Reference originality review

Reference Originality Review

Determine whether a declared subject is independently expressed relative to its declared references. Produce one Reference Originality Review grounded in exact source pairs, not a visual impression or unsupported legal conclusion.

Read references/reference-originality-method.md before collecting or classifying similarity evidence.

Method

1. Freeze the subject revision, reference set, authorized access, intended audience, requested comparison boundary, and excluded legal questions.
2. Build the source registry before reaching a verdict. Include current artifacts and material history on both sides when history could distinguish independent evolution from late convergence.
3. Inspect applicable expression across wording and numeric claims, branding, imagery and media, assets, page or flow structure, interaction, motion, responsive behavior, and implementation-visible composition.
4. Record exact paired evidence with stable locators. Separate common product grammar, functional constraints, and standard platform conventions from distinctive expression and distinctive combinations.
5. Test credible alternative explanations: shared requirements, a common upstream library or template, chronology, convergent usability choices, authorized reuse, and incomplete evidence.
6. Classify the declared comparison only as clear_on_reviewed_evidence, changes_recommended, hold_for_redesign, or evidence_incomplete. State confidence and denominator; no verdict proves universal originality.
7. Where change is needed, identify the smallest coherent redesign that removes the distinctive dependency while preserving the product's own requirements. Re-review the exact changed artifact against the same evidence set.

Evidence rules

• Never infer copying from a mood, genre, color family, or common layout alone.
• Never clear a subject after inspecting only a screenshot when wording, motion, state transitions, assets, or history are material.
• Treat image hashes, text similarity, DOM diffs, and motion traces as leads; a metric is not the verdict.
• Preserve uncertainty and contradictory evidence. Missing source access is an evidence gap, not permission to claim either copying or originality.
• Keep private references and internal analysis within their authorized audience. A public derivative must contain only intentionally releasable evidence.
• Describe design risk and observed similarity. Legal infringement, fair-use, trademark, and jurisdiction-specific conclusions require qualified legal authority.

Output contract

Produce one Reference Originality Review containing:

• exact subject identity, revision, scope, authorized audience, reference denominator, exclusions, and evidence gaps;
• source registry with current and historical locators, chronology, provenance/authorization, and confidence;
• pairwise evidence ledger covering each applicable expression category;
• alternative-explanation analysis distinguishing conventions, constraints, shared sources, and distinctive expression;
• calibrated verdict, rationale, confidence, and the boundary within which the verdict is valid;
• ranked redesign actions with the distinctive dependency each action removes; and
• exact re-review evidence for a changed candidate, or a clearly owned next action when evidence remains incomplete.

Routing boundaries

• design-system-extractor owns evidence-backed recovery of the subject product's own existing tokens, components, states, and design rules.
• interface-craft owns designing, building, polishing, or reviewing a bounded interface or flow; this Skill supplies an originality finding when that is an independently requested artifact.
• product-asset-production owns rights-safe production, provenance, and exact QA of a declared media pack.
• Comparative product research may inform the reference set, but does not authorize copying another product's identity or private material.
• Route legal conclusions to qualified legal review; do not manufacture a legal artifact because no specialist is available.

Complete only when another reviewer can locate every material source pair, understand why alternatives were accepted or rejected, and reproduce the declared verdict without relying on the author's intuition.