Skip to content
Live Status: Temporarily unavailable

SLOs & System Reliability

We publish our service level objectives so you know exactly what to expect. When we miss, here's what happens.

Our Service Level Objectives

These are the targets we hold ourselves to. Live component health is shown above and on the status page.

Manifest/API Availability

Target: 99.9%

Percentage of successful responses (2xx) from manifest and API endpoints over a 30-day rolling window.

Decision Latency (p95)

Target: <200ms

Time from cue receipt to pod decision completion, measured at the 95th percentile.

Pod Deadline Hit Rate

Target: > 98%

Percentage of ad pods that complete stitching before the playback deadline.

Invalid Manifest Rate

Target: < 0.01%

Percentage of manifests that fail HLS/DASH validation or cause player parse errors.

Tracking Integrity

Target: 100%

Exactly-once semantics: no duplicates, no missed events. Verified by reconciliation.

How We Measure

# SLI Calculation Pipeline
┌─────────────────────────────────────────────────────────────────┐
│                     Real-Time Metrics Flow                       │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  ┌─────────┐    ┌─────────────┐    ┌──────────────┐             │
│  │ Request │───▶│ Prometheus  │───▶│  SLI Bucket  │             │
│  │  Logs   │    │   Scrape    │    │  Aggregator  │             │
│  └─────────┘    └─────────────┘    └──────────────┘             │
│                                            │                     │
│                                            ▼                     │
│                     ┌──────────────────────────────┐            │
│                     │    30-Day Rolling Window     │            │
│                     │    ─────────────────────     │            │
│                     │    availability = good/total │            │
│                     │    latency_p95 = histogram   │            │
│                     │    error_rate = errors/total │            │
│                     └──────────────────────────────┘            │
│                                            │                     │
│                     ┌──────────────────────┴───────┐            │
│                     ▼                              ▼            │
│              ┌─────────────┐              ┌─────────────┐       │
│              │  Dashboard  │              │   Alerts    │       │
│              │  (Grafana)  │              │ (PagerDuty) │       │
│              └─────────────┘              └─────────────┘       │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

What Happens When We Miss

Automated mitigations kick in before humans need to act. Here's the protection stack.

Safe Mode

When error rates exceed thresholds, the system automatically enters safe mode: reduced fanout, strict deadlines, and slate fallback for ad breaks.

Trigger: Error rate > 5% or latency p95 > 200ms

Automatic Rollback

Configuration changes that cause SLI regression are automatically rolled back within 60 seconds of detection.

Trigger: SLI regression > 20% from baseline

Partner Quarantine

Partners with repeated timeouts or errors are temporarily quarantined with exponential backoff cooldown.

Trigger: 5+ timeouts in 60 seconds

Circuit Breaker

Per-partner circuit breakers prevent cascading failures. Trips after threshold breaches, recovers automatically.

Trigger: Failure rate > 50% over 10 requests

Live Status Snapshot

Status temporarily unavailable

For real-time status and incident updates, visit our status page.

View Status Page

Trust starts with transparency

Download our Security Pack for the full picture: architecture, threat model, compliance, and more.

Download Security PackVerification Guide