Founder Signal OS
A creator-owned intelligence layer that records what makes you irreplaceable and holds that line across every campaign your agents touch.
Origin. The specific history that makes your position earned rather than claimed.
0 of 12
Answer all three to continue.
The problem
Ask any competent model to write as a founder and it reaches for the category. It has read ten thousand people in your position and it returns the median of all of them, in your font. That is why AI output feels close enough to ship and never sounds like the person who built the thing.
Sabrina Stocker's Impossibly Human thesis names the problem correctly: the founder is the brand, and the least copyable thing is the human. This is the infrastructure question underneath it. What has to be written down, and who has to own it, for that to survive contact with an agent fleet?
What a brand guide captures
Adjectives. Three tone words. A list of phrases to avoid. Enough for a freelancer to sound plausible for one campaign.
What a signal record captures
The argument you lost that changed your mind. The client you fired and why. The sentence you keep repeating because it keeps working. Retrievable, attributable, yours.
What gets recorded
Not a summary of your personality. A specific claim, where it came from, the evidence behind it, the conditions under which it holds, your actual phrasing, and what an agent is never allowed to say instead.
A few hundred records like this is the difference between a model impersonating you and a model working for you.
authority-graph
# signal/beliefs/pricing-refusal.yaml
claim: "I do not price by deliverable."
origin: 2019-retainer-collapse
evidence:
- call/2026-03-11#t=00:14:22
- post/why-i-stopped-quoting-hours
holds_when: buyer asks for a rate card
breaks_when: procurement requires line items
voice:
phrasing: "You are not buying hours. You are
buying the call I make at hour three."
never: "flexible pricing options"
approval: humanThe four layers
The specific history that makes your position earned rather than claimed.
Without a recorded origin, every model writing in your name reaches for the category story instead of yours. You become interchangeable with anyone who read the same books.
The decision rules that produce your calls, separate from the calls themselves.
Judgment that lives only in your head cannot be delegated to a person or a model. Every piece of work routes back through you, and the business stops scaling at your calendar.
The actual language, not a description of the tone.
A tone-of-voice document produces an average. Your real phrasing gets smoothed into the median of everything the model has read, and your audience feels the swap before they can name it.
Whether what the audience does changes what the system knows.
Without a feedback path, every campaign starts from zero. You publish more and learn nothing, and the tenth month of output is no sharper than the first.
The pilot
The pilot runs underneath something already scheduled, because a system that only works on a sample never survives a real deadline.
Two sessions and an archive sweep. Your talks, calls, drafts, and dead ends become a sourced record instead of a personality summary. Nothing synthetic enters at this stage.
That record becomes an authority graph: beliefs, mechanisms, refusals, proof, and the contradictions you have earned the right to hold. Averaging is the failure mode, so nothing gets smoothed.
One real campaign runs on top of it. Thesis, sequence, and distribution brief all generate from the same governed source, and every publish still passes through you.
Replies, objections, and conversions feed back into the graph. Month six is sharper than month one because the system learned, not because you worked more hours.
The boundary
Delegation without a boundary is just slower replacement. These stay with you, and the architecture is built to make that enforceable rather than aspirational.
Everything else is research, recall, drafting, sequencing, and distribution. That is the part worth automating, and it is most of the hours.
Open proof
These are repositories, not slides. Fork them, run them, or decide from the source that this is not for you.
The six-layer substrate for research, strategy, production, distribution, and learning. The creator holds the audience model, the voice record, and the performance history.
The execution layer: skills, workflows, specialist agents, and safety hooks that run inside Claude Code, Codex, Cursor, and Gemini rather than a closed product.
Sovereign memory and governance across the agent fleet, so decisions and source material outlive any single tool or session.
The public specification the architecture answers to: portable modules, explicit approval gates, evidence trails, and creator-owned state.
Apply
I am selecting one founder whose body of work is substantial and scattered, with a real campaign on the calendar. No price is listed because the first one is a proving run, and I would rather agree the terms with the person than publish a number at a stranger.
Started at Mindvalley University 2026 in Tallinn and shipped as an entry to the Social Media Mastery 7 Days Launch Challenge run by Sabrina Stocker and Vishen Lakhiani. The scan, the scoring model, and the underlying stack are open. This is an independent FrankX project and is not organized, sponsored, or endorsed by Mindvalley.
Read the Tallinn field journal