coo-service-level-misses

Service-level misses

Every miss is a risk to activation and to paying customers. Classify it, get the right owner moving, tell users clearly, and fix what repeats.

Look up first

• The service levels and red thresholds: excom_knowledge_search "service levels". If missing, use the defaults in the weekly ops review skill, label them "default", and propose them to the owner.
• Who executes production fixes (operator, CTO) and how to reach them: excom_knowledge_search "escalation" or "on-call".
• The user reply tone and the support mailbox: excom_knowledge_search "communication patterns".

Steps

1. Detect:
   • melaya_run_diagnosis and melaya_eval_report through melaya_tools.
   • cloudflare_traffic_analytics by status.
   • melaya_approval_list.
   • melaya_pipeline_trigger for paused reasons.
   • The support inbox.
   • grafana_list_firing_alerts or sentry_list_issues, if granted.
2. Set the severity:
   • SEV1: down for all users.
   • SEV2: a major feature degraded for many users.
   • SEV3: a minor feature affected for some users.
   • SEV4: low impact. Signup, checkout and payments are at least SEV2 while a customer outcome is open.
3. Find the likely cause before proposing anything:
   • Edge 520 and 522 mean the edge could not reach the origin. 524 means a wait over 100 seconds.
   • App responses carry CORS headers; edge error pages do not, so a "CORS error" is often an edge error.
   • Correlate with the last deploy (your code host's commits) and with firewall events (cloudflare_firewall_events).
4. Route the fix with excom_send_handoff (kind request), with the run id kept inside the team:
   • Production to the CTO and the operator.
   • Support patterns to the customer experience role.
   • Vendor blockers to the supply role.
   • Possible credential exposure straight to the security role, never quoted.
5. Draft the status update and the user reply (zoho_email_save_draft or gmail_create_draft): plain text, what happened, what was done, no internal ids.
6. After resolution, write a blameless postmortem using references/postmortem-template.md, and save it with excom_save_draft.
7. Record the miss (service level, duration, severity) with excom_record_measurement. A third miss of the same kind becomes a fix proposal (excom_propose_task), and a pipeline candidate if recovery is manual.