reference-intelligence

name

reference-intelligence

description

Analyze and translate design references, award-site videos, screen recordings, screenshots, transcripts, AI observation reports, precedent collections, and inspiration links into brand-specific decisions. Use for requests such as "analyze these references", "what should we borrow", "compare these award websites", "turn these video reports into a plan", or "what should we keep, adapt, reject, or defer". Do not use for competitor pricing/market dossiers, approved motion-library lookup, brand diagnosis without references, or implementation from an already approved plan.

Reference Intelligence

WHEN TO USE THIS

• Load after brand diagnosis and a creative brief when references may expand concept vocabulary.
• Load when one or more external references must be observed, compared, translated, or converted into implementation guidance.
• Load when attractive mechanics risk being copied without evidence, brand fit, responsive logic, or production judgment.

NEVER DO

• Never let references define positioning before brand evidence does.
• Never describe an inference as observed behavior.
• Never identify a library, framework, plugin, or exact implementation from appearance alone.
• Never select a mechanic because it looks prestigious, novel, or award-winning.
• Never copy another site's geometry, sequence, section architecture, or identity system literally.
• Never promote a pattern without asset, mobile, reduced-motion, accessibility, performance, and maintenance checks.
• Never treat an AI report as direct visual evidence; record it as a mediated source.
• Never generate implementation plans before the brand translation is approved.

OPERATING DEPTH

Choose the smallest depth that answers the research questions.

Depth	Use When	Required Result
Lightweight gate	A full spatial inception has no supplied corpus or needs no precedent study	Named reference questions, rationale, and corpus not required decision
Focused analysis	One or two references answer bounded questions	Source analysis plus concise translation ledger
Corpus analysis	Three or more references, mixed media, or explicit comparison	Per-source analyses, cross-corpus synthesis, evidence ranking, translation ledger, and approved handoff

Escalate depth when sources conflict, recordings are incomplete, mechanics are trigger-ambiguous, or the decision creates substantial asset or engineering cost.

INPUT CONTRACT

Gather:

• approved evidence dossier, diagnosis, creative brief, or equivalent brand constraints;
• named research questions and the decision they will inform;
• source inventory, provenance, format, duration or page scope, and access limitations;
• current experience or implementation when adaptation must fit existing work;
• real asset, accessibility, responsive, performance, maintenance, and authority boundaries.

If brand context is missing, perform source forensics only. Stop before recommending brand adoption.

PROCESS

1. Scope the questions before inspecting mechanics.
2. Register every source and distinguish direct media, screenshots, transcripts, reports, and second-hand summaries.
3. Reconstruct each source chronologically and label observation, inference, ambiguity, and confidence.
4. Normalize communication jobs, layout, typography, imagery, interaction, transition, motion, responsive clues, and likely cost.
5. Compare recurring patterns, unique mechanics, strongest evidence, contradictions, and source-quality limits.
6. Translate candidates through brand truth, chapter job, proof burden, asset reality, stillness, accessibility, mobile, performance, maintenance, and scale.
7. Classify every material candidate as KEEP, ADAPT, REJECT, or DEFER.
8. Approve the translation before it affects territories, storyboards, production plans, or code.
9. Handoff only the decisions and evidence required by the next workflow.

TRANSLATION CONTRACT

For every accepted or deferred mechanic, record:

• communication job;
• strategic source and evidence confidence;
• retained principle;
• changed geometry, intensity, trigger, duration, content, and section ownership;
• required proof and assets;
• mobile equivalent;
• reduced-motion/static equivalent;
• accessibility and performance constraints;
• implementation and maintenance cost;
• failure and rejection conditions.

Use a removability test: if the recommendation becomes indefensible when the reference's prestige or visual polish is removed, reject or translate it further.

ROUTING BOUNDARIES

Request	Route
Competitor positioning, pricing, SEO, or market dossier	competitor-profiling
Technical tool/framework comparison	research-analysis
Brand truth, perception gap, audience, or premium posture	brand-strategy
Approved motion job needs an existing effect candidate	motion-library
Approved mechanic needs motion planning or code	cinematic-motion
Post-project lessons may change capabilities	learn

ANTI-PATTERNS

Anti-Pattern	What It Is	Fix
Inspiration Shopping	Collecting what looks attractive without research questions	Define the decision and questions first
Prestige Transfer	Assuming an award-site mechanic will make another brand premium	Test brand truth, proof, and removability
Report As Vision	Treating an AI report as if the agent watched the source	Mark mediation and preserve uncertainty
Mechanic Collage	Combining favorite effects without one governing system	Set motion ownership and a pattern budget
Scale Imitation	Importing the density or complexity of a larger brand	Translate to actual content, capacity, assets, and maintenance
Implementation Guessing	Naming GSAP, WebGL, or plugins from appearance	Describe likely pattern as inference only
Mobile Afterthought	Approving desktop spectacle before an equivalent exists	Design static/mobile meaning before enhancement

REFERENCE LOADING RULES

Load references/source-forensics.md whenever analyzing recordings, screenshots, transcripts, AI reports, or incomplete/mediated evidence.

Load references/corpus-synthesis.md when comparing three or more sources, ranking evidence, resolving contradictions, or checking selection and prestige bias.

Load references/brand-translation-and-handoff.md when brand context exists and the task requires Keep/Adapt/Reject/Defer decisions, concept inputs, storyboards, production guidance, or implementation plans.

OUTPUT SHAPE

Lightweight gate: Research decision -> named questions -> corpus requirement -> rationale -> next workflow.

Focused analysis: Source register -> chronology -> observed/inferred inventory -> candidate patterns -> translation ledger -> ambiguities.

Corpus analysis: Analysis contract -> per-source findings -> recurring/unique mechanics -> evidence ranking -> bias audit -> brand translation ledger -> approved handoff.

Implementation handoff: Master sequence -> shared rules -> section plans -> assets and evidence gates -> motion ownership -> fallbacks -> acceptance criteria.

NON-NEGOTIABLE CHECKLIST

1. Research questions precede reference selection.
2. Source type and provenance are explicit.
3. Observation, inference, ambiguity, and confidence remain separate.
4. Corpus claims trace to source evidence.
5. Every material mechanic is kept, adapted, rejected, or deferred.
6. Adaptations state exactly what changes and why.
7. Brand truth and proof remain upstream of reference prestige.
8. Mobile, reduced motion, accessibility, assets, performance, and maintenance are covered.
9. Translation approval precedes implementation planning.
10. The handoff contains no unsupported claim or copied architecture.