Implement a performance change

Performance work fails two ways, both producing a diff that looks like a win: a number that moved on the benchmark but not in production (measured under conditions that don't match where the system is slow, or against a baseline taken under a different protocol), and a speedup that quietly broke correctness — a faster wrong answer is still a defect. This guide adds the performance discipline on top of the base implement-task rules. These are conventions the review packet inspects — nothing enforces them at edit time.

Plan the transformation first — the workspace's change plan covers baseline, waves, and rollback; this guide is the execution half. Numbers, not vibes: nothing is faster until a baseline and a final figure, under the identical protocol, say so. No number and no protocol is tinkering, not optimizing — get the target before you start.

Before you start, open references/task-template.md and copy it into your task file — it is the session frame for this work; fill it in as you go (don't reconstruct the structure from memory). It scaffolds the baseline block, the target and ceiling, the hypothesis, the measurement protocol, the per-change ledger, and the self-review. The task packet itself uses the kit's task template.