Decoder. plain-English AI glossary

Metrics

● Core

Numerical measurements over time—latency, throughput, error rate, cost—aggregated for trend analysis.

Think of it like

Your heart rate monitor showing beats-per-minute over the day so you can spot workout vs rest patterns.

Example

Inference latency p50=100ms, p95=250ms, p99=500ms; throughput 1000 requests/sec; cost $0.001 per inference.

How it actually works

Time-series databases (Prometheus, InfluxDB) store high-cardinality metrics. Cardinality explosion (unique label combinations) makes storage expensive and queries slow. Rules aggregate (sum, rate, histogram) to make alerting feasible. Dimensionality (service, model, region) enables slicing but explodes cardinality.

For product teams

Quantifies system health and ROI—track latency, cost, and quality metrics to guide decisions.

For engineers

Instrumentation emits counters (requests), gauges (queue length), histograms (latency); time-series DB stores; query engine aggregates for dashboards and alerts.

Dashboards — Displayed in.
Alerting — Triggering.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome