ONSET

RELIABILITY · OPERATIONAL REALITY

Our committed numbers, our actual numbers, and what we do when something breaks.

The SLA tells you what you can hold us to. This page tells you how we actually run — the budgets, the cadence, the playbook. Verifiable against status.onset.my.

  • 99.9% Growth
  • 99.95% Enterprise
  • RPO <24h
  • RTO <4h
  • DR drills quarterly

COMMITTED TARGETS · 12 NUMBERS

Twelve numbers, no hedging.

  • Uptime target (Growth / Scale)

    99.9%

    ~43 min/month max downtime

  • Uptime target (Enterprise)

    99.95%

    ~21 min/month max downtime

  • P95 latency budget — chat reply

    < 3.0s

    Enforced on 24h rolling window

  • P95 latency budget — voice TTS

    < 800ms

    Real-time conversation quality

  • Snapshot retention

    30 days

    Encrypted · AWS ap-southeast-1

  • Recovery point objective (RPO)

    < 24h

    Worst-case data loss in DR

  • Recovery time objective (RTO)

    < 4h

    Time to restore service in DR

  • Incident response — SEV0

    < 15 min

    War-room declared + comms started

  • Postmortem deadline

    < 7 days

    Blameless writeup + action items

  • DR drill cadence

    Quarterly

    Restore + smoke test, logged

  • Quality sampling

    5% / hourly

    Per-module response audit (quality monitor)

  • Cross-tenant isolation tests

    Weekly

    Automated RLS assertion

WHEN SOMETHING BREAKS · 6 STAGES

Detect → Declare → Contain → Communicate → Recover → Learn.

Every SEV0/SEV1 follows this sequence. No improvisation, no heroics — the playbook is in tasks/incident-runbook.md and every on-call engineer can recite it.

  1. 1

    Detect

    Automated health checks flag the regression and auto-pause the affected module. 15-minute mean-time-to-detect target.

  2. 2

    Declare

    On-call engineer opens the incident in audit_log, starts a Telegram war-room channel, and posts the SEV class.

  3. 3

    Contain

    Isolate the affected surface — revoke compromised secrets, snapshot logs, freeze pipelines that depend on the broken module.

  4. 4

    Communicate

    status.onset.my updated within 30 minutes of declaration. Affected tenants emailed within 1 hour. SEV0/SEV1 also WhatsApp-pushed.

  5. 5

    Recover

    Restore service in priority order. Confirm via independent quality replay before un-pausing the module.

  6. 6

    Learn

    Blameless postmortem published within 7 days. Action items tracked in tasks/lessons.md. Same incident never happens twice — that is the contract.

WHERE TO VERIFY

Don't take our word for it.