"What if…" candidates

A catalog of eight directional bets that uplift Phase 2b selects from when authoring stardust/uplift-questions.md. Each candidate is a captured-but-underused brand trait that variant B or C can amplify.

The eight candidates are the worked-example floor, not a ceiling (see § Extension rule). What anchors the agent is not a closed list but the evidence shape every candidate must carry: the failure mode of "what if…" questions is improvisation toward editorial vocabulary regardless of the captured brand register, and the evidence shape is what keeps improvisation out.

The eight candidates

1 · Display-typography amplification

What if we extended <captured display family> from eyebrows into the page's structural voice?

Triggering signal. The captured surface carries a display font that's reserved for very small uses (eyebrows, all-caps section labels, badge text) while body / headings use the workhorse family. Detected by inspecting _brand-extraction.json.type.families.display usage count vs. its capability — if the display family ships axes or weights that the captured site never exercises, the family is underused.

Direction. Promote display type into section headers, navigation labels, and key headlines. The captured family stays the same; its scope changes.

Natural register for C. kinetic-display.

Disqualified when. The captured display family is generic (Inter, Roboto, Helvetica vanilla) without a signature axis — no trait worth amplifying.

2 · Photography re-foregrounding

What if the captured photography breathed at editorial scale?

Triggering signal. The captured site has high-quality photography but the layout crops it to thumbnail size — 3:4 portrait cards in a 5-up rail, 16:9 wide tiles with thin captions, hero photo behind text without enough share-of-canvas. Detected by inspecting pages/<slug>.json.images for naturalWidth/Height of captured assets ≫ rendered card sizes.

Direction. Restore photography to 3:2 or 4:3 with title overlays in the lower band. Hero photo becomes the layout, not a tile. Cards gain photo-first composition.

Natural register for C. editorial.

Disqualified when. Photography is generic stock with no identity (e.g. Unsplash-generic team photos, smiling-customer templates). Amplifying generic photography exposes the weakness.

3 · Live-data promotion

What if real-time operational signals were hero priority?

Triggering signal. The captured site exposes live data (security wait times, inventory levels, departure boards, status indicators, queue counts) but the live data lives below the fold. Detected by inspecting pages/<slug>.json.sections for operational-language signals (numbers + time stamps + per-row identifiers) that don't appear in the hero region.

Direction. Promote the live-data band into the hero adjacency. Tickers, dashboards, count-up numerals, refresh-pulse indicators. Often pairs with a configurator-search composition.

Natural register for C. live-systems.

Disqualified when. The captured site has no live data — marketing brand surfaces without operational signals can't fake liveness.

4 · Signature-gesture extension

What if <captured signature gesture> extended from one surface into a system?

Triggering signal. The captured site has a defining gesture — a giant monogram, a stadium identity, a transport wordmark, a custom illustrated motif — appearing only at the hero. Detected by inspecting _brand-extraction.json.motifs.wordmarkAsMotif or signature SVG / illustration appearing on pages/<slug>.json's landmarks[hero] but absent elsewhere.

Direction. Extend the gesture's vocabulary into section headers, dividers, accent surfaces, footer wordmark. The gesture is no longer an exception; it's the brand's structural voice.

Natural register for C. arrival (composition-led) or kinetic-display (motion-led).

Disqualified when. There is no captured signature gesture — the page reads as a template without a defining mark.

5 · Voice-register pivot

What if <captured underused tone> was foregrounded over <captured dominant tone>?

Triggering signal. Voice samples in _brand-extraction.json.voice carry a tonal contrast — formal + casual, civic + warm, technical + plain. The captured page leans heavily on one side and uses the other only in margins (footnotes, support copy, employee-program blurbs).

Direction. Foreground the underused tone in section heads and primary CTAs. The dominant tone retreats to legal / governance / secondary surfaces.

Natural register for C. editorial.

Disqualified when. Voice samples are uniform — no tonal contrast to lean into.

6 · Color-ladder re-weighting

What if <captured reserved shade> became the primary surface?

Triggering signal. The captured palette has a deep / accent / muted shade that's used in 1–3 places (a button hover state, a single footer band, an icon fill) but never as a primary ground. Detected by inspecting _brand-extraction.json.palette.roles for shades whose computed count is < 10% of the dominant background.

Direction. Promote the reserved shade to one or two large surfaces (a band, a header, a footer pre-band). The palette is unchanged — only its proportions shift.

Natural register for C. Any (color re-weighting is composition, not motion).

Disqualified when. The captured palette is monochromatic + neutrals only — no shade to lift.

7 · Audience-routing reframe

What if the page foregrounded <one captured audience> over the multi-audience compression?

Triggering signal. The captured page tries to serve multiple user intents in one composition (e.g. an airport homepage serving departing-today + arriving-today + parking + trip-planner + access

• shopper + civic — seven intents, one surface). Detected by inspecting pages/<slug>.json for > 4 distinct CTA verbs in the above-the-fold region.

Direction. Pick the most operationally-critical audience and foreground its path. The other audiences stay accessible but are demoted below the fold. The IA changes; the content set does not.

Natural register for C. live-systems (when the foregrounded audience is operational) or arrival (when civic).

Disqualified when. The captured page serves a single clear audience — no compression to untangle.

8 · Motif vocabulary swap

What if the layout primitive switched from <captured primitive> to <captured alternate>?

Triggering signal. The captured page uses one composition primitive heavily (e.g. cards-in-a-grid, badges-in-a-row, chips-in-a-pill-rail) while a different primitive appears once or twice (e.g. a list with index numbers, a banner with overlay text, a panel with a side rail). Detected by inspecting _brand-extraction.json.systemComponents for under-represented patterns.

Direction. Promote the alternate primitive to the layout default. The card grid becomes an indexed list; the badge row becomes a labeled rail; the chip pill becomes a panel-and-rail.

Natural register for C. kinetic-grid.

Disqualified when. The captured page has only one composition primitive — nothing to swap to.

Extension rule

An out-of-catalog candidate — recorded as derived — is admissible when it carries the same evidence shape as the eight worked examples above:

1. ≥ 2 captured citations — concrete pointers into the capture (_brand-extraction.json, pages/<slug>.json, brand-review.html tension IDs) that make the trait real, not imagined.
2. An explicit disqualification test — the condition under which the candidate must NOT be picked, stated up front the way every catalog entry states its "Disqualified when."
3. The variant role it serves — B's composition bet or C's cinematic bet, with a Natural register for C declared when it serves C.

Record each candidate's source in uplift-questions.md as Source: catalog or Source: derived. A derived candidate missing any of the three parts is improvisation and is refused.

Selection procedure (used in Phase 2b)

1. Walk all eight catalog candidates, plus any derived candidates admitted per § Extension rule. For each, read the triggering signal and check the captured brand surface for the trigger.
2. Mark disqualifications. A candidate is disqualified when its disqualification clause fires. Record the reason in uplift-questions.md so the audit trail is preserved.
3. Rank remaining candidates by trigger strength: how underused is the trait? How clearly does the captured evidence support amplification?
4. C picks first. The cinematic register was selected in Phase 3a; C takes the candidate that register naturally amplifies (per each candidate's Natural register for C field above — the single source of truth for the register→candidate mapping).
5. B picks from remaining. Prefer candidates whose move is composition / IA / voice (not motion) so B and C differentiate by axis, not intensity.
6. A needs no candidate. A is the faithful + improvements variant; its bet is the improvements list, not a trait amplification.

If after disqualification fewer than two candidates remain, the brand surface is too thin for three differentiated variants — trigger the stop condition (uplift SKILL.md § Stop conditions (b) / (d)) and switch to --two-variants (A + C) or refuse.

Format of uplift-questions.md

---
_provenance:
  writtenBy: stardust:uplift
  writtenAt: <ISO-8601>
  againstInput: <URL>
  readArtifacts:
    - stardust/current/_brand-extraction.json
    - stardust/current/pages/<slug>.json
    - stardust/current/brand-review.html
---

# "What if…" candidates — <URL>

## Picked

### Variant C · <candidate name>
Source: catalog | derived
What if: "<verbatim "what if…" phrasing>"
Cinematic register: <register>
Evidence: <captured citation, e.g. "_brand-extraction.json#type.families.display
shows 'Sharp Grotesk Semibold' used in 5 elements only">
Motion bet: <one-line>

### Variant B · <candidate name>
Source: catalog | derived
What if: "<verbatim "what if…" phrasing>"
Captured trait amplified: <name>
Evidence: <citation>
Composition bet: <one-line>

## Disqualified

- **<candidate name>** — disqualified because <reason>.
- **<candidate name>** — disqualified because <reason>.
...

## Considered but not picked

- **<candidate name>** — strong trigger but B/C differentiation
  better served by the picked candidate.

The disqualifications section is mandatory. Listing what was excluded (and why) is the audit trail that proves Phase 2b walked the catalog rather than reflexively picking the first match.