Positive example — bounded measured pipeline
- Context: large inputs exceed a verified memory limit.
- Mode:
budget-constrained.
- Decision: measure peak memory, process bounded chunks, apply backpressure, and verify throughput and correctness.
- Why it fits: the implementation directly serves stated budgets.
Negative example — speculative cache
- Context: no latency target or measurement identifies repeated expensive work.
- Mode:
clarity-first.
- Bad decision: add a shared cache with invalidation and synchronization complexity.
- Better decision: keep clear computation and instrument it if performance becomes material.