Skip to content

Operational pre-decisions sign-off register (ID-61.8)

Operational pre-decisions sign-off register (ID-61.8)

Section titled “Operational pre-decisions sign-off register (ID-61.8)”

Created: 05/06/2026 (S314). Owner: Liam (product owner). Scope: the four operational pre-decisions (a)–(d) that must be on the record before the ID-45/T7 full-corpus re-ingest (Spine S5). Each row records the decision, its ratified value, the decision source, and sign-off. Acceptance for ID-61.8 is this register itself — the decisions are recorded; there is no code test.

#DecisionRatified valueDecision sourceSign-off
(a)/walk cadence (OQ-1)Nightly at 02:00 — cron 0 2 * * * — on both envs; the existing Coolify scheduled task (created DISABLED both envs, S312) is enabled at cutoverDECIDED 05/06/2026 (Liam, S314) — this register is the decision recordSigned off 05/06/2026 (Liam)
(b)datapath-watch cadence + thresholdsVercel cron app/api/cron/datapath-watch every 15 minutes (*/15 * * * * in vercel.json); STALL_THRESHOLD_MINUTES default 30. At cutover: set INGESTION_EXPECTED=true so silence-condition (C) goes liveRATIFIED S311 ({66.15} re-home as Vercel cron)Recorded as ratified 05/06/2026
(c)Alert routingSentry-only: MONITOR_ALERT_WEBHOOK_URL deliberately unset → logger.error Sentry bridge (datapath-watch alert sink); recordPipelineRun emits Sentry.captureMessage on failed (error) / completed_with_errors (warning) (lib/pipeline/record-run.ts)RATIFIED S311Recorded as ratified 05/06/2026
(d)Kill-switch drill + crash-loop backstopKill-switch order of convenience: Coolify control stop (MCP) / UI Stop / SSH docker stop; crash-loop backstop = lifespan-only boot (boot never walks) + restart: unless-stopped kept for availabilityDOCUMENTED — runbook §“B3 — crash-loop backstop + kill-switch” (docs/runbooks/onprem-b1-deploy.md)Recorded 05/06/2026; one-time DRILL flagged POST-go-live, not now

(a) /walk cadence — decision detail and rationale

Section titled “(a) /walk cadence — decision detail and rationale”

Decision (Liam, S314, 05/06/2026): nightly at 02:00 (0 2 * * *), both envs, enabled at cutover. Rationale as recorded with the decision:

  • The 02:00 walk clears before the 03:30 LMDB cold-backup host cron (30 3 * * *, runbook {66.14} — the schedule there already notes it runs “after the /walk 02:00 window”).
  • The walk is incremental (memo’d), so a no-change night is cheap.
  • The bearer-gated on-demand POST /walk remains available between scheduled runs.
  • Single-flight rejects overlap — a long walk cannot stack a second pass.

Pre-existing state this decision lands on: the bearer-gated POST /walk + boot-never-walks burn-valve shipped as ID-83 (done, S311); the Coolify scheduled task was created but left DISABLED on both envs (S312); COCOINDEX_SOURCE_PATH is set on both envs (burn-safe because boot never walks). Enabling the task at cutover is operator checklist item (f) in docs/runbooks/onprem-b1-deploy.md. This decision resolves OQ-1.

Do NOT build new monitors — the following are all already-covered / explicitly out of scope per the ID-61.8 recon binding list:

  • datapath stall monitor {66.15}
  • failed/completed_with_errors Sentry alerting
  • kill-switch + crash-loop backstop (runbook B3)
  • pipeline-runs UI panel
  • coverage-alerts cron
  • [RATIFIED-DO-NOT-BUILD] pipeline_failures table
  • [RATIFIED-DO-NOT-BUILD] Q4.12 cost dashboards