Repository navigation
the fragmentation ratchet's class-name and rule-count dimensions measure size, not duplication, so they cap new UI instead of ratcheting sprawl #15455
Description
Activity
- added a commit that references this issue
on Sep 1, 2026 - added a commit that references this issue
on Sep 13, 2026 What landed today (#16596, fcd03d0)
Progress note for this umbrella. Checked against
origin/mainat b688b3e. #16596 referenced this issue (Refs). Its body says it is only a baseline move after batch 2b, not a fix for the measure.AC Evidence on origin/mainStatus 1. Replace distinct_class_names/css_rule_declarationswith a duplication measureStill the size counts: repo_tests/frontend_fragmentation_ratchet_test.py:92(5619) and:94(9401). #16596 changed only these two figures.❌ 2. Scope inline_genericsaway from calls typed from the generated contractrepo_tests/frontend_api_contract_ratchet_test.py:194uses the same_inline_generic_count. #16596 moved its baseline 577 → 564 and reformatted two comprehensions; it did not change the measure❌ 3. A test shows the new measure rises on a duplicated rule and not on a new component None ❌ 4. Baselines re-derived from the new measure #16596 re-derived baselines from the existing measure ❌ What remains: all four ACs.
- added a commit that references this issue
on Sep 18, 2026
Problem
Two dimensions in
repo_tests/frontend_fragmentation_ratchet_test.py, and one infrontend_api_contract_ratchet_test.py, count size rather than duplication. Combined with the both-directions assertion they act as a cap, not a ratchet: any legitimate new component fails them, and the only way to pass is to delete something else.distinct_class_names.foo-barnamescss_rule_declarationsinline_genericsapi.*<Shape>call sitesThis was demonstrated rather than theorised. #15429 added a load-failure banner to three views. The banner's CSS was duplicated across all three — real sprawl, correctly caught. Consolidating it into
assets/main.cssremoved 67 lines and broughtcss_rule_declarationsdown, butdistinct_class_namesdid not move: merging three copies into one removes declarations, not names. The five names are new UI and no consolidation removes them.inline_genericshas the same shape. The fourapi.get<WorkItem>calls #15429 added are correct typed detail fetches. The metric counts them as unverified assertions, which they are — but so is every correct call.What works and should be kept
components_declaring_stylesheld at 381 through the same change, because styles were added to files that already had them. That is the duplication signal. The per-family counts (btn116,status114, …) behaved correctly too — they rise when a family gains another spelling of the same idea.Acceptance criteria
distinct_class_namesandcss_rule_declarationsare replaced by a measure that rises with duplication and not with legitimate growth. Candidates: rules declared in more than one file, or names differing only by a known synonym pair (modal/dialog,toast/alert).inline_genericsis replaced by, or scoped to, calls whose shape is not drawn from the generated contract — a call typed fromtypes/generated/api.tsshould not count.Notes
Until this lands the three dimensions stay, at their current figures, so they still catch a regression — they just cannot tell growth from sprawl, and anyone adding UI will have to raise them with a note. That cost is the reason to fix the measure rather than keep paying it.
Related: #12730, #12731 (the umbrella these ratchets serve), #12363, #15429 (the change that exposed this).