Correlation Monitoring

Individual metrics in isolation tell you that something changed. Correlated metrics tell you why. The monitoring system tracks 23 metric pairs with expected correlation coefficients, and flags when any pair diverges beyond expected range.

High-value correlation pairs in our system: throughput vs. quality (expected: weakly negative, r = -0.23 — higher throughput slightly decreases quality). Handoff completeness vs. downstream rework rate (expected: strongly negative, r = -0.71 — more complete handoffs mean fewer revisions). Message volume vs. coordination efficiency (expected: weakly negative, r = -0.18 — more messages slightly reduce efficiency). Escalation rate vs. quality scores (expected: moderately negative, r = -0.42 — more escalations correlate with lower team quality).

When a correlation breaks — say throughput vs. quality suddenly shows r = -0.67 instead of the baseline -0.23 — it means the system is under a stress it was not designed for. Investigate the cause before optimizing individual metrics. Improving throughput in a broken-correlation state will accelerate quality degradation.