanban; orchestrator gates only. IMPLEMENTATION IS AUTHORIZED FROM THE FIRST RUN — do not stop at reconnaissance; a handoff without commits+tests will be rejected (lesson from MC-4913/MC-4965 first-run false-stops). THREE SLICES, one branch, one handoff: 1. task_detail.html FAILED stat conflates timeout+failed (Tessa #9): show them separately in the 7d stats strip ('Failed N · Timeout N'), matching the one-vocabulary rule from MC-4942 U10. models.get_task_run_stats may need a timeout count — additive only. 2. radio-daily-brief fails without emitting an error line (Tessa #7, task-side): in the radio-show build script (~/workspace — find via tasks/radio-daily-brief.md command), ensure every failure path prints a final 'Error: <cause>' line to stdout before non-zero exit, so /runs failure_preview extracts the real cause. Touch ONLY error/exit paths — no logic changes. 3. /tasks server-side pagination (item 31 follow-up): tasks_page renders all 116+ rows (~480KB). Server-side window: ?limit= (default 100, clamp 500) + offset, totals from full counts (pattern: /runs pagination from MC-4942 U4); keep client windowing/search working over the served rows; 'Show all' link sets limit=500. PRESERVE: problems-first default, healthy-collapse toggle (MC-4965), cross-tab search, sticky header. GATES: pytest tests/test_mc4942_u10_tasks_ux.py tests/test_mc4942_u4_runs.py tests/test_mc4942_u6_task_detail.py + new assertions per slice; py_compile; the documented mc4056 flex-wrap failure is pre-existing — don't chase. NO service restarts, NO merging — review-required handoff with branch+SHA+test evidence. Slices 1+3 = mission-control repo; slice 2 = workspace repo (separate branch kb/<id>-radio-error-line, note it in the handoff). Expected check-in: within 2h of dispatch.