What problem does it solve? Performance work on a C++/CPython JSON engine often repeats failed experiments or accepts unmeasured changes. This Skill records every optimization that won (with commits and measured effects), every negative result, and ranked salvageable work, so engineers do not retry rejected ideas or accept changes without evidence. ## Core Features & Use Cases - Win catalog: Chronological tables of accepted techniques (hybrid SAX parsing, PGO+ThinLTO, SIMD escape scans, schema caches) with commits and measured effects. - Negative-results ledger: Failed experiments (schema recovery, file-newline writes, nested-mappings dispatch) preserved as patches with the exact gates they failed, preventing rework. - Measurement protocol: Rules of engagement requiring before/after make bench-small runs, ABBA blocks, A/A floors, and canonical regression gates before any optimization is accepted. - Use Case: Before attempting to speed up Strata's serializer, load this Skill to check whether the idea was already tried, what the measured outcome was, and which evidence gates a new candidate must pass. ## Quick Start Load the performance playbook before starting any optimization work on the Strata JSON engine and check the negative-results table for prior attempts.
