Service operations
SLA & Support
Performance, uptime, throughput, scaling, support and incident response targets must be provider commitments validated under PoC/load test and documented contractually.
PoCProvider commitment required
p95Measure every module
TPSConcurrent + batch
ElasticVolume growth test
| Service metric | Definition | Prototype target | Acceptance evidence |
|---|---|---|---|
| Availability | Monthly KYB API availability excluding agreed maintenance | To be proposed | Provider SLA + monitoring export |
| Document module latency | p50/p95/p99 per document | Measured in PoC | Load-test report |
| External intelligence latency | Per connector and end-to-end | Measured in PoC | Trace / timing evidence |
| Batch throughput | Cases/hour at stated concurrency | Measured in PoC | Load-test report |
| Incident response | Severity-based acknowledgment and restoration | Provider response item | Support SLA |
| Model issue response | Accuracy regression / drift investigation | Provider response item | Model ops runbook |
Load-test scenarios
- Normal onboarding volume.
- Morning/evening peak.
- Bulk re-screening.
- One external connector slowed down.
- One connector unavailable.
- Large document and multi-page Akta.
- High EDD queue accumulation.
Operational dashboards
- Case throughput and queue depth.
- Module success/error/timeout rate.
- Connector availability and quota.
- Model confidence distribution and drift.
- Human-review SLA and override rate.
- Deletion/DSR job SLA.
- Security/authentication anomalies.