Pipeline Health Monitoring

A pipeline that fails silently is worse than a pipeline that does not exist — it produces false confidence in stale or incomplete data. Pipeline health monitoring tracks four categories: collection health (are sources being reached and returning data?), processing health (are normalization and analysis completing without errors?), quality health (are credibility scores distributing normally and contradiction detection functioning?), and delivery health (are briefs being produced and consumed on schedule?). Each category has automated alerts that fire when metrics deviate from normal ranges.

## Pipeline Health Metrics

COLLECTION HEALTH
- Sources active: [N] / [Total] (alert if < 90%)
- Collection success rate: [%] (alert if < 95%)
- Sources stale (>2x expected update interval): [N]
- New data volume (24h): [N records] (alert if < 50% avg)

PROCESSING HEALTH
- Normalization queue depth: [N] (alert if > 1000)
- Normalization error rate: [%] (alert if > 2%)
- Analysis completion rate: [%]
- Dead letter queue depth: [N] (alert if growing)

QUALITY HEALTH
- Contradictions detected (7-day): [N] (alert if zero)
- Credibility score distribution: [normal / skewed]
- Synthesis findings per cycle: [N]
- AI-flagged hallucination rate: [%]

DELIVERY HEALTH
- Briefs produced on schedule: [Y/N]
- Consumer consumption rate (24h): [%]
- Urgent alerts delivered < 1hr: [Y/N]