Optimization Pattern — Problem
Pattern: Optimization
Component: problem.md
Version: 1.2 | Updated: 2026-07-29
Statement
Performance work ships on intuition: no baseline, many variables changed at once, p50 celebrated while p99 or correctness quietly regresses.
Measurable symptoms
| Symptom | How you detect it |
|---|---|
| No baseline | Cannot prove improvement |
| Multi-knob PRs | Cannot attribute the win/loss |
| Correctness break | Faster path fails oracle |
| Contract blind | p99/cost SLO ignored |
Root cause
Optimization prestige rewards visible activity. Measurement and oracles feel like bureaucracy until a silent break hits production.
What breaks when this pattern is skipped
- Perf “wins” roll back after incidents
- Eval/quality gates fail after hot-path edits
- Arguments replace dashboards
Non-goals of this pattern
- Not premature micro-opts without a measured hotspot
- Not a substitute for algorithmic redesign when complexity dominates
- Not auto-tune via MCP in production
Related
Changelog
- 1.2 (2026-07-29): Optimization problem page expanded with skip-impact and operator symptoms.
Version: AIES v1.0.0✏️ Edit this page on GitHub