SLO Definitions

Aligned with: PRD NFR-1 (Performance) Created: 2026-02-14 Owner: Platform Engineering

---

Overview

This document defines baseline SLO targets for the i4g platform. Cloud Monitoring alert policies (see infra/modules/monitoring/) fire when burn rates exceed these budgets over rolling windows.

---

API Latency

Endpoint Category	p50 Target	p95 Target	p99 Target	Budget Window
Search / List	< 500 ms	< 2 s	< 5 s	28 days
Case Detail	< 300 ms	< 1 s	< 3 s	28 days
LLM-backed (RAG)	< 2 s	< 5 s	< 10 s	28 days
Dashboard widgets	< 1 s	< 3 s	< 5 s	28 days

Error Rate

Service	Target	Measurement
API (overall)	< 1 % 5xx	HTTP 5xx / total requests
Ingestion job	< 5 % failures	failed records / processed records
Report job	< 2 % failures	failed reports / attempted reports
Dossier queue	< 2 % failures	failed dossiers / attempted dossiers

Queue & Job Health

Metric	Target
Review queue depth	< 500 unassigned items (alert at > 500)
Report generation duration	< 30 min per batch (alert if stuck)
Dossier generation latency	p95 < 10 min per dossier
Ingestion throughput	≥ 100 records / hour in steady state

Contact Decryption Access

Metric	Target
Contact decryption calls / user / h	≤ 10 (alert above threshold)
Unusual-access alerts / day	< 5 (investigate if sustained)

Mobile (Future)

Metric	Target
Dashboard load on 4G	p95 < 3 s
Image evidence download	p95 < 5 s

---

Monitoring Implementation

SLO monitoring is implemented via:

1. Structured log events emitted by the Observability and AlertingService classes in core/src/i4g/.

2. Cloud Monitoring log-based metrics that parse structured JSON logs for alert=true events, keyed by alert_type.

3. Terraform alert policies in infra/modules/monitoring/ that define notification channels, burn-rate windows, and escalation rules.

4. StatsD / OpenTelemetry metrics for latency histograms and counters, configurable via ObservabilitySettings.

Key Structured Log Fields

Field	Type	Purpose
alert	bool	true for alert-worthy events
alert_type	str	Category: contact_decrypt_access, ingestion_failure, dossier_stuck, dossier_failure
severity	str	warning or critical
component	str	Emitting service component

---

Revision History

Date	Change
2026-02-14	Initial SLO baseline (WS-8 / F51)