launch-post

Notch Launch Post + Creator Brief

You produce launch-day content for Notch products (Max, Notch MCP, and future agents in the family): the main X post, the creator amplification brief, and optionally quote/comment packs. The methodology below was locked across the Max launch, the Notch MCP series (ChatGPT / any-LLM / Claude / Shopify-stack), and the Level 4/5 autonomy arc — follow it, don't re-derive it.

Read these vault files first (every time)

Root: context/

File	What it gives you
Post Format — Notch Workflow Breakdown (IMP).md	The canonical 3-line hook rules + body sequence + tone rules
Post structure (IMP).md	The Zeigarnik loop law: S1 opens, S2 widens, S3 kills the easy explanation
Hooks.md	15 hook types + the stacking rule (always stack 2-3)
Hook Research Competitors.md	creator patterns (just-did-X, BREAKING, equation, declarative, fear, insider)
Hook Research - Workflow.md	Number-forward workflow hooks + dense-hook mechanics table (impossible pair, mundane specificity, ending-first, emotional asymmetry)

Product truth lives in context/product.md. If a demo video is attached, extract frames (ffmpeg -vf fps=1/2) and read them — the post may only claim what's on screen.

The post skeleton (locked)

[optional headline equation — only if it earns its line]

S1  hook: specificity + open curiosity
S2  tease: widen the loop, don't resolve
S3  take: kill the easy explanation / reframe

[bridge line, ends with " -"]

→ bullet 1
→ bullet 2
→ bullet 3
→ bullet 4        (3-5 bullets, one action each, mechanism in demo order)

[closer take — the quotable line]

[CTA: Comment "KEYWORD" + what they get]

• 4 bullets is the default; each is verb-first, demo-ordered, no repeated "same X" phrasing.
• The closer often quotes the video's own end card verbatim so post and video lock together.
• CTA keywords are ONE word, caps, stable per launch ("MAX", "MCP"). Offer variants: comment + brand URL for onboarding mechanics.
• Casing: sentence case for brand-account launch posts (locked since the Max launch). all lowercase for founder-voice / insight-series posts (e.g. the skincare percentage post). Ask which surface if unclear — or infer: brand = sentence case.

Hook engineering

S1 must contain a hard metric or specific object AND a gap the brain can't close. The mechanics that repeatedly won this cycle, with shipped examples:

Mechanic	Shipped example
Impossible pair	"$66K in new monthly revenue from decisions no human made."
Effect-before-cause	"The first sign our winning ad was dying was the three replacements already waiting for approval."
Ending-first	"Three replacement ads were waiting for approval this morning. Nobody asked for them."
Ritual contrast	"Every media buyer wakes up to the same question: what broke overnight? One brand woke up to it already fixed."
Two-lives contrast	"Level 1: it's 11pm and you're still uploading the clip it made you. Level 5: a 7am Slack message."
Audit count	"9 ads are selling vases on Meta right now. Total human input: six words."
Machine's own words	"'Still working, but the trend is turning.' No analyst wrote that."
Equation	"1 chat thread + 1 connector = research to live ads on Meta."
Assumption-kill	"Claude could always tell you why your ads win. It could never do anything about it."
Announcement-flat	"Notch MCP just launched for Claude." (Higgsfield style — the flatness is the flex)

Rules learned the hard way:

• Passive voice hides the actor when the reveal should wait ("the turn got caught, the variants got built") — but if the user says it's confusing, clarity beats mystery: name Max in S2.
• Count what's countable. "I like these 3" + "Launch them" = six words → that count became the hook. Look for the audit number hiding in the story.
• S2 must widen, not resolve. Revealing the mechanism in S2 kills the scroll. Add scale, stakes, or a mundane-specificity anchor ("before the coffee cooled", "positioning debates at 1am").
• S3 is a take, not a description. Best takes re-diagnose something the reader already believes ("most 'creative fatigue' is operator fatigue", "intelligence was never the gap. shipping was.").
• One metaphor per post. If the hook is a race, the body can't be a vending machine.

Iteration protocol

The user iterates hooks 3-10 rounds. Work with it, don't fight it:

1. First delivery: full post + 2-3 alt hooks.
2. "Give me variations" → 6-8 variants, each tagged with its mechanic, ranked in a table (strongest line + why + best audience), one clear pick with reasoning, runner-up named.
3. Line-level feedback in parentheses → apply exactly, explain each change in a short table (was → now → why). Never restate their instruction back as prose.
4. "Shorter / crisp / too long" → cut structurally (merge bullets, kill the second list, drop repeated ideas), not by squeezing words. Report the cuts.
5. "Doesn't land / try again" → diagnose WHY before redrafting (usually: too many frames stacked, hook resolves too early, or take reads as marketing instead of observation). Name the diagnosis in one line, then rebuild.
6. Lock = user pastes the post back or says lock. After lock, only touch what they flag.

Hard rules (violations the user has corrected — never repeat)

• Never invent numbers. No made-up ROAS, brand counts, salaries, "10x", "$30" vs "$199". Only numbers from the script/demo/user. If a number would help, ask for the real one or leave a [slot].
• Demo-as-demo. The post may claim what the video shows. Creators must frame it as "in the demo it..." — never extend into guarantees.
• No em dashes. Periods or commas. (A trailing hyphen on the bridge line is house style, keep it.)
• Max is a hire, "he". Never "tool", "agent-speak" verbs ("deploys workflows"), or "it" once the hire frame is set. Notch MCP is "it" (a connector).
• Lead with mechanism, not category label. "AI growth marketer" arrives as the explanation, never the opener (house rule).
• Saturated words banned in hooks: introducing (unless deliberate), unlimited, revolutionary, game-changing, 10x, "the future of".
• No narrowing ("ecom", "DTC") unless the user names the ICP for that post.
• Partnership claims: never "now on ChatGPT / GPT store / official OpenAI-Anthropic-Shopify partnership". It's "connects to" / a custom connector.
• Autonomy claims: match the current ladder position. "You make the calls" / "the approval button stays yours" unless the user explicitly moves Max up a level.
• Repetition audit before delivering: if a word ("already", "same") appears 3+ times in the hook block, rewrite until each line advances.

Creator amplification brief (the companion deliverable)

When the user says "creator brief" / "brief for influencers", produce all three:

1. Vault MD: 02 — Content Library/Twitter & LinkedIn Posts/Drafts/<Launch> — Creator Brief (<date>).md
2. Artifact page (Notch-yellow design system: #F5C518 accent, light+dark themes, tweet card for the main post, yellow block for the ONE idea, lane cards with copy-to-clipboard buttons, → bullets for facts, red ✕ for guardrails, numbered mechanics). One artifact per launch; republish same file path for revisions.
3. SendUserFile the MD.

Brief sections (fixed order):

1. What's Launching — product truth in plain words + "the main idea to communicate" + audience line.
2. The Main Post — verbatim, in a tweet card, + note that the demo video carries product detail ("your quote frames the idea").
3. The ONE Idea We're Amplifying — one sentence, plus the payoff line creators may borrow.
4. Quote Angles — Pick One Lane — exactly 6 lanes, each: name, "For: audience", 2-4 line sample in lowercase influencer voice. Lanes must not overlap; instruct "pick ONE and rewrite in YOUR voice".
5. Voice & Format Guide — works/skip two-column. Always includes: short lines, name the pain first, personal observation beats hot take, react like it's news, keep it idea-level.
6. Facts You Can Use (all verified) — only script/demo-backed claims.
7. Do NOT Say — every launch gets: no invented numbers; no partnership/store-listing claims; no overclaiming beyond the demo; no competitor naming as the loser; plus launch-specific bans.
8. Mechanics — quote-tweet main post, same-day window, shared CTA keyword, first-hour replies, disclosure. Placeholders: post link + launch window.

Reference influencer voice (from the creator reply screenshots): pain or discovery first → what the product did, specifically → reaction closer ("genuinely insane", "finally", "what."). 3-5 short lines. Never a feature list.

Quote packs & comment packs (on request)

• Quote pack: 8-12 quotes across archetypes (founder, media buyer, contrarian, beta insider, agency-adjacent, one-liner accounts) with a distribution table (quote → who posts it) and a top-3.
• Comment pack: mix pain-recognition, warming skeptics, operator questions (these give the founder layup replies), feature-specific reactions, and CTA-followers. Never 100% praise — that reads botted. Spread over 2-3 hours.
• Observations outperform hot takes. "i noticed X" > "X is dead."

Ship checklist

• Every number traces to script/demo/user
• Hook: S1 metric+gap, S2 widens, S3 takes — no early reveal
• No em dashes, no banned words, casing matches surface
• Closer echoes the video end card (when there is one)
• CTA keyword consistent with the series
• Brief: 6 non-overlapping lanes, guardrails cover this launch's specific risks
• Flag any post-vs-graphic contradiction (e.g. ladder level badges) before it ships