performance-metrics

name: performance-metrics description: Exposes performance metrics endpoint and implements load testing to verify p99 < 100ms and 50 txn/sec throughput NFRs.

Performance Metrics and Load Verification

Overview

This skill defines how to expose measurable performance indicators and verify non-functional requirements through repeatable load testing.

It focuses on proving, not assuming, that transaction evaluation meets p99 latency and throughput targets from the challenge specification.

When to Use

Use this skill when:

• Implementing a metrics endpoint for runtime performance visibility
• Computing latency percentiles from evaluation results
• Measuring sustained throughput under load
• Creating reproducible load-test scripts and acceptance checks
• Documenting evidence that NFR targets are met

Key Concepts

• Latency Percentiles: p50/p95/p99 computed from evaluationDurationMs
• Throughput: Transactions processed per second over a measurement interval
• Metrics Endpoint: API route exposing operational counters and latency statistics
• Load Test Scenario: Controlled traffic profile (for example 50+ txn/sec for 60s)
• Acceptance Gate: Explicit pass/fail criteria aligned to NFR thresholds

Guidelines

• Implement GET /metrics (or GET /health/metrics) returning JSON with p50/p95/p99 latency and throughput
• Include at minimum: active rule count and alert count in the metrics payload
• Compute metrics from persisted evaluation data using deterministic query windows
• Add load test script using k6 or Artillery under scripts/ with documented execution steps
• Configure load test to sustain at least 50 txn/sec for at least 60 seconds
• Add explicit assertion/check for p99 latency below 100ms during test run
• Document baseline hardware/environment assumptions for fair interpretation

Explicitly Forbidden

• Claiming NFR compliance without measured evidence
• Using one-off manual checks without reproducible script
• Mixing warm-up and measurement windows without clear separation
• Reporting averages only when percentile requirement is p99
• Hardcoding fake metric values in endpoint responses

Design Rules

• Keep metrics read path efficient and index-aware for production-like usage
• Maintain clear separation between business evaluation logic and observability layer
• Prefer additive API design: metrics endpoint MUST NOT break existing endpoints
• Time windows for metric computation SHOULD be configurable
• Load scripts SHOULD be deterministic enough for repeated CI/local comparison
• Store and share test outputs as auditable evidence of compliance

Expected Outcome

After applying this skill:

• Performance metrics are available via API for real-time inspection
• Load tests can be run repeatably from the repository
• p99 latency and throughput targets are verified with evidence
• Performance non-functional requirements are objectively validated