You can now read UGC and script one honest video. The next problem is volume. After five shoots you have forty clips named final-v2-REAL.mp4, three conflicting ingredient lists, a testimonial somewhere in email, and nobody remembers which hook actually worked or whether you may still use Maya's face.

That pile is not a library. A library lets the next creator make the next variation without inventing facts, re-asking permissions, or re-editing from zero.

What belongs in the library

Think in reusable pieces, not finished videos. For the shop moisturizer or a ByeBuy lesson campaign, collect:

• Approved product clips: the 3-second pump texture, the half-face application, the screen recording — trimmed, labeled, export-ready.
• Product facts: ingredients, size, price, how to use, what it does not do. One canonical list with a version date.
• Proof points: demo footage, before/after pairs with lighting notes, test reports, source links.
• Brand rules: vocabulary to use and avoid, visual system, disclosure wording, claims you never make.
• Real customer permissions: who said what, where the original lives, what usage they approved and until when.
• Creator notes: who films well in which setting, pacing quirks, pronunciation, what each creator will not say.
• Hooks that worked: the exact opening shot + line, with the signal it earned (completion, saves, comments).
• Captions and on-screen text: approved phrasing, spelling of names, disclosure placement.
• Music/voice rights notes: track name, license type, where it may run, expiry.
• Performance learnings: one line per variant — what you tested, what happened, what to do next.

A ten-second texture close-up reused across six videos is a library asset. A vague folder called "content stuff" is soup.

Markdown and JSON discipline: stop reconstructing truth in chat

Connect this directly to Part II. Creative work gets faster when facts live in files, not in chat memory.

• Use Markdown for human-readable direction: briefs, scripts, review notes, learning logs. A future creator or AI session can open UGC-SCRIPT.md and understand the story.
• Use JSON for machine-readable structure: asset indexes, claim banks, publishing metadata. Consistent fields mean you can filter, sort, and check expiry without re-reading every file.

Example: Maya's testimonial lives once as a video file, once as a Markdown quote with context, and once as a JSON row with status, approval, and rights-expiry. Three views, one truth. Nobody retypes her words from memory and accidentally upgrades "feels gentle" to "cured my skin."

Keep one rule: facts are written once, referenced everywhere. If the price changes, you update the fact file and the claim bank — not twelve captions by hand.

A library vs. a pile: metadata is the difference

A pile has files. A library has files plus answers. Every asset should answer seven questions at a glance:

If an asset cannot answer status, source, approval, and rights, treat it as unusable. That single habit prevents the two classic disasters: publishing an unapproved claim and reusing an expired face.

Name files so humans can scan: 2026-09-02_maya_texture-closeup_v01_approved.mp4. Keep a clean master, then cut channel crops from it — the same discipline Class 58 taught for images and Class 62 taught for delivery packs.

Practice: index ten assets + build a claim bank

Make UGC-ASSET-INDEX.md. You need ten rows minimum. Mix types so the next shoot is actually covered: at least three clips, two facts/proofs, one permission record, one hook, one caption, one rights note, one learning. Use this shape:

## Asset index

| ID | Asset | Type | Status | Source | Approval | Audience | Campaign | Rights / expiry |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| A-01 | Texture close-up 3s | clip | approved | Filmed 2026-09-02, Maya | Owner+Maya 09-04 | Sensitive-skin beginners | Winter launch | Organic+paid, until 2027-03-01 |
| A-02 | Day1 vs Day14 split | proof | approved with note | Same light, no filter | Owner 09-04 | Same | Same | Organic only; no medical claim |
| ... | ... | ... | ... | ... | ... | ... | ... | ... |

Then add the claim bank below the table — the only sentences anyone may say about the product:

## Claim bank

- "Fragrance-free moisturizer, 50ml." [approved — product sheet v2, 2026-08-20]
- "Pea-size covers full face in demo." [demonstrable — see A-01]
- "Maya: 'No sting around my nose after two weeks.'" [approved quote — permission P-01, until 2027-03-01]
- "Clears acne in three days." [BANNED — unverified, never use]

Include at least one banned claim. Libraries teach what not to say as clearly as what to say.

For JSON discipline, keep a matching machine-readable copy for at least the index — even a small ugc-assets.json with id, asset, status, source, approval, rights_expiry fields counts. The point is not tooling sophistication; it is that status, source, and expiry are fields, not folklore.

Finish line

Deliver UGC-ASSET-INDEX.md with ten indexed assets (each with an approved/not-approved status field) plus a claim bank with approved, demonstrable, sourced, and banned entries.

Quick verification: pick a banned claim and search your index — is there truly no clip or caption still carrying it? Then pick an approved asset and check its rights-expiry — could you legally post it next month? If either answer is unclear, the metadata is incomplete.

Common failure mode: indexing filenames without decisions. An index that lists ten clips but never says approved, expired, or banned is just a prettier pile. Status and rights are the whole job.

Check your understanding

1. What is the difference between a library and a pile of assets? 2. Why keep facts in Markdown/JSON files instead of reconstructing them in each chat session? 3. Which four metadata fields make an asset safely usable, and what goes wrong when rights-expiry is missing?

Next

You now have the full Class 63 loop: read UGC honestly, brief a truthful 30-second script, and store the proof so the next video starts from approved truth. Class 64 builds on exactly this foundation — using AI to research angles, vary hooks systematically, and test creator content — without letting automation invent the customer, the quote, or the result.