The scorecard

c1352026-11-10

Every layer component against Mark's five checks, with its cousins and its resolution.

Source: c135 initial-prompt.md (Mark's five checks, in his order); c135 plan.md §The scorecard, §Resolution strategies, D1–D8 (D8 revised 2026-11-09: the waiting-list extractions happen now); brand/component_philosophy.md (the eight principles the checks measure against); qdash ui/LAYER-CANDIDATES.md (the waiting list); app/data/component-registry.ts (the names). The cell values are illustrative.

The five checks

modular · clean · documented · shaped · docs page. Mark's, in his order; one row per component, generated by a script, kept as a living page.

The catalog

23 groups, about 250 components, 193 demos with adminProps; released by tag (v0.9.37). Measured, not rebuilt.

The waiting list

StateEmpty, JsonTree, StatTile, Panel, Spark, BarList, Attention*, SkeletonRows and more, staged in qdash. Extracted now, ahead of their groups (D8).

The Opus track's 59 fixes

UxIcon id, a danger tone on UxPill, a logo-only AgentAvatar, highlight.js optimizeDeps. Merged into the census at Phase 1.

Component	Modular	Clean	Documented	Shaped	Docs page	Cousins	Resolution
UxButtonux · every consumer	passes	passes	no doc comments on props	passes	passes	UxIconBtn, UxAddBtn	keep
UxPillux · every consumer	passes	passes	no doc comments	passes	no danger tone (Opus list)	UxTag, UxBoxTag	keep; UxBoxTag merges into it, alias two releases
DataTabledata · qdash, design	passes	one arbitrary value	slots undocumented	passes	no state port; flavors not harvested	UxTableList	keep both, with a note: dense and roomy
DataMetricValuedata-metric family · qdash, design	passes	passes	no doc comments	passes	passes	DataStat, DataVal, DataCard	merge DataStat and DataVal into the family; alias and deprecate
DataSparkChartOlddata · no consumer in the scan	not assessed	hex colors	none	not assessed	passes	DataSparkChart	retire, after the scan confirms
FilterTriggerBarfilter · qdash, design	passes	passes	emits undocumented	passes	passes	DataListingBar (contains it); the filter-old group	keep; retire filter-old
NavVerticalSubmenunav · qdash	passes	provisional subGap prop	no doc comments	passes	passes	—	keep; drop subGap
StateEmptywaiting list → DataStateEmpty	not assessed	not assessed	not assessed	not assessed	not assessed	FilterNoMatches; the #empty slots	extract now
JsonTreewaiting list → DataJsonTree	not assessed	not assessed	not assessed	not assessed	not assessed	UxCodeBlock	extract now
StatTilewaiting list → DataStatTile	not assessed	not assessed	not assessed	not assessed	not assessed	DataStat, DataMachineMicro	extract now; joins the DataStat merge

Six ways to resolve a duplicate or a cousin

1Merge

two components show the same thing with different props: one survives, the other becomes an alias that warns

2Split

one component carries two purposes: a prop that changes what it is

3Alias and deprecate

consumers cannot move in one release; two minor releases, a console warning, a note (D3)

4Retire

no consumer uses it and no pattern needs it; checked against the scan, not memory

5Extract

qdash staged a library-grade piece the layer lacks; the waiting lists, now (D8)

6Keep

with a note when the overlap is deliberate: a dense and a roomy variant a pattern documents

Group order (D1): the most used first, one release per group (D7)

interactive, display, feedback

data

table, data-metric, data-machine

filter · nav · layout

the page furniture

form

inputs, selects, save bars

domain groups

actor, agent, policy, compliance, security, …; each shows a qdash flavor or is a retire candidate (D4)

icon · typography · prose

last; the least churn

passes the checkwork leftnot assessed yet (still in qdash)extracted from qdash now (D8)the five checks

The problem

The layer grew with the work, c92 through c128: duplicates and near cousins, shapes that predate the component philosophy, code without annotation, docs pages that show one default where the dashboard uses five flavors. The control plane web app (c134) will stand on this layer for years and starts from a blank slate, so the layer must already hold what it needs.

The core idea

Measure before changing. Every component gets one row against the five checks, plus who uses it, which cousins it overlaps and one of six resolutions; the census is ruled with Mark, then the groups are brought to five passes in D1's order, one release each. The cells above are illustrative: the script fills the real ones in Phase 0.

What is open

Release cadence during the cleanup, per group or fewer and larger (question 2); which families Mark already knows are duplicates, so the census starts from his list (question 4); the annotation lint's exact rule, written after the first group passes (D6).