Audit Reference Originality

Audit Reference Originality

Compare the shipped experience with the complete reference corpus. Treat the audit as an evidence exercise, not a vibe check.

Preserve the audit boundary

• Audit only unless the user also asks for fixes.
• Call findings originality risks, overlaps, or red flags; do not declare legal plagiarism from visual similarity alone.
• Pair every red flag with exact current-site evidence and exact reference evidence.
• Keep fact, inference, and unknown access separate.
• Do not clear a site after checking only its homepage screenshot.

1. Build the source registry

Start with the materials explicitly supplied in the task or recorded by the project:

• manifests, prompt packs, and originality matrices
• representative stills, full-page captures, and section crops
• MP4s and extracted motion frames
• reference URLs and named creators
• source brand assets, copy, numbers, screenshots, and downloads
• project briefs, IMAGE_CREDITS.md, licenses, and attribution files

Prefer local evidence captured at the time of the brief. Use a live reference only to fill a real gap, because it may have changed. Record each reference's path or URL, role, date when known, and which categories it can prove.

Stop and report an access gap when a promised reference is missing. Do not silently reduce a full-page or motion audit to one cover image.

2. Inventory the current site and its history

Inspect:

• rendered text, metadata, navigation, calls to action, legal copy, and hidden accessible labels
• brand names, wordmarks, logos, icons, people, companies, URLs, and product/interface data
• prices, metrics, dates, counts, percentages, package names, and repeated proof claims
• every rendered image, background, texture, screenshot, font, logo, icon, audio file, and downloadable asset
• every video, poster, frame sequence, shot order, transition, duration, and playback treatment
• layout hierarchy, section order, distinctive compositions, typography behavior, motion grammar, shaders, cursor effects, and interaction sequences
• current source, built output when available, asset provenance, and repository history

Do not trust filenames as proof of originality. Inspect the bytes, visible result, and history. Search renamed, deleted, and replaced files with git log, git show, git log -S, and git log --all --name-status.

Run the deterministic inventory helper when local files are available:

python <skill-dir>/scripts/build_evidence_inventory.py \
  --site <site-root> \
  --reference <reference-file-or-directory> \
  --reference <another-reference> \
  --output <temporary-output.json>

The helper finds current and historical exact-file matches, suspicious basename reuse, normalized text overlap, and repeated number tokens. Treat its output as leads for human review, not an automatic verdict.

3. Compare category by category

Read references/audit-rubric.md before judging findings.

Audit at least these categories:

1. Text — headlines, body copy, labels, CTAs, captions, legal text, alt text, metadata, and decorative wording.
2. Brands — names, marks, wordmarks, proprietary icons, people, companies, partnerships, URLs, and distinctive verbal identity.
3. Numbers — metrics, percentages, prices, dates, counts, plan structures, durations, and interface values.
4. Images — exact files, crops, generated derivatives, screenshots, people, poses, objects, signature compositions, and color treatment.
5. Assets — fonts, icons, logos, textures, mockups, downloads, code bundles, and third-party media with unclear provenance.
6. Videos — exact files, frames, shots, timing, camera moves, edit rhythm, transitions, overlays, posters, and audio.
7. Structure and motion — section order, unusual layout devices, pinned sequences, cursor interactions, shaders, and combinations of signature elements.
8. History — copied material that was later renamed, recolored, cropped, hidden, deleted, or replaced.

Common patterns such as black backgrounds, large sans-serif type, ordinary pricing tables, standard fade-ins, or a conventional footer are not red flags by themselves. Escalate combinations of distinctive elements or direct evidence.