Augment It
Search/

← Context/ plans

0.0.1.0Partially-Shipped

Tidy every consumer onto the shared primitives

Button is done across nineteen units. This sequences what is left: Chip everywhere, the stale annotations, and the three containment jobs that are not component work at all.

Where this starts

Button is adopted in 19 of 19 units — 297 call sites, 39 declared holdouts, 824 net lines of CSS deleted, and not one member needed an override rung above variant+size. The loop that did it has been run nineteen times and revised four times from its own engineers' reports.

The federal layer moved underneath it today too: the Tier 1 palette collision is repaired, destructive has a boundary, ghost and outline have a surface-independent hover, and the type scale and info tone shipped — both of which had been proposed in DESIGN.md and never written to theme.css.

So the primitives are real. This plan is about the consumers.

Phase 1 — Chip, everywhere (this wave)

Twelve units carry chip-shaped markup, across roughly forty distinct class names (cc-tag, cc-pill, ow-chip, cr-domain-chip, srq-chip, badge, source-badge, pe-affiliation-pill, …).

Executor: Adopt The Shared Chip In One Member. One agent per group, one commit per member, issues raised not chased.

GroupUnits
Aresponse-reviewer, search-results
Bcorpora-curator, sort-filter-lens
Corg-workbench, record-collector
Dchat, pack-runner, search-and-add
Edocs-portal, enhanced-records-list, shell

The classification risk is the whole risk. A filter chip that toggles a filter is a Button that happens to be chip-shaped, and nineteen members already made that call. The most likely way this wave does damage is by converting interactive controls back into spans. The loop leads with the decision tree for that reason.

Phase 2 — the stale annotations (this wave, folded in)

Seven data-deviation declarations are now wrong. They live in person-enrichment, records-surface and search-and-add, all migrated before rung 0 existed, and every one says some version of "the ladder has no rung for layout." It has one now.

These are not defects — they are annotations that outlived their reason, and they are actively harmful because the Deviations catalog is supposed to be short enough to read. Remove the attribute; keep the layout fix; if the fix is a member class on a wrapper, that is still rung 0 and still not a deviation.

After this, the federation's deviation count should be 1.

Phase 3 — containment, which is not component work

Three jobs that keep surfacing in migration reports and do not belong in a component diff. Named here so they stop being rediscovered.

  1. docs-portal styles its guests. It mounts other members' galleries into its own document while carrying bare section / h2 / code selectors and generic .cell / .chip / .grid classes. A member can be entirely inside its own boundary and still restyle its neighbours. (The shell, by contrast, was measured clean: 141 selectors, every one hashed.)

  2. sort-filter-lens leaks 49 class names. Down from 61 incidentally via the Button work. The live-collision names — .row, .error, .muted, .empty — are all still there. Its own refactor doc has the phases.

  3. Bare element selectors, federation-wide. .X-app h2, code, pre, textarea, input[type=file]. Each reaches past every component boundary beneath it. One of these was silently widening every control in record-collector by 8px.

Phase 4 — the next organs, in evidence order

Not started. Listed so the ordering argument is on the record rather than re-litigated.

OrganSightingsNote
ListRow / SelectableRow5+ membersFar past the three-member threshold. The single most-cited holdout of the Button rollout: full-bleed, wrapping, variable-height rows measured at 39, 56, 89, 93px against a 32px control
MenuItem3response-reviewer, shell, sort-filter-lens — and two of them declare role="menu" with no menuitem children and no arrow keys
StatusIndicator16 units observe connection_statusRendered eleven ways. Phase 1's Chip covers the label; the widget is separate
Combobox2Tag-suggest surfaces with no listbox semantics at all
ToggleGroup2Segmented controls; also where confidence-as-a-control would live

Standing decisions, so nobody re-opens them

  • Layout stays with consumers. The Federation Has No Layout Layer. Not drift — the platform never posed the question, so there is nothing to promote. Expires when three members reach for the same shape; the list-row organ is already past it.

  • Literals are ignored for now. F8 reports one hit per file, and var(--token, #fallback) is a documented member convention. Fixing it turns 72 into ~700.

  • F7 has no mechanical check, and every control-boundary defect this week was found by an engineer with a browser. The contrast gate walks 30 text pairs. This is the highest-value gate to build and it is not in this plan.

Decisions saved for the operator

Flagged rather than blocking; the wave proceeds with the stated assumption.

  1. --color-surface in dark mode. outline's border measures 2.73:1 on it, under F7's floor. Fixing the palette collision did not move this, because it is a property of graphite-700. Changing --color-surface touches every member. Assumption: leave it; outline is rare on that surface.

  2. Is ConfidencePill a Chip tone? It predates Chip and speaks a vocabulary Chip deliberately lacks. Assumption: keep it separate; do not merge.

  3. The 1px type-scale cost. Folding 12px into --text-body (13px) increases dense table text and can reflow fixed-width columns. Assumption: the scale ships as DESIGN.md specifies; members snap deliberately, not in this wave.