You instrument every system you run. The most important one, you read by feel.
Priors is a friction console. Log a hard interaction and it commits to a forecast that resolves against what the person does next, then hands you the behavioral read behind it and a message tailored to that person at that stress level.
You have a model for everything except the variable that decides most outcomes. The co-founder who went quiet. The report who says fine. The board member who keeps asking the question you already answered. You navigate them on instinct because no one ever gave you an instrument. Priors is the instrument.
How you operate everywhere else
Here is how you operate everywhere else. You isolate the variable. You make a forecast. You let it resolve against what actually happens. You update. It is the only way you trust anything, and it is why the systems you run do not surprise you twice. Then the decisive variable turns out to be a person, and every tool you were handed for that is the wrong shape.
Every tool is the wrong shape
The advice from a smart friend is a story about how the friend would have handled it. The language that asks how it landed and how it made you feel describes a register you do not operate in, and you can feel it not fitting. The books hand you scripts that assume every person across the table is the same person. None of it survives contact with the specific human in front of you, so you fall back on instinct and call it reading the room.
You keep no score
Call it what it is. Every read you make is a forecast, and you have never once let one resolve on the record. You predict how she takes the news, how he plays the negotiation, what the quiet means. Then reality runs, and you keep no score. If you ran your systems the way you run your reads on people, you would have no monitoring, no postmortems, and no idea why anything fails. You would have been fired for it years ago.
Behavior under load is observable
This is not a character flaw, and it does not get fixed by trying harder to be perceptive. Behavior under load is observable. The patterns are stable enough to forecast. A read can resolve against what the person actually does next. The only thing missing was the instrument. Nobody had built the console for the one system that decides most of what happens to you.
You log a friction event the way you would log any other telemetry. Who, the state they were in, how loaded the situation was, what happened, what stopped. Minutes, not a journal. The instrument returns four things.
The pattern under load
The behavioral pattern under that load, what the behavior actually signaled, and the misread you were about to make.
A claim that can be wrong
A concrete, confidence-stated prediction about what the subject does next, inside a stated window. You watch what happens. The forecast resolves against reality, you record the outcome, and your calibration record updates. Misses are kept. A read that cannot be wrong is worth nothing.
Tailored to the subject
One version of what you could say, tailored to that person at that stress level, written for the person receiving it.
Pre-meeting orientation
The pre-meeting orientation. A short read spent on the highest-leverage minutes of the conversation.
Under the hood it is Bayesian, and it does not hide that. Priors seeds each read with a prior, then updates it against what the subject actually does. Your logged events are the evidence. The read is that prior after the evidence has moved it. Month one, you have opinions. Month six, you have a calibration record you can see.
Sunday night you sent a 1,400 word strategy doc. Monday 9:14am he replied "Let's discuss." Nothing since.
Daniel did not go quiet because he disagrees. He went quiet because he is running failure scenarios and has not finished. The terse reply is a placeholder, not a verdict, left so you would not read the silence as him ignoring you while he stress-tests the doc against everything he believes could go wrong.
The currency here is not enthusiasm. It is rigor. He trusts the person who brings him the problem with the plan before he has to find it himself.
You read "Let's discuss" as resistance and walk into Tuesday defending the strategy. That is the error. He is not opposed to the strategy. He is opposed to committing before he has found the failure mode, and right now he suspects you have not looked for it as hard as he has.
Tracked the moment the read generates. When the window closes it resolves against what Daniel actually did, and you record the outcome: hit, miss, or partial.
- →He will not raise the doc in standup. He waits for a lower-stakes channel.
- →Left alone, he surfaces a concern within 48 to 72 hours, framed as a question, not an objection.
- →If you open by selling the upside, he disengages.high likelihood
- →If you open by naming the weakest part of your own plan, he engages substantively in the first 10 minutes.high likelihood
Reach him before the meeting and hand him the rigor he was about to spend two days manufacturing alone. Name the weakest part of your own plan first, then send one tailored line.
"I want your read on the strategy doc, specifically the migration timeline in section 4. That's the part I'd attack first if I were you. Free for 20 min before the 2pm?"
It hands him the failure hunt as the agenda, names the section so he does not re-skim 1,400 words to find his footing, and sets a bounded container that he reads as respect for his time and a signal you are not about to make him absorb open-ended risk.
This is one read on one event. The instrument runs the same way on the report who says fine, the board member who keeps asking the question you already answered, the partner who got a clean fix instead of being heard. Yours is the read it has not run yet.
Priors was built by Johanna Farrimond. Physics and electrical engineering degrees, an MSc in Information Assurance with Honors, and 20 years in IT including a decade in classified government networks. She spent that career reading systems by their telemetry, on the principle that behavior under load tells you what a system actually is, regardless of what it claims to be.
People are systems under load. They have been the one system nobody built the console for. So she built it.
Your log is yours. Real events about named people, handled like the sensitive material it is.
Every read commits to a forecast, and every forecast resolves · hit, partial, or miss, all of them kept. Month one you have opinions. Six months in, this is the record you can see.
month 1 → month 6The waitlist is the first cohort
- →Early access before public launch.
- →Direct input on the instrument while it is still taking shape.
- →The founding operator rate, held for the life of the account.
You're on the record.
You're in the founding cohort. Expect a few worthwhile emails before beta, your launch invite when access opens, then occasional updates. Unsubscribe anytime.