# 2026-05-29 Next Batch 11 Patterns

Scope: 11 TikHub/Douyin selected samples with downloaded source videos and DashScope qwen3 ASR.

Inputs:

- Research DB: `/home/deano94/codex/material_research_db/material_research.sqlite`
- ASR root: `/home/deano94/codex/material_research_db/asr`
- Videos root: `/home/deano94/codex/material_research_db/videos`

Progressive disclosure:

- `references_read`: `/home/deano94/.codex/skills/mayagi-live-procut/SKILL.md`
- `references_read`: `/home/deano94/.codex/skills/mayagi-live-procut/references/material-strategy-research.md`

## Batch Status

- Total candidates in DB: 255
- Selected candidates: 24
- Downloaded videos: 20
- ASR complete: 20
- Analyses after this batch: 20
- Four selected samples had no usable source video URL in the TikHub response and were left untranscribed.

## What This Batch Adds

### 1. 素材不是越多越好，是变量要真

Operator samples repeatedly say the same thing from different angles:

- repeated remakes of an old winning script get identified as the same crowd/material pool;
- hundreds of low-quality stitched clips can dilute the account and steal exposure from better materials;
- new material should introduce real variables: hook, script, actor, scene, prop, proof order, or product angle;
- old materials can suppress new materials if they keep the weight and the new clips are not given a clean test path.

MAYAGI transfer:

- Do not render 20 cosmetic variants for L7825.
- Build 5-10 distinct tests by viewer mind: toe pain, foot line, summer scene, wide-foot trust, brand taste.
- Record what the changed variable is before uploading.

### 2. “素材本身行不行” comes before amplification

Several operator samples separate content quality from投放操作:

- tools such as追投/抖加 are for supplementing signal, not saving weak material;
- potential material is recognized through relative signals: display, click, conversion, entry cost, product click/live entry, and trend against account baseline;
- a clip can have high ROI but poor scale, or high spend but weak conversion; the action should differ.

MAYAGI transfer:

- Add a scoring sheet after each test: hook retention, first-proof clarity, product-page/shop/live-entry signal, and whether the buyer comments match the intended audience.
- Do not keep polishing a topic that has no early signal; change the viewer problem or proof order.

### 3. 素材库要按“句子能证明什么”管理

The editing/tutorial sample shows a useful workflow for winners:

- first version should verify the structure, not become visually overbuilt;
- after a product/video proves it can sell, preserve the winning opening or sentence as a reusable preset;
- enrich the same speech with product images, details, masks, split screen, and order changes;
- do not use enrichment before the core structure works.

MAYAGI transfer:

- L7825 needs a sentence-to-visual-job asset library: toe protector, T-strap line, heel memory, outfit scene, color decision.
- Reusable modules are allowed only after a winning structure appears.
- No mirror/de-dup trick may distort shoes.

### 4. 鞋类最好从脚型/身材焦虑切，不要只说好看

The two shoe samples are close to the MAYAGI problem:

- creator self-identifies as wide foot, fat foot, pear shape, thick calf, no ankle, sweat foot, etc.;
- the hook often starts from anxiety or contradiction, then uses on-foot proof;
- proof is concrete: width, toe space, strap coverage, sole softness, anti-slip, stability, whether the foot slides, whether toes are protected;
- honest caveats increase trust when framed correctly.

MAYAGI transfer for L7825:

- Strong hooks:
  - “夹趾鞋最怕的不是丑，是夹脚。”
  - “脚宽脚背高，别先看跟高，先看带子怎么走。”
  - “想显腿长但不想穿累脚高跟，可以先看这双的脚背线。”
- Must show real proof:
  - transparent soft toe protector;
  - toe-post contact position;
  - T-strap vertical line;
  - low heel/stability;
  - on-foot walking or weight shift.
- Avoid crude body-shaming language; keep empathy and MAYAGI tone.

### 5. 穿搭/奢品内容提醒我们：场景和品味可以卖，但不能替代实证

The outfit and luxury-accessory samples work by taste mapping:

- object is explained through wearer identity, outfit scene, occasion, and color relationship;
- the strongest value is not a spec, but “this choice says something about the wearer”;
- fashion content builds desire by showing a complete ecosystem, not an isolated product.

MAYAGI transfer:

- Use generated model/outfit images for “where would she wear it” and “what taste does it signal”.
- Use real A/B/C proof for comfort, material, toe, strap, and heel truth.
- Scene words should be concrete: summer restaurant, commute trousers, city vacation, black dress, white skirt, cropped pants.

## Promotable Rules

These are repeated enough to enter the skill reference:

1. Do not count cosmetic remakes as new material. A new material needs a changed hook/script/proof order/scene/actor/product angle.
2. Before scaling a clip, identify whether the material itself has signal. Tools amplify signal; they do not create product belief from nothing.
3. For MAYAGI shoe clips, start more often from foot/body anxiety or scene desire, then prove with visible shoe-foot contact.
4. Preserve winning openings/sentences as reusable modules only after they have shown signal; before that, keep edits structurally clean.
5. Generated outfit images are desire projection. They must not replace comfort/fit proof.

## Not Promoted

These stay only in the DB/report:

- exact operator追投 budget numbers;
- aggressive creator slang;
- luxury celebrity examples and brand names;
- platform claims such as “8x revenue” without MAYAGI account validation.

## Immediate L7825 Implications

The next L7825 render batch should not start from the three previous v7 drafts. Build from this sequence:

1. Pick one viewer state.
2. Choose a complete sentence path from A-bank, not one A group.
3. Assign one visual job per sentence.
4. Use B/C only where it proves that sentence.
5. Add generated outfit stills only for scene/persona projection.
6. End with MAYAGI / Summer Season as a quiet brand tail, not a hard low-end sale.