Phase 1 before locking the 3rd item, theory, read zmm/references/理论框架选型与装配.md §1–4, first compare genuinely different explanations, then score, verify, and confirm per the existing「lock four items」. §5 mapping is merged into the one-pager, no separate approval; the concept must still provide transferable judgments and cannot substitute a philosophical label for a definition.
First read zmm/references/家族公约.md(if unreadable → state plainly「cannot read the convention, red lines cannot be guaranteed」and stop,do not substitute memory for the convention），
This skill's built-in criteria are in references/规则卡.md(criteria / why / how to check / strength), read them once before starting;{vault} when there is a corresponding rules file in the vault, the vault prevails and the rule card is the baseline.
Then read zmm/references/交互规范.md（🔴 reading once doesn't count: at wrap-up follow §4 the three-piece set — Recap · Before/After · next step: give numbered options; when information is missing follow §4 usemultiple-choice questionsto ask,ask only one at a time; for cases where it doesn't apply see §5), then read {vault}/00-规则与索引/文案框架-总入口.md(the only navigation).
🔴 Its position on the shortest path:replaces /zmm-script, not added alongside it
选题 → 【重讲概念型走这儿】→ 红线审核 → 发
/zmm-concept 代替 /zmm-script
⚠️ zmm routing remembers a measured fact: this system once had 7 pre-publish gates, and the result was 13 finished drafts, 0 published.
The number of gates itself is the cause of not publishing.
🔴 This skill adds no new gates. It does only one thing:make the rework that would already happen happen earlier, on one page.
Why this skill exists (measured cost, do not delete)
N22 "The Lower the Price, the Worse It Sells" went through 12 versions. Among the 12 versions only two really mattered: ① Zhan Mingming said「my experience can't be the main body, Douyin's mass audience can't reach it」 ② he spoke himself。
The other nine versions were guessing what he wanted, and every wrong guess had the same cause: the whole draft was written first, only to find the proposition level was wrong.
| Version | Died at |
|---|---|
| v1 · v2 · v4 | Ontology definition not locked, propped up with「his experience」 |
| v3 | Theory not locked(Weber explained「it dropped but didn't feel like it」, not「it dropped and nobody bought」) |
| v5 · v6 | Proposition not locked(shifted from「telling one point」to「telling the whole thing」) |
⇒ Six versions of rework, all dying on the same four items. And each of those four items needs only one sentence.
🟢 N23《没有精准流量》validated that it works — but pushed the problem downstream (2026-08-29)
N23's retelling v1 was basically usable, iterated only to v4, with no rework at the proposition level. Locking four items holds.
🔴 But everything this entry lost was lost in Phase 3–5:
| What was lost | Died at which step | Cost |
|---|---|---|
| The line spoken on behalf of the audience(The only source of the urge to share) | Phase 4 — I wrote it into "suggest adding it back," he approved it,I only implemented the other one; this line never made it into the main text | It's not in the final cut. This item's share criterioncan't be read out |
| The landing point changed but wasn't treated as a rewrite | After Phase 3 — capability knife → classification knife, and the landing-point sentence is "the draft's ID card" | frontmatter / title / filename were still carrying the old landing point for a while |
| Rejected the right opening based on a wrong speech-rate parameter | Phase 2 — used the ÷5.1 point value to calculate the first 5-second budget, judged "and high-ticket" wouldn't fit | He added it back in the final cut,it fits. Measured 6.88, parameter off by 35% |
⇒ Once the upstream is locked, rework doesn't disappear—it moves downstream.
The four places in this update (Phase 2 speech rate · Phase 3 post-check landing point · Phase 4 landing in main text · Phase 5 cover and saving to disk) All come from N23, all are downstream.
Who it's for
A solo knowledge creator who researches and edits everything themselves. Three hard constraints: don't assume he has a team · no upward reporting (give judgments directly) · zero jargon(use theory as-is, but the terms don't appear in the final draft).
🔴 The cross-cutting layer is always there：CLAUDE.md Floor rule 7 — After finishing each paragraph, reread it as "I" and write out the inner line he'd say at that moment. If you can't write it, the paragraph isn't done. No tone of blame or preaching anywhere in the piece.
Process
Phase 0 接题 + 两道准入闸门
Phase 1 🔴 锁四样（一页纸，不写整稿） ← 本技能的全部价值
Phase 2 写整稿（一次写完，不逐段确认）
Phase 3 🔴 交给他口述（D1）
Phase 4 只剪不扩 + 红线全表 + 差异表 + 规则入库
Phase 5 /zmm-title 归口 + 封面 + 落盘 + 建取数日历
Phase 0 · Take the topic + admission (no pointless questions)
The topic comes from {config.paths.topic_pipeline} 's directory's topic pool, or newly given by the user.
Two gates, no work without both(criteria are in pool §2.3, go read them, don't rely on memory):
| # | Gate | One sentence |
|---|---|---|
| ① | Anchor point | Before he hears your new formulation,is there a word he's already familiar with that you took apart first？ |
| ② | Ruler criterion | Can he take the thing you handed overand use it to measure something you never mentioned? Can't measure = it's still just a definition |
Also run it through the dedup gate（总入口 §2 Layer 1): compare core conclusion × main analogy + write out which is the sister item。
Phase 1 · 🔴 Lock four things (the core of this skill, see references/锁四样.md）
Ask everything at once, one page, don't write the full draft.
| # | What to lock | Who provides |
|---|---|---|
| 1 | Specific question | AI proposes candidates,he decides |
| 2 | 🔴 Ontological definition / the counterintuitive line | Only he can provide |
| 3 | Classic theory + disciplinary common-knowledge target | AI finds and scores on five dimensions,he confirms |
| 4 | 2–3 plain-language analogies | AI proposes,he swaps |
🔴 If all four aren't locked, you may not write the full draft.
For the theories in item 3, each one gets a card; any card missing the "core to which level" line may not enter the final draft:
| Line | What to write |
|---|---|
| What problem does it solve | One sentence |
| Mechanism | One sentence of cause and effect |
| Source | Name · Work · Year · Chapter or page (go as deep as you can) |
| How far it was verified | 读过原文 / 权威二手（教材、综述，写明是哪本） / 没核到 |
| Verbatim or paraphrase | The sentence quoted in the final draft: is it the verbatim translation or a condensed paraphrase; for a paraphrase, say “the gist is” in the draft |
Rules: If the same source cannot be verified twice in a row, write 没核到 and stop; do not expand the search, and do not name it in the final draft; for quotations seen through secondary literature, state “quoted in” and do not pretend to have read the original; when a person's name first appears, write “Country · Role · Chinese name (common foreign name)”.Naming a book in the final draft means handing the audience a verifiable source, and the audience really will check it. “I remember he said” and “I have read the original” must look different on this card.
⚠️ Item 2 is the easiest to skip, and it is exactly the root of the six rounds of rework. AI cannot write a sentence like “pricing itself is part of the product” — it can only write a feature description (“pricing answers three questions”). If you cannot get item 2, stop here and wait; do not substitute theory or personal experience.
Phase 2 · Write the full draft (all at once)
Follow {vault}/04-模板/概念纠偏型口播稿模板.md six sections.
Within section ⑤, follow zmm-script/references/概念型口播.md seven beats —— 🔴 must be opened and read; two beats are the easiest to miss：
② Say the viewer's answer for them(“the easiest beat to miss, decides success or failure”) · ⑥ A second scenario from another industry(“all the sense of gain comes from this beat”).
When the draft is out, report three things:duration range · gap distribution table(count seconds) · full red-line scan result。
🔴 Give only a range for duration, not a point value（2026-08-29 总入口 §3 #2 recut):
原速区间 = 稿面汉字 ÷ 6.9 ~ ÷ 5.1
🔴🔴 2026-08-31 correction: the basis for the passage above was wrong; do not follow the old version. Testing confirmed the final video used 1.2× speed ⇒ 6.88 is the final-video rate, not the spoken rate。 Original speed = 1583 ÷ (230×1.2) = 5.74 characters/second.
⇒ The reason written on 08-29 when overturning the old ruling was “N19 5.1 vs N23 6.88,35% spread」—— after removing the speed multiplier it is 5.1 vs 5.74, a 12.5% spread. That reason no longer holds. 🔴 The original-speed speaking rates of the three conceptual pieces are actually quite concentrated：N19 5.1 · N20 5.32 · N23 5.74. ⇒ Closed:calculate original-speed speaking rate as the 5.3–5.8 characters/second range(multiple measurements); for the first 5 seconds, use the lower bound of the range for capacity and the upper bound for duration; the final video is the accelerated figure, so divide out the speed multiplier before entering it into the parameter table.
🔴 Draft actions (write all four frontmatter fields together; missing one does not count as backfilled)：
稿面字数 / 成片秒数 / 倍速系数 / 原速语速 = 字数 ÷ (成片秒数 × 倍速)
⚠️ Do not swap the two rates: to calculate script length useoriginal speed; to calculate how many characters fit in the first 2/3/5 seconds usefinal-video speed
(Douyin's 2s 跳出 / 5s 完播 / 观看趋势 are all on the final-video timeline).
🔴🔴 Before rejecting a sentence based on duration, ask one question: would a 35% error flip this conclusion? N23 measured cost: I calculated at ÷5.1 and only 21 characters fit in the first 5 seconds, so on that basis I rejected the 「and high average order value」 he wanted to add. He added it back into the final cut, and it fit. —— When a parameter is uncertain, don't use it as a veto.
🔴 When handing it over, you must say explicitly: this version is a draft for you to narrate, not the final script.
Phase 3 · 🔴 Hand it to him to narrate
No filming without a retelling. Retelling comes back → archive it verbatim → then process it.
🔴 First thing after the retelling comes back: check whether the landing point changed (added in N23)
Check the landing point, not the wording. 口播稿输出格式：The landing-point sentence is this script's ID card — changing the landing-point sentence means changing the script, not revising it.
Ask one question: the landing point he stated, and the one you locked,are they the same sentence?
N23 actual test: no, and his is better.
| AI script | His retelling | |
|---|---|---|
| Landing point | When you don't run paid traffic, you don't have the action of 「aiming」 | You can have precision marketing, but you shouldn't have precision traffic |
| What kind of knife | Capability knife(you can't do it) | ★ Classification knife(that thing is called something else) |
⇒ Landing point changed = change four things on the spot: frontmatter landing-point sentence · work title · cover headline · file name。 🔴 Don't put it off until before publishing — N23 was still carrying the old landing point for a while.
⚠️ When the landing point changes, it's usually for the better(he is reasoning, the AI is describing).This step is about capturing it, not correcting him.
Phase 4 · Cut only, don't expand
Optimization can only cut from the retelling script, never expand upward from any AI script.
Four must-dos：
- Full red-line table（
违禁词避雷表10 categories + absolutes + symbol level) — 🔴 not grep four words - Unify person(when explaining a mechanism, use a generic term consistently throughout; only switch when discussing a specific industry scenario, and switch only for that paragraph)
- Internal consistency（
/zmm-reviewQuestion 6 (skeleton and breakpoints): do all numbers/times/facts in the text match up front and back) - Diff table + rules into the repository（
总入口§5 four steps: archive the original words → determine which layer they belong to → write them as executable criteria → attach the actual cost)
🔴🔴 The fifth item (added in N23, because I missed it):Anything approved to be added back must be put into the body on the spot and verified with grep
Because of the rule 「never expand upward from an AI script」, the things he deleted in the diff tablecan only be listed and asked about, you're not allowed to add them back yourself. But after he approves? — the original process broke off right here.
N23 actual cost: the sentence spoken on the audience's behalf (theonlysource of the share impulse) — after he approved adding it back, I only landed the other spot; this sentence never entered the body ⇒ it's not in the final cut ⇒ that criterion couldn't be read this time.
🔴 Three steps, none can be skipped：
① 他批准 → ② 当场落进正文（不是写进"建议"栏） → ③ grep 原句验证在不在
⚠️ The same thing has already been taught once: 📣 The checklist quotes the original sentence, not the line number, becauseline numbers silently point to a different sentence。 This entry is of the same kind: a sentence written in the suggestion columnwill silently not exist。 Criterion: after editing, grep that original text. If grep finds nothing = it is not there.
Phase 5 · Title cover + visuals + saving to disk
Title coverGoes through /zmm-title as the owner(must mark structure numbers S1–S12, and must answer “under this title, does the opening of the body need to change”).
Save the final draft to {vault}/_草稿/NN_认知_{标题}.md, format per 00-规则与索引/口播稿输出格式.md。
The filename and the draft's save location must match — if the location changes, rename it; don't drag it to just before publishing.
🔴 Where the cover assets are (N23 addition: I once looked in the wrong directory)
docs/brand/design-system.md ← 视觉系统（暖白纸底 · 档案绿 · 宋体黑 · 刻度线母题）
docs/brand/douyin/cover-template.html ← 抖音封面模板，改帧和标题即可
docs/brand/douyin/cover-00X.* ← 已发的封面，看它们对齐风格
⚠️ Not in _装修/。 N23 I only searched _装修 and _已发布，didn't search docs/,
and as a result made a plain-text cover with black text on a white background, completely detached from the whole visual system, and had to scrap and redo it.
The template's two hard rules(written in the template comments; go read them):
- Each line of the cover's large text ≤6 characters, at most two lines（
口播稿输出格式2026-08-29 wording corrected: originally wrote “≤6 characters”, omitting “each line”) - When both lines are 6 characters, the font size must be reduced(long-sentence mode 108/148), otherwise the size contrast is gone and the template falls apart
🔴🔴 The pitfall of grabbing a cover frame: subtitles are burned into the picture
The template's title area sits right on the subtitle band. N23 measured: the subtitle band is at y1400–1520 (75% down the frame).
Don't just grab a random frame. Sample the whole video at 1.5-second intervals → programmatically filter for frames where the subtitle band is clean → then sort by contrast in the header area. N23 measured: out of 143 frames, only 4 had no subtitles.
🔴 After publishing (N23 addition)
| What to do | Why |
|---|---|
Register _已发布/已发布内容库.md — core conclusion + main analogyFill in both keys |
With one missing, the next dedup check can't be done. ⚠️ N20 was published but never registered; the dedup gate is blind to it |
| Create a T+24h / T+7d data-collection calendar | T+24h is a snapshot; all accounts collect in the same window, so reading late makes cross-comparison impossible |
| Hard-code in the calendar descriptionthe data-collection checklist + minimum view-count threshold | Not reaching the threshold = don't read this time,not that a negative result was read out |
| Copy “what this post can't attribute” into the T+7d entry | To prevent skipping over it during the review |
🔴🔴 In the next draft's frontmatter, open a 承接上条复盘: column, copy this post's review's “N things that carry over to N+1” one by one, and mark each as ✅ 已执行 / ❌ 已否决（理由） |
measured: The N20 review fixed “remove the 1.15× speedup in the next one” —N23 not only didn't remove it, it raised it to 1.2×. The instruction wasn't rejected; it was silently lost.Cost: the editing layer and the copy layer have never once been separated to this day; N23's 24pp retention collapse could have been a clean reading.Pre-publish gate: this field must not have blank entries. Rejection is fine; silent loss is not. |
🔴 Six "must nots" (all have measured costs; not a single one may be softened)
| # | Must not | Cost |
|---|---|---|
| 1 | Write the full draft before all four are locked | N22 v1–v6 Six rounds of rework |
| 2 | Use "my experience" as the skeleton | Three criteria: ① If the entire "I" section is deleted, does the argument still hold? ② If it takes up >1/4 of the body, it is the main subject ③ 🔴 Does the general Douyin audience have this scenario? If not = it can only serve as a single supporting sentenceN22 only passed after being cut from 31% to 5% |
| 3 | Only grep the four banned words | 「US dollars" slipped all the way into the example for 作者方法论 §5.2 — Once a banned word gets into a long-term file, it will copy itself in every session afterward |
| 4 | Cite rules from memory | I once cited a line from 语言风格档案 without having read that file (I got it right by luck that time).§6 loading list says writing the body requires reading it, so you must actually read it |
| 5 | Reconcile conflicting rules yourself | Check 总入口 §3 ruling table;list and ask about anything not in the table, then write the ruling back into the table(this skill has already produced #13/#14/#15) |
| 6 | He decided to keep it, then added it back himself after deleting it | 裁定 #12: execute, and write the exemption next to it |
On-demand loading list (save tokens, do not read it all at once)
| What I am doing now | Read at minimum |
|---|---|
| Judge whether this concept can be made | This skill's Phase 0 + topic pool + 已发布内容库 |
| Lock the four | references/锁四样.md(the only mandatory file included with this skill) |
| Write the body | 04-模板/概念纠偏型口播稿模板.md + zmm-script/references/概念型口播.md + 语言风格档案 |
| Pair with theory | Three admission criteria + five-dimension scoring + theory card "how far it's been verified" (general disciplinary knowledge does not follow citation discipline） |
| Red-line self-check | 规则总表 condensed checklist + 违禁词避雷表 full table |
| Review | /zmm-review four decisions and eleven questions (when the vault has 发布前审核流程.md , that takes precedence) |
Absolutely never do
- Do not write the full draft before all four are locked.
- Do not draw the conclusion "this is a misconception" for the audience — you only point out the crack; that sentence is for them to say themselves.
- Do not perform any conversion action — never include anything like "comment 1 / DM me / link in the comments / contact me if you need it".
- Do not use examples the audience cannot reach (for luxury examples, checkwhether the audience stands in the seller's or the buyer's position）。
- Do not force in a book just to have an endorsement (
§1.65mechanism match <2 is eliminated outright). - Do not add new gates.
Don't know the next step → go back to /zmm。