The short-video SOP

Imperative steps, each with a named precedent. The judge here is the least legible on the axis (open-sourcing-the-judge) — so more of this SOP is empirical rule, less is read-from-code.

Phase 0 — pick the machine (once)

1. Decide the content's factory: (a) the product's own output screen (cal-ai-the-screen-is-the-ad — supply scales with usage), (b) a fixed reaction/format pointed at an infinite input stream (khaby-lame-the-reaction-arbitrage), or (c) a repurposing line off one weekly production event (hormozi-the-repurposing-machine). Choose by what you already produce daily.
2. Run a person, not a logo — founder accounts dodge category restrictions and out-reach brand accounts (okamoto-the-cadence-extremum).
3. Mute-proof by construction: the payoff must read with sound off in 3 seconds — 63% of highest-click-through videos hook within 3s, 71% of users decide in the opening seconds (TikTok's own data; converges with Jenny's first-frame rule).

Phase 1 — the production line

4. Test ideas in text first: the cheapest screening layer for video is a post (Hormozi's tweet-test); only proven ideas earn production.
5. Atomize: one production event → many platform-native artifacts, each re-hooked per platform (hormozi-the-repurposing-machine; Jenny's per-platform retuning — the same video did 1M on TikTok and 1K on YouTube until retuned).
6. Cadence is the experiment budget, not a risk: the judge scores artifacts, not accounts — post as much as you can afford to produce (okamoto-the-cadence-extremum at 80–100/day is the measured extremum; account-warmup and over-posting penalties are unresolved folklore).
7. Per-artifact metrics or it didn't happen: 3s-retention and completion are the binding gauges (industry consensus thresholds ~70% at 3s / ~70% completion circulate unverified — treat as directional, calibrate on your own data).

Phase 2 — distribution accelerants

8. Seed with nano-creators paid for labor: $50 to unknowns whose video carries your format — when the format carries reach, cheap creators rent the same judge as famous ones (cal-ai-the-screen-is-the-ad).
9. Scale winners into a managed network (Cal AI's 250 retainered creators) or into a bounty market — per-view clipping bounties at ~$1–8 CPM are a functioning labor market (the-clipping-economy); post the bounty, don't hire the clipper.
10. Never buy followers — the purchasable outcome is the vanity axis; judges score posts (brendan-kane-the-purchasable-outcome).

Phase 3 — convert and hand off

11. The organic ceiling is real and measurable: Cal AI's influencer channel capped at ~$2M/month before paid took over — instrument the plateau and plan the paid/other-channel handoff instead of discovering it.
12. Convert in-app, not in-bio: the screen-as-ad loop closes only if the video's payoff is one search away ("App Store: 搜索名") — the store listing is the landing page (cal-ai-the-screen-is-the-ad).

What this SOP cannot give

The judge's weights are unpublished and the folklore layer is thick (warmup, shadowban, over-posting — all unresolved). The floor is the pipeline above; the ceiling is idea supply and format taste, as everywhere (the-mrbeast-factory §6).