A free diagnostic · Opens September
There’s a line through your AI workspace. Below it: everything your AI is told. Above it: everything it’s held to.
By Pierre Boutquin · Founder, Curio Chat Academy
Monday you fix it. Thursday you’re fixing it again. If you use AI every day, you know the ritual: you corrected the tone on Monday — we don’t open with “I hope this finds you well” — and by Thursday it’s back. So you fix it again, a little more tersely, and some part of you starts keeping a list of things you apparently teach forever.
Before any framework, two counts from last week. How many corrections did you type into an AI that you had already typed before? And how many drafts did you read end-to-end, personally, before they shipped? The first number is what re-teaching costs. The second is the going rate for being the only no in the building — and unlike the first, it grows as your business does.
This page explains where both numbers come from — and the one line that decides whether they ever change.
01The Part Nobody Says
You got here by doing everything right
Here’s what makes the Thursday ritual maddening: it happens to the organized ones. You kept the record. You wrote the brief that says whose business this is. You corrected carefully, in writing, more than once. Every upgrade you bought made your AI better informed than the version before it.
And none of it changed your Thursdays. So the standard advice arrives — write a longer brief, add more instructions, correct more thoroughly — and it all pours into the same half of the workspace that was already full.
The fullness and the repetition are the same fact. Everything you built sits on the informed side of a line you were never shown. And information, however complete, does not hold drafts.
02Two Pieces of Furniture
The cabinet is full. The desk is empty.
Picture a new hire. You gave them the filing cabinet — every project, every deliverable, the client list, the style guide. You even gave them the corrections binder: six months of your rulings, dated and explained. They read all of it, carefully.
And on Tuesday of week two, their confident proposal — the exact thing that blew up last spring — goes straight to the client anyway. Because nothing in your firm was built to stop it. Reading a correction and being stopped by one are different events.
Every real business already knows the missing piece: the desk the work must cross before it leaves the building, where someone with the authority to say not like that — not to this client, not with that claim can hand it back before the courier arrives.
In your practice, that desk exists. It’s you. Every draft crosses your attention or ships unexamined — which works exactly until Thursday, when the seventh draft crosses a tired desk.
I was that desk for most of a career — 36 years on bank systems where a wrong number wasn’t a typo: it was an incident. What fixed it was never a better filing cabinet, and it was never me reading more carefully on Thursday. It was a step the work had to cross that didn’t get tired.
Two questions place any AI workspace
Told: no · Held: yes
The empty gate
A checker with nothing to check against. Theatre, until the record and the brief exist underneath it.
Told: yes · Held: yes
Installed
The only box where a draft can fail — and the only one where your corrections outlive your attention.
Told: no · Held: no
The stranger
Smart in general, blank about your business. Thursday isn’t even its fault — it never knew your rules existed.
Told: yes · Held: no
Briefed, not bound
The standard profile in this market: full cabinet, real corrections — and nothing that can stop a draft. Home of the Monday re-brief. Every upgrade so far moved you deeper into this box, not out of it.
03The Framework
Where the line falls
There is a line through every AI workspace, and it doesn’t fall where people expect. Below the line sits everything your AI is told: the record of the work, the sources under your claims, the brief it reads before it starts. Above the line sits everything your AI is held to: corrections written where a draft can break them, a step that can stop the draft when it does, a no that stays in force after you stop watching — and a wrong no that gets overturned on the record instead of quietly mutating.
Below the line, your AI is informed. Above it, it is governed.
The two halves fill by different mechanisms. The bottom fills just by working — diligence fills it, note-taking fills it, even a memory tool can fill it on your behalf. The top half never fills that way, because a no only exists where someone decided the system should be able to refuse.
Above the line — what your AI is held to
-
Reversal — a no that proved wrong, overturned on the record: dated, with the old version kept.
-
Reach — a no from months ago still binding work you weren’t watching.
-
Hold — a step that can stop a draft, and has — one that isn’t you reading every line.
-
Verdicts — corrections written where a draft can break them, with what happened attached.
Fills only by deciding.
The line
Below the line — what your AI is told
-
Brief — the system reads whose business it’s standing in — every session, not once.
-
Grounds — the claims you ship rest on sources you can put on screen.
-
Trace — the work left something behind.
Fills by working.
04How To Read It
Read the shape, not the score
The ladder is not a completeness scale, and the instinct on hearing “seven rungs” — I must not have written enough down — is exactly backwards. The most common profile in this market is full below and empty above: informed, ungoverned. The record is complete, and nothing can hold a draft. More files is the shape of the problem, not the distance from the solution.
That profile never crosses on its own, because everything it does well is the half that fills by working. And the two most expensive states on the whole ladder aren’t the empty rungs — they’re the ones you’d swear you have: the brief that exists and is never loaded, and the checkpoint that exists and has never held a single draft. An empty rung gets fixed. A sworn-but-unproducible one gets defended.
So the only instruction this framework ever produces is one rung: the first one you cannot put on screen, counting up from the bottom. Never a list.
And until you can draw your own ladder, one question sorts every tool you’ll ever be pitched: where’s the step where a draft can fail? If the answer is a score, a dashboard, or “you review it” — you just heard the answer.
05The Diagnostic
The Thursday Test finds your first broken rung
The Thursday Test is that line turned into an instrument: seven questions, ninety seconds each, one for each rung. The rule that makes it honest — you answer by putting the artifact on screen, not by remembering that it exists somewhere.
Every answer scores one of three ways. Grounded: on screen inside ninety seconds. Absent: nothing there — honest, and fixable. And the one the whole test exists for — unshowable: you believe it exists and cannot produce it. Unshowable is worse than absent, because an absent rung gets fixed while an unshowable one gets defended — and everything above it inherits the fault silently.
One scoring rule catches almost everyone, kindly: you are not the gate. If the only step between a finished draft and the world is you, reading every line — that was question five, and the honest score is absent. Your attention is real work. It just isn’t installed anywhere except in you, and it goes home when you do. That’s the whole Thursday mechanism: the information survived; the enforcement was never installed.
I’ll be honest about which rung I expect that to be. Five is where most people land, and it’s where I landed — record kept, corrections written, brief in place, and still nothing between a finished draft and a client except me, reading. That is not a diligence problem, and it does not get fixed by trying harder on Thursday.
When it opens
- It runs in the AI you already use — nothing to install, nothing to upload.
- The test itself won’t ask for your email. Taking it requires nothing but the ninety seconds per question.
- It will tell some people, on the strength of their own answers, that there’s nothing here to sell them. On purpose. That part is what makes the rest of it worth taking.
The Thursday Test opens in September
Leave your email and you’ll get one message the day it opens, with the test in it. That is the list’s entire job — no drip, no countdown theatrics.
🔒 100% private. Unsubscribe anytime. No spam, no sharing.
What vanishes by Thursday was never the information. It was the enforcement.
Put me on the list →Pierre/
Founder, Curio Chat Academy
The Thursday Test · Evidence diagnostic
Seven questions your workspace answers — not you.
It runs inside the AI you already use. You paste one prompt; it walks the seven rungs and asks you to put an artifact on screen for each one — the file, the source, the rule, the held draft. Ninety seconds per rung. What comes back is one line you bring here, and the first rung you couldn’t produce.
It opens in September. Here is exactly how it will run, so you can decide now whether it’s worth your ten minutes then.
-
Copy the prompt
Published in full, on this page, before you copy it. You read exactly what you’re about to run — there is no hidden half.
-
Open the AI you use most
Whichever one that is. If yours can reach your files it will check them; if it can’t, it asks you instead — and the result says which of the two happened.
-
Bring back the last line
One line, and it’s the only thing that travels: seven state tokens, whatever name you gave it, and two flags. No file names, no paths, no quotes from your work — there is nothing in it to redact, which is why we don’t ask you to.
Two things that will still be true in September
- No email to run it. The prompt sits on the page; reading it, copying it and running it require nothing at all.
- Two of the four results tell you there’s nothing here to sell you. That isn’t modesty — it’s what makes the other two worth believing.
Or — it’s the same page.