The X account SOP

The industrializable procedure for building an X account, assembled only from moves with a named precedent — each step cites its case or the judge's own code (open-sourcing-the-judge). Format: imperative steps; no step that can't be checked.

Phase 0 — pick the seat (once)

1. Audience = buyers. Choose the niche so the people the judge assembles contain the people who pay (when-the-playbook-sells-products principle 2; Tibo sold a Twitter tool to Twitter creators; Marc Lou sells boilerplate to the founders watching him).
2. Fix one reprintable format. One visual/textual grammar, rigid enough that every artifact is recognizably a unit of the same series — Butcher's black-and-white geometric card is the extreme proof (faceless, 1.2M followers, the brand IS the format). Blendtec discipline: the format survives unchanged for years.
3. Faceless is viable; persona is a cost. The persona route (Naval/Koe tier) buys reach with a character that must be fed forever (when-the-playbook-sells-products principle 1). Default faceless-format unless the persona already exists.

Phase 1 — read the judge, scrape the winners (week 1, then continuous)

4. Read the rulebook literally. The judge's form is published (~/Develop/reference/x-algorithm): the 19-term action vocabulary, the pre-scoring filters, the author-diversity decay. Diff the repo every 4 weeks — the update cadence is a standing subscription to the judge's drift.
5. Scrape the niche's winners. Jenny's method verbatim (jenny-hoyos-craft-is-scrapeable): crawl the top accounts' posts, extract invariants — length, media type, hook shape, thread vs single, posting windows. The idol is a dataset; numeric gates come from the corpus, not from folklore.

Phase 2 — the production line (per artifact)

6. Inventory → brutal filter. Keep an idea list an order of magnitude larger than output (Jenny: 1000 → 10; Tibo: 11 products → 1 survivor). First filter: "do I want to make it."
7. Engineer each post backwards from the score terms (signs are public, weights are not):
   • dwell / continuous dwell — the card must reward a stop, not just a glance;
   • share-via-DM — make things one person sends to one person ("this is so you");
   • follow-author + profile-click — every post is an ad for the profile; the profile is the landing page (pinned post = the conversion surface);
   • reply/quote — leave one deliberate open edge (Trahan's unresolved arc, ryan-trahan-narrative-for-capital);
   • never buy positives at the price of negatives — not-interested/mute/block/report subtract; engagement-bait is a taxed good.
8. Don't die before scoring. Pre-scoring filters are hard walls: no duplicate reposts, no stale content, respect the seen/served dedup — a filtered post scores zero regardless of quality.
9. Respect the diversity decay. One author's candidates attenuate within a single feed response ((1−floor)·decay^position + floor): spacing beats stacking; a burst of posts cannibalizes itself.
10. Every artifact is an experiment. Pre-register what the post is testing (hook shape? format variant? window?); log per-post metrics; change one variable at a time (Jenny's experiment-per-artifact; the arbiter makes this gradient descent, industrialization-needs-a-mechanical-arbiter).
11. Repeat the winners. Re-post and remix proven artifacts on schedule — "great advertising wears in, not out"; the 1,001st exposure of the format is a +150, not a +1 (jack-butcher-the-format-factory).
12. The cadence floor (when in doubt): one subject you know better than anyone, one insight under 5 sentences, daily, for 6 months (nikita-bier-sell-the-spike).

Phase 3 — distribution accelerants (the ladder moves)

13. Lay-ups, not asks. Pre-produce zero-effort content FOR accounts one tier up (airrack-the-collab-ladder; Butcher's permissionless apprenticeship — illustrate the big account's ideas, they become your distribution). The X-native cash-free variant: performance-vested equity — milestone unlocks against MRR (monthly recurring revenue), which buys sustained promotion and self-corrects when the promoter's interest wanes (tibo-the-reflexive-pickaxe: $3k → $20k MRR in weeks, magnitudes self-reported).
14. Run a counter with provenance. A public, dated, falsifiable goal + a verifiable metric series (Levels' 12-in-12, Marc Lou's Stripe-rendered dashboards, metrics-carry-provenance as a market force). Serialized persistent state turns followers into holders of plot context — the return-visit engine.
15. Give write-access. Let the audience choose the next artifact (Blendtec's "what should we blend"; donation-triggered plot events). Money/attention in = content out.
16. Buy the cold-start. A content corpus, a $500 micro-tool that already has traffic, monthly free-tool launches, programmatic pages — starting assets are purchasable (tibo-the-reflexive-pickaxe).

Phase 4 — convert the stock

17. The feed is the loss-leader; the stock monetizes off-feed (the-mrbeast-factory exhibit 5). Three proven conversion shapes, pick by product:
    • launch into your own audience — Bier: two apps to acquisition (~$30M + Discord) with zero paid marketing, then the terminal rung: becoming the judge's product head (nikita-bier-sell-the-spike);
    • sell the meta-layer — Butcher: courses/community at ~99% margin, "build once, sell twice" (jack-butcher-the-format-factory);
    • sell the pickaxe — Tibo: the tool for the very genre you perform in, ~$8M exit (tibo-the-reflexive-pickaxe).
18. Convert on the machine's clock, not the calendar. Every conversion shape has a decay curve: Bier sells the spike before it decays; Butcher's meta-layer window lasted ~2 years; Tibo priced the platform risk into his exit. Know which clock your machine runs on.
19. The audience ignites; channels sustain. Tibo's mature acquisition is ~70% non-audience (SEO, affiliates, user-generated content) — plan the handoff from account-driven to channel-driven acquisition instead of riding the account forever.

The measurement stack

Per-post log mirroring the judge's terms as far as X analytics exposes them (impressions, profile clicks, follows, replies, bookmarks/shares); weekly single-variable review; monthly re-scrape of the niche winners; 4-weekly repo diff. Numbers carry provenance or they are self-narration.

What the SOP cannot give

The weights are withheld and the ranker is all taste (open-sourcing-the-judge): the procedure guarantees the floor (nothing dies to filters, nothing bleeds to negatives, every artifact learns), not the ceiling. The residual is what it always is — idea supply, taste, position (the-mrbeast-factory §6). Jenny's caveat stands: the rules are real, and a 70%-retention video still did 10M.

Sources: Bier: two viral apps, zero marketing · Bier → X Head of Product · Butcher: 1.2M followers breakdown · Butcher: $180k/mo courses · Tibo: equity-for-promotion, exit · Tibo: 8-figure exit interview