The federation has no layout layer, so seven of its eight deviations are about placement

Why Care?

Eight data-deviation declarations exist across the whole federation after nine Button migrations. Seven of them are about placement. Alignment, margin, grid column, stretch-in-a-flex-column. Exactly one is an actual appearance deviation.

That ratio is not a sign that members are undisciplined about layout. It is a sign that the escape hatch was the only door in the building, so everything walked through it — including the things that were never deviations at all.

The immediate cost was legibility. A catalog's Deviations section that lists five margin adjustments teaches its reader to skim, and skimming is precisely where the one real deviation hides. Rung 0 fixed that by ruling layout out of the deviation mechanism entirely. But rung 0 answers "where does this get recorded", not "who decides it" — and the second question is still open.

What is actually missing

packages/theme ships a ten-step spacing scale. packages/shared-ui ships three components: Button, ConfidencePill, ToggleHeader__PromptOrPackage--Icons.

Every one of those is a thing that goes inside a box. None of them is a box.

There is no Stack, no Row, no Cluster, no Grid, no Container, no Panel, no Toolbar. There are no container widths, no breakpoint tokens, no density scale, no rules about what a page's outermost frame looks like. The federation standardised the contents and left the containers to nineteen independent authors.

Counted, 2026-09-13

Two rows deserve to be read twice.

padding: var(--space-*) is zero. Not low — zero. The spacing scale was shipped deliberately and has been adopted by nothing except the components that shipped with it. A scale nobody consumes is not a standard, it is a proposal with good formatting.

Seventeen of twenty members have no responsive behaviour whatsoever. Not "responsive in a way we disagree with" — absent. Nobody has decided this is fine; nobody has decided it isn't.

What this is not

It is not drift, and the convergence loop's five verdicts do not apply to it. PROMOTE, CONVERGE, SANCTION, DEMOTE and WATCH all presume two or more members solving the same problem differently, where the work is picking a winner. Here nineteen members solved a problem the platform never posed. There is no candidate to promote, because nobody diverged from anything.

This is the same distinction the user drew about the members outside the registry: "We started with the Design System because there is cleanup to do. For the others, there's just nothing. Nothing needs to be cleaned, just needs to be implemented." Layout is a nothing-yet, not a mess. Twenty-one max-width values are not twenty-one bad decisions; they are twenty-one local answers to a question that was never asked centrally.

Reading it as drift would invert the governance model. diverge → promote → enforce starts with divergence, and divergence is only meaningful against a standard. Enforcing before exploring would skip both of the first two steps.

The interim ruling

Consumers keep layout. Deliberately, on the record, and with an expiry condition.

1. Layout is the parent's job (rung 0). A shared component never positions itself; the member aligns it, or a plain wrapper div does. This is already true and already cheap — every one of the seven cases costs zero or one declaration.

2. Layout is not a deviation and does not get a data-deviation. The Deviations section stays scarce enough to be worth reading.

3. No member is asked to adopt a spacing token it does not already use. The 912 raw paddings stay. Converting them before a layout layer exists would be churn that has to be redone once the containers arrive.

4. Raise, don't chase. Layout observations from migrations land here, not in the migration diff.

The expiry condition is the same forcing function the rest of the practice uses: when three independent members reach for the same layout shape, it stops being theirs. A Toolbar has already been sighted twice (scanbar rows in two members); a SelectableRow / list-row organ has been sighted four times across record-collector, prompt-template-manager, docs-portal and search-results, and is now the single most-cited missing organ in the federation. That one is arguably already over the line.

The path, when it opens

The right sequence is exploration → spec → plans, and none of it exists yet. Naming the shape so it can be picked up cold:

• Exploration — what is a layout layer for a federation of sovereign micro-frontends? The question is genuinely open and probably has a house-specific answer. Module Federation means a member is mounted into a shell it does not control, which makes a page-level container more of a negotiation than a component. Prior art worth reading rather than recalling: Every Layout's primitives, Tailwind's container/flow model, Radix's layout escape hatches, Open Props' sizing scales — all pinned in studies/frontend-ui-kits-component-libraries.

• Spec — the organ registry entry per shape, plus whatever the federation's equivalent of --container-* and breakpoint tokens turns out to be. Slots into Design System Convergence beside the control organs.

• Plans — a per-shape rollout, one organ at a time, the way Button went. The Button arc is now the proven template: author the primitive, prove it on one member, then one subagent per member with issues raised not chased.

Nothing above should start before Button is finished and Chip is planned. This file exists so the gap is recorded rather than rediscovered a third time.

Related

• Component API Contract And The Control Scale — rung 0, and the ladder it completes

• Adopt The Shared Button In One Member — where the seven deviations were written

• Converge The Federated Design System — the five verdicts, and why they don't apply here

• Design System Convergence — the organ registry this would extend

• The Federal Layer Never Shipped Space Radius Or Z — how the spacing scale got here

• Setup Microfrontend with Proper Monorepo Configurations — where a layout default would eventually be wired in