Engineering Reliability Metrics Operational

Latency (Response Time)

Latency (Response Time): Time taken to respond to requests.

Formula

P50 Latency = Median response time
P95 Latency = 95th percentile response time
P99 Latency = 99th percentile response time

Computable definition

The canonical formula in the GASP formula grammar. The MCP server compiles it to SQL over the datasets listed below.

percentile(field.response_time_ms, 0.95)

Unit: milliseconds

P50 variant: percentile(field.response_time_ms, 0.5)
P99 variant: percentile(field.response_time_ms, 0.99)

Notes: Canonical value is P95. Variants p50 and p99 give the other percentiles engineering.md lists. Report percentiles rather than averages, which hide tail latency.

Data requirements

Grain: Per service per interval, aggregate daily · Frequency: Daily · Aggregation: aggregate (day) · Staged: no · Main source: Monitoring.

Field	Dataset	Type
`period_start`	service_metrics	date
`response_time_ms`	service_metrics	integer

Example SQL (Postgres)

SELECT (SELECT PERCENTILE_CONT(0.95) WITHIN GROUP (ORDER BY response_time_ms) FROM service_metrics WHERE period_start >= :period_start AND period_start <= :period_end) AS latency_response_time

From an agent: with the GASP MCP server, call lookup_metric({ "name": "Latency (Response Time)" }) or generate_query({ "metric": "Latency (Response Time)", "adapter": "<your adapter JSON>" }).

Benchmarks

P50: < 100ms
P95: < 500ms
P99: < 1000ms

What It Tells You

User experience. Slow responses impact satisfaction and conversion.

← Back to Engineering metrics