go-turbo-improvelisted

# go-turbo-improve Apply the fix, keep the behavior, prove the win. Two things are easy to ship accidentally when someone says "just make it faster": an optimization that changes behavior is a bug, and an optimization without a measurement is a guess. The procedure exists to catch both. ## Procedure Copy this checklist into your response and tick items as you complete them. Each unticked box is a way to ship a performance claim you cannot defend: ``` Fix progress: - [ ] Step 1: Diagnosis in hand (which path, which rung) - [ ] Step 2: Baseline captured to old.txt, tests green - [ ] Step 3: One coherent change at the highest applicable rung - [ ] Step 4: Behavior verified (go test, -race if concurrent, go vet) - [ ] Step 5: new.txt captured, benchstat run - [ ] Step 6: Reported the number, reverted unmeasurable complexity ``` **1. Know what you're fixing.** Work from an existing diagnosis — from `$go-turbo-analyze`, a profile, or the user. Without one, spend the first step finding the bottleneck; optimizing the wrong function is pure cost. **2. Baseline first.** Before touching anything: ```sh go test ./... go test -bench='BenchmarkTarget$' -benchmem -count=10 -run='^$' ./pkg > old.txt ``` Where no suitable measurement exists, add the smallest one that answers the question: a focused benchmark for local code, or a trace/load test for a queueing, network, or system effect. A microbenchmark is one valid form of evidence, and it supports only a microbenchmark claim. **3.