The mapping brief — contract

stardust/reskin/mapping.md is the decision record between the two captures: every content slot from the content model assigned to a donor module from the enumerated vocabulary. It is authored in Phase 3, before any rendering, and it gates rendering: a page whose brief fails the bars below is not rendered.

The brief exists because the alternative — deciding layout slot by slot mid-render — is where silent improvisation happens. The validation experiment (H5) showed every mapping decision is decidable from the two artifacts alone: slot anatomy (content-model/<slug>/content-model.json) vs module anatomy (stardust/reskin/donor-modules.md). No render context needed, no judgment arbitrary enough to require a human.

Entry schema

One table row (or block) per slot, for every slot in the content model — no slot may be absent from the brief:

Plus a Stats block (mapped ratio, new-modules list), a Deltas block (chrome swaps + the normalization ledger entries), and a Judgment calls list. See the worked example that validated this contract: the experiment's mapping of 12 slots onto a 16-module donor vocabulary at 91% mapped, 1 new-module, 1 chrome swap.

Status semantics and gates

• mapped — the slot renders as a named donor module, possibly with a module transformation noted in the rationale (e.g. carousel slides → static card grid when the donor has no carousel vocabulary — a layout decision, not a content delta; the slides' text and images are all carried).
• new-module — the donor vocabulary has no module for this anatomy (the experiment's case: a search form on a donor with no search anywhere). Requirements:
  1. The entry names the donor tokens it composes — input spec, primary button spec, radii, spacing. A new module is donor vocabulary recombined, never a third design system.
  2. It appears in the Stats block's explicit new-modules list.
  3. Nothing is ever improvised mid-render: a renderer needing a module the brief doesn't grant stops, and the brief is amended first.
• chrome — nav / footer swapped for donor chrome carrying the source's links (labels and destinations preserved — the ../../migrate/reference/content-preservation.md rules apply to chrome too). Chrome slots are excluded from the content-gate scope and each swap is a documented delta (D1, D2, …) in the Deltas block.
• carried-invisible — non-visual content inside the byte scope: sr-only / visually-hidden a11y landmarks, skip-target headings. Real sites keep their only h1 here. None of the other statuses fit: it isn't chrome (it is inside the gate scope, so its bytes must survive), mapped is wrong (no donor module has an anatomy for invisible text), and new-module is wrong (a new-module entry must compose donor tokens; this composes none). Contract: the slot is carried verbatim into an equivalent visually-hidden element (the rendered page's own sr-only utility), stays inside the content-gate scope, and adopts no donor tokens. Note the heading level carried — an sr-only h1 is still the page's h1.

Gate: mapped ratio ≥ 80% over content slots (chrome and carried-invisible both excluded from the denominator — one is out of scope, the other has no module to map to by definition). Below 80% the donor vocabulary doesn't cover this content — stop per SKILL.md § Stop conditions: widen the donor capture (more donor pages usually means more modules) or get explicit sign-off on a larger new-module list. Do not lower the bar silently.

Slot splitting

Section ≠ slot. When one captured slot holds two logical regions (the experiment's single fullwidth div holding News and Events), split it into sub-slots in the brief (s07-news, s07-events) so each mapping decision is atomic. The split is a brief-level construct — the content model is not edited — and slot-coverage still asserts the parent slot's full text, so a split can't shed content.

Casing / text-transform policy

Declare it once in the brief; it binds the renderer:

• Captured text is rendered-case (innerText reflects text-transform — see reference/content-model.md § Rendered-case text). The byte gate compares rendered text on both sides.
• The underlying text is never edited to change case. Casing is applied via CSS text-transform only, mirroring what the source rendered.
• When source casing and donor style collide (uppercase-tracked source eyebrows vs a donor's sentence-case eyebrows), byte fidelity wins: keep the rendered casing, adopt the donor's eyebrow color/weight/size tokens. Note the tension as a judgment call in the brief.

Icon-ligature text

Icon fonts of the Material Symbols family put the ligature name in the DOM as text — a source "more" link is literally chevron_right Read more in innerText, so the ligature name is in the content model, in the byte-gate reference, and (via the ordered stream's text nodes) in whatever the renderer emits. Decide its treatment in the brief, per ligature family, one of:

• carry as text (the default — byte fidelity): the renderer emits the ligature name in a marked span; without the icon font it renders as the literal word. Honest, gate-clean, ugly — name it as a judgment call and as the eyeball verdict's weak point if it shows;
• carry + adopt the icon font: same bytes, but the span sets the icon font family so the ligature renders as the glyph. Only if the font's license allows (same rule as donor fonts);
• normalize it away: a shared-ledger entry (N*) that strips the icon text nodes on the source side — then the renderer must not emit them either. This is a content removal, so it takes an explicit ledger entry with a why; never just edit the strings.

What is never allowed: silently dropping or rewording the ligature text in the renderer — that is a byte-gate failure by construction (reference/gates.md § Failure modes).

Deltas block

The single audit surface for every deviation between the live source page and the gated comparison:

• D* chrome swaps (what was swapped, what carries over, why it is out of gate scope);
• N* normalization ledger entries, copied from the page ledger (stardust/reskin/normalize/<slug>.mjs) — the brief cites them, the ledger executes them.

If a delta is not in this block, it does not exist: any other source↔render difference is a gate failure, not a delta.

Flag, don't fix

Source garbage discovered while mapping — broken JSON-LD URLs, wrong-locale labels, duplicated links — is carried verbatim (fidelity over repair) and flagged in the brief's judgment-calls list for the human. Silently fixing it breaks the byte gate; silently keeping it without a flag ships a known defect unreviewed.