Algorithm Optimization

Use when

• The task is to speed up or scale an existing implementation.
• You need to reduce memory pressure, improve asymptotic behavior, or remove practical bottlenecks.
• The work should be benchmarked and justified rather than guessed.

Operating defaults

• Measure before optimizing.
• Start with the actual bottleneck, not the most interesting code.
• Preserve correctness with regression tests and reference outputs.
• Consider asymptotics, constant factors, memory locality, and parallelism in that order of impact.
• Make benchmark inputs representative and reproducible.

Workflow

1. Define the performance target and workload shape.
2. Profile or benchmark to locate the dominant cost.
3. Remove algorithmic inefficiencies before micro-optimizing.
4. Re-measure after each meaningful change.
5. Record tradeoffs in readability, memory, portability, and maintenance.

Quality bar