/nexus

Marquee Blueprint — the lp domain preset of wish

/nexus marquee "<product/offer>" ≡ /nexus wish domain=lp — the lp domain preset of the wish quality-ceiling engine. marquee is a named alias kept for discoverability; the engine (crystallization → tournament → gauntlet-converge → One-Shot Gate, termination bound, Q9 separation) is owned by reference/wish-recipe.md and is not re-derived here. This file owns only what the preset fixes: the 5-dimension LP Ceiling Rubric with machine oracles, the GROUND and PRODUCE phases, the LP skeptic angles, and the Marquee Dossier.

Read this file before executing the marquee Recipe.

1. Nature / When to Use / Boundaries

Marquee is wish machinery specialized to one deliverable class: the acquisition landing page. It inherits wish's quality apparatus (crystallization dialogue → exemplar anchoring → cross-engine blind tournament → calibrated gauntlet-converge → reception → One-Shot + Comparative exit gates) and swaps wish's generic rubric authoring for a fixed 5-dimension LP Ceiling Rubric whose Performance dimension is anchored to machine-checkable oracles — the objective corner of the ceiling that no reviewer's taste can inflate or deflate.

Inherited engine elements (owned by reference/wish-recipe.md, not re-derived here). The preset does not re-implement them; it states only where its own phases already discharge them:

Intentional difference from wish: no Scarcity Gate. LP production is a legitimately repeatable demand — every product, campaign, and launch can warrant one — so marquee does not ration invocations. The anti-inflation anchor is objectivity instead: dim ④ cannot ACCEPT without Lighthouse Perf/SEO/Best-Practices ≥ 95, all Core Web Vitals green, and WCAG AA. This is a deliberate design choice, not drift.

Use marquee for

• The flagship acquisition LP — the page that defines how the product is first seen and must convert.
• A launch with no revision runway: the page ships once and the campaign fires against it.
• When the user asks for "the best possible landing page" and means it — ceiling quality across message, craft, conversion, performance, and trust simultaneously.

Not this — route to

Overloaded anchor: bare "landing page" with no ceiling framing → REDIRECT per reference/intent-clarification.md (normal quality → funnel[premium]/funnel; ceiling one-shot → marquee).

Scale: 26-55 agents × ≤4 cycles (3 + ≤1 exit-gate bonus), 10-24× cost. Per-phase: P0-P1 ≈ 2 · P2 ≈ 4-6 (Ground sweep) · P3 ≈ 9-12 (3 directions × structure+direction+prototype, + judge panel) · P4 ≈ 6-8 (craft-axis production) · P5 ≈ 5-7 Evaluators/cycle + 2-3 skeptics on near-ceiling cycles · P6-P7 ≈ 2. Confirm-before-launch always (unconditional — intentional parity with wish/summit).

Model selection: as wish §1, judgment-heavy phases at plan-tier / rest at Sonnet 5 default — mapped here to Phase 1 Crystallize + the Phase 6 One-Shot Gate verdict (plan-tier) vs tournament generators, craft-axis producers, and Evaluators (Sonnet 5).

2. Termination Bound

As wish §2 (reference/wish-recipe.md) for the exit-reason vocabulary and best-so-far reporting, with two preset-fixed deltas: N=3 is fixed (no L/XL N=5 escalation) — hard total 4 cycles including the One-Shot bonus cycle; and ACCEPT requires dim ④ to reach 3 only via its machine oracles (Lighthouse/CWV/WCAG) — a candidate ACCEPT with a failing Lighthouse run is structurally impossible.

3. Confirm / Safety Gate

As wish §3 (reference/wish-recipe.md) for the confirm-always parity, the contract-level/AUTORUN-cannot-skip checkpoint mechanic, and the F8 cost-gate subsumption. Marquee-specific: Phase 0 surfaces the 5-dimension rubric skeleton + preliminary cost envelope; Phase 1 Crystallize excavates the conversion contract (goal / single CTA / offer / audience / brand constraints / disappointment criteria).

• No Scarcity Gate — intentional difference from wish (§1).
• Standard Ask First tiers apply unchanged (L4 security, 10+ files, external publish/deploy — shipping the LP to a live host is an external system modification and always confirms).

4. Resume

Checkpoint-resume (≥ 4 phases): marquee resume — persist the Marquee Contract + Ceiling Rubric at Phase 1 exit; the Evidence Ledger at Phase 2 exit; the winning direction's full prototype content plus the salvage list at Phase 3 exit (resume must never re-run the tournament); the produced page + per-dim score trajectory + surviving attacks at each Phase 5 cycle boundary. An interrupted marquee resumes mid-convergence with its trajectory intact.

5. Phase Contract (AUTORUN chain template)

Phase 0  FRAMING ★Confirm-before-launch (always)
         Nexus[scope + rubric skeleton + preliminary cost envelope]
   ▼
Phase 1  CRYSTALLIZE ★contract-level dialogue (AUTORUN cannot skip; plan-tier model;
         reference/dialogue-protocol.md)
         → Marquee Contract: conversion goal + single CTA + offer + audience +
           brand constraints + disappointment criteria (each mapped to a score-0 trigger)
         → lock the Ceiling Rubric (5 fixed dims, §5a; reachability sanity-checked by Magi)
   ▼
Phase 2  GROUND ‖ Compete[competitor LP teardown] ‖ Cast+Field[personas — the Echo
         walkthrough roster] ‖ Growth[search intent + GEO] ‖ Evidence Ledger per
         reference/research-grounding.md (every claim/number/logo/testimonial cited
         and trust-tiered; ungrounded claims are barred from copy)
   ▼
Phase 3  TOURNAMENT ‖ 3 orthogonal directions: narrative-led / design-led / proof-led
         (replace with dialogue-surfaced axes when Phase 1 finds better ones)
         each direction: Funnel[structure] + Vision[direction] + Forge[hero + 1 section
         prototype] — prototype-level competition, never 3 full builds
         → judge panel = Echo×personas ‖ Palette ‖ Growth[CRO] ‖ Magi[brand-fit],
           scoring vs the frozen Ceiling Rubric
         → winner + salvage list (informal cherry-pick from runners-up, fed to Phase 4)
   ▼
Phase 4  PRODUCE — funnel[premium] craft axes as the production substrate:
         Prose[copy — Evidence-Ledger-grounded] ‖ Artisan[build] ‖ Flow[motion] ‖
         Ink/Builder[image assets] ‖ Muse[token-first] ‖ Growth[SEO/OGP/JSON-LD]
         (parallel branches per _common/PARALLEL.md; Artisan owns page integration)
   ▼
Phase 5  GAUNTLET-CONVERGE loop ≤ 3 cycles (evaluator-loop-protocol.md machinery;
         Generator excluded from Evaluators)
         every cycle   ‖: Echo[persona re-walk → ①] ‖ Palette[a11y → ④] ‖
                          Bolt+Lighthouse[perf oracles → ④] ‖ Growth[CRO/SEO audit → ③] ‖
                          Voyager[real browser + responsive → ②④] ‖ Judge[craft → ②] ‖
                          Canon[legal]?[claims/legal → ⑤]
         staged        ‖: refutation panel per _common/ADVERSARIAL_REFUTATION.md —
                          spawned ONLY once all dims ≥ 2; LP-specific skeptic angles:
                          (a) bounce-reason enumeration — each persona names why they
                              would leave without converting;
                          (b) differentiation refutation — "does this page still hold
                              with a competitor's logo swapped in?" (a YES kills ①);
                          panel also ratifies candidate ACCEPTs (§2)
   ▼
Phase 5.5 RECEPTION — the GROUND personas (Cast/Field) meet the produced page COLD:
         no rubric, no loop history, first-impression only → friction points enter as
         evidence on dims ① message and ⑤ trust; unresolved failures become residuals
   ▼
Phase 6  EXIT GATES (both run; together they grant at most ONE bonus cycle, total 4) —
         MATERIALLY-BETTER test and loss semantics as wish §6 Phase 4 (`reference/wish-recipe.md`).
         (a) ONE-SHOT GATE — fresh-context independent verifier (plan-tier), spawned with
             ONLY the Marquee Contract + Ceiling Rubric + final page — no loop history.
         (b) COMPARATIVE GATE — blind head-to-head, provenance stripped: the produced page
             vs the strongest Phase 2 exemplar LP, and vs the retained runner-up direction.
         → either fires → exactly one bonus cycle → re-gate once (advisory only)
         → both pass → deliver
   ▼
Phase 7  SHIP — Guardian [+ Launch? if deploying] + Marquee Dossier;
         Experiment[A/B test design] recorded as a handoff to growth-acceptance

5a. The Ceiling Rubric (5 fixed dimensions, score-3 anchors)

Disappointment criteria from Phase 1 map onto these dims as score-0 triggers (wish mechanism, inherited verbatim). The dimension set is fixed — Phase 1 tunes the anchors to the product, never removes a dimension.

6. Output Report — Marquee Dossier (named)

Emitted inside NEXUS_COMPLETE on top of the base ## Nexus Execution Report:

• Marquee Contract recap — conversion goal, CTA, offer, disappointment criteria, each classified per Acceptance Provenance (Q15)
• Per-dim score trajectory — per-cycle scores across the 5 dimensions; machine-oracle raw values (Lighthouse / CWV / WCAG) attached for dim ④
• Tournament comparison — the 3 directions' judge scores + why the winner won + salvage items applied
• Evidence Ledger — every copy claim with citation, trust tier, and verification status
• Gauntlet ledger — bounce-reasons raised / killed / fixed; differentiation-refutation verdict; open attacks listed on any plateau exit
• Exit gates — One-Shot redo assessment with cited evidence + Comparative verdict (vs exemplar / vs runner-up, with the property named on any loss) + bonus-cycle usage
• Calibration record — per evaluator: exemplar score, control score, verdict (dim ④ marked calibration-exempt)
• Reception simulation — per-persona cold verdict + friction points + unresolved residuals
• Exit reason (§2 vocabulary) + residual gap vs ceiling (empty only on ACCEPT)
• A/B handoff — the Experiment test design queued for growth-acceptance

7. Failure Modes Prevented

Inherited from wish §9 (reference/wish-recipe.md) unchanged: vibe acceptance of "best possible", first-idea anchoring, tournament cost blowup on small deliverables, generator grading its own work, unbounded ceiling pursuit.

8. Shared-Protocol References

(Middle column — what each protocol contributes — is as wish §10; see reference/wish-recipe.md.)

9. Decision Tree vs Neighbors

Is the deliverable an acquisition landing page?
  NO  → wish (any-deliverable ceiling) / runway (in-product UI) / podium (docs+slides)
  YES ↓
Does an LP already exist and just needs improvement?
  YES → kaizen (metric-driven) / restyle (visual direction)
  NO ↓
Is ceiling one-shot quality actually required (launch fires once, page defines the product)?
  NO  → funnel[premium] skill (standard-to-premium studio) / funnel skill (single page or section)
  YES → marquee — crystallize → ground → tournament → produce → gauntlet-converge
        → reception (cold personas) → One-Shot + Comparative gates → ship (+ A/B handoff to growth-acceptance)