Business plugin composition audit

Summary

Third plugin audited against SPEC-145, after SPEC-147's storytelling and SPEC-148's places. Business is 602 lines across five runes and is the cleanest of the three by a wide margin: no custom content models, no postTransform, no pipeline hooks, no behaviors, no component overrides, no requiresParent.

Four of its five runes compose today. One is gated on SPEC-146, and only for two of its properties. And it is the first plugin with no misfiled capability — which is a useful negative result, because it shows the pattern the first two audits found is "at most one", not "always one".

It also forces a refinement to SPEC-148's method, which changed three verdicts in this audit alone.

Method refinement — path B splits in two

SPEC-148 D1 established the five resolution paths and the per-property unit. Applying it here showed path B ("properties from a node", the one composition breaks) is not one thing:

B-intrinsic — the value exists only as authored content. There is no attribute carrying it; the author wrote an H1 and the rune harvested it. event's and organization's headline / blurb via pageSectionProperties are this. Only SPEC-146 fixes them.

B-by-choice — the value originates in an attribute, and the rune's transform chose to render it as a node which the schema then stamps. cast-member's name, role and portrait are built as nameTag / roleTag / portraitTag from attrs.name / attrs.role / attrs.image; timeline-entry's label and date the same. A composition declaring the same attributes can map them through the bag — path A — and works today.

The distinction matters because it is the difference between "blocked" and "free", and three of business's five runes look blocked under the coarse reading and are not. Stated as a rule: a property is only genuinely blocked when its value has no attribute to fall back to.

The audit

cast — no schema at all

The parent declares sections, attributes (layout) and a content model, and no schema. Nothing to resolve, nothing to break. Composes today.

cast-member — Person, all three properties B-by-choice

Every value is already an attribute the author writes; only the rendering is a node. A composed cast-member maps them from the bag and publishes correctly with no SPEC-146.

One detail to carry over: attrs.image runs through resolveImageScheme, so placeholder: / icon: sources resolve to inline <svg> (SPEC-106). A composition placing {% slot %} or an <img> directly must keep that, or scheme sources silently become broken image URLs.

timeline — ItemList, children only

{ type: 'ItemList', lists: ['itemListElement'],
  children: { 'timeline-entry': { type: 'ListItem', property: 'itemListElement',
                                  generated: { position: 'index' } } } }

No properties on the parent at all. children is path E, which works across a composition boundary (measured in SPEC-145's worked examples), generated: index rides items.forEach((item, i) => …), and lists is a root attribute untouched by nesting. The rune's transform does build pageSectionProperties refs, but the schema table maps none of them, so they carry no schema weight. Composes today.

Also already declarative where plot was not: business/Timeline was one of the six runes using sequence: 'connected' in SPEC-147's count, so there is no postTransform here to retire.

timeline-entry — ListItem, both properties B-by-choice

organization — the only gated rune, and only partly

Its schema table is the variant form, { by: 'type', rows, fallback }, selecting a schema.org type from the type attribute. The source comment records why that form exists: "organization was already doing what this milestone proposes… by: 'type' says the same thing where a reviewer can read it, and keys the rows off the one list that also feeds matches, so the accepted values and the published types cannot drift."

So the part that looks hardest — a schema.org type that varies by attribute — is composition-safe, because row selection reads attrs. What breaks is the same pair that breaks in event: the name and description harvested from authored content.

Verdict

And there is no misfiled capability. Nothing in business needs a client lifecycle, a behavior, a custom content model or a pipeline hook. This is the first plugin that is purely domain content, so unlike storytelling (which sheds storyboard) and places (which dissolves around map), business retires cleanly and completely.

That is a useful negative result for the cross-audit finding: the pattern is a domain plugin is mostly domain content plus at most one misfiled capability — not "always one". Two of three had one; this one does not.

Decisions

D1 — path B is recorded as B-intrinsic or B-by-choice

Per SPEC-148 D1's per-property unit, with the split above. A bare "path B" verdict overstates the blocker by counting values that merely render as nodes, and it mis-stated three of this plugin's five runes before the distinction was drawn.

D2 — a B-by-choice property is mapped from the attribute in its composed form

Not preserved as a ref for fidelity's sake. The attribute is the value's origin; stamping a rendered span was the old transform's choice, and reproducing it would import a blocker the composition does not need.

D3 — business retires completely, staged as replace-not-delete

Same staging as SPEC-147 D1 and SPEC-148 D2: the composed implementations ship alongside until an explicit later decision. Nothing relocates, because nothing is misfiled.

D4 — resolveImageScheme behaviour is part of the contract, not an implementation detail

cast-member resolves placeholder: / icon: sources to inline SVG (SPEC-106). A composed replacement that drops it fails silently on exactly the sources that look most like ordinary strings. Any composed rune taking an image attribute inherits this.

Implementation notes, deliberately not yet work items

1. Compose cast + cast-member — unblocked, and the simplest schema-bearing composition available (one Person, three attributes). A better first proof than itinerary if a schema-emitting example is wanted.
2. Compose timeline + timeline-entry — unblocked; exercises path E and generated: index together, which nothing else audited does.
3. Compose organization — after SPEC-146; the variant schema form should be checked against a composed tree specifically, since selectRow's attribute read has not been measured across a boundary.
4. Carry resolveImageScheme into any composed rune taking an image attribute (D4).

Non-goals

• Deleting plugins/business/ — same staging as the sibling audits
• Auditing the remaining plugins (design, learning, media, docs, marketing)
• Changing organization's variant schema form; it is cited as already correct
• Resolving whether cast's carousel layout needs behaviour — it is a CSS scroll-snap track today and stays one

Acceptance Criteria

• cast, cast-member, timeline and timeline-entry exist as compositions with fixtures, publishing graphs identical to the plugin's recorded baseline
• Every B-by-choice property resolves through the bag in its composed form, asserted at path A rather than assumed (D2)
• organization composes after SPEC-146, with headline and blurb publishing and the by: 'type' row selection verified across a composition boundary
• timeline's generated: { position: 'index' } numbers entries identically in composed form, asserted on a list of three or more
• A composed rune taking an image attribute resolves placeholder: and icon: schemes to inline SVG (D4)
• plugins/business/ is unchanged and passing its own tests until an explicit removal decision (D3)
• Lumina's 150 lines of cast / timeline / organization CSS are re-keyed via contextModifiers with no unstyled regressions, or each accepted change is listed
• The rune authoring guide documents the B-intrinsic / B-by-choice distinction, since it decides whether a rune is blocked

References

• SPEC-148 — the places audit; establishes the per-property method this refines
• SPEC-147 — the storytelling audit; the first data point for the misfiled-capability pattern this one qualifies
• SPEC-145 — composed runes; the mechanism audited
• SPEC-146 — name resolution across boundaries; what gates organization
• SPEC-130 — the schema table and its applier; the five paths and the by / rows variant form
• SPEC-106 — image src schemes; the resolveImageScheme behaviour D4 preserves
• SPEC-150 — the design audit; the first plugin that stays a plugin, and the derived-data blocker
• SPEC-151 — the marketing audit; counts the pageSectionProperties headline pattern across the corpus, of which organization is one