--- name: parked-code-quality-audit description: "CLOSED — the maintainability/tech-debt audit ran and the owner's CSS-split question was answered and executed. Kept for the refactor discipline it established." metadata: node_type: memory type: project originSessionId: 0b7c5f41-bb34-4ab4-93be-fdb363f5eb40 --- # CLOSED — ran; the owner's open question was resolved Parked 2026-06-28 (owner: "Add 1-6") as a plan-only maintainability sweep — dead code, duplication, fragile patterns — producing a triaged refactor backlog. **The owner's specific question (asked 2026-07-04): should the stylesheet be split into multiple files the way the main UI script was split?** It was evaluated, answered yes, and carried out. Nothing here is pending. *Correction 2026-07-20: removed the study-file list, the prohibited-pattern enumeration, protocol numbers, the post-split file inventory, and the verdict scaffolding. Current structure lives in `library/CODE_MAP.md`; the rules live in `CLAUDE.md` + `rules/`.* ## The reasoning worth keeping **Why splitting a stylesheet here is not the same decision as splitting a script.** There is no build step and no module system — everything is global scope, served as-is. So a split is paid for at *load time*, not at build time: stylesheet `@import` chains serialize requests and cost real latency, whereas separate link tags load in parallel but must then be kept in sync with the service-worker asset list and the cache-bump discipline, and their order still matters because the cascade depends on it. The split was worth it anyway, but the tradeoff is a load-order and cache-invalidation cost, not a free reorganization. **Anyone proposing another split must price it the same way.** **Behavior-NEUTRAL means the test suite is the proof, not the intent.** The standing discipline for refactor work: if a refactor changes any test, it was not behavior-neutral — regardless of how obviously safe it looked. The suite is the definition of "no behavior change", which is why refactors are allowed to be aggressive at all. **Dead code has a specific tell here.** Retired features leave remnants that still *look* live — orphaned handlers, commented-out blocks, and stale test description strings that reference things which no longer exist. Stale descriptions are the worst of these, because they make the suite lie about what it covers. Feeds [[parked-synthesis-masterplan]].