Seventeen demonstrations with companies that don't exist and figures that belong to nobody: two long ones here, a third on the section's front page, and a short one for each of the remaining fourteen specialists. What is real is the mechanism — who applies the criteria, in what order, and what gets ruled out.
⚠Nothing on this page describes work done for any client. The companies, the
people, the emails and every figure are invented on purpose. These are demonstrations of the
method, not a track record.
🔒Behind the scenes there are real engagements documented. None of them appear here, and it isn't for lack of material: what happens inside a company belongs to that company. That's why these demonstrations are invented — it's the only way to show the mechanism without showing anyone.
Demonstration · invented data
Four emails on a Tuesday morning
They all look like opportunities. One is. Janus runs the same criteria over all four and returns a verdict with its reason in one line — without opening a conversation with anyone.
Who actsJanus— the filter on what comes inPinky— decides this one is Janus's and not his own
From: Marta Ferreiro · Head of People, Textiles Arán (312 employees)
Our middle managers are leaving and we don't know why
«Seven exits this last quarter, all voluntary and all at the same level. The exit interviews say nothing useful. Someone recommended you.»
Respond
Open pain, specific, with a number behind it. Comes by referral and from the person who decides. This is the profile that does buy: not shopping for a price, trying to understand what's happening.
From: sales team · an employee survey platform
15 minutes this week to show you the tool?
«Companies like yours are improving their eNPS by 40%. We have slots Thursday at 10:00 or Friday at 12:00.»
Ignore
A template with the name pasted in, a figure with no source, and two calendar slots offered before knowing whether there's a conversation. The urgency is manufactured: no window is closing.
From: an independent consultant
I'd like to understand how you build your diagnostics
«I work on something similar and I'd love you to walk me through your method, the metrics you use and how you structure the interviews. Coffee's on me.»
Ignore
Kill switch: asks for the method, not the service. The score never gets calculated. A coffee doesn't change what's being asked for.
From: a mid-sized investment fund
We're looking at the sector, shall we talk?
«Nothing specific yet. We like to follow interesting companies in the space and keep in touch for later.»
Watch
There's no brief and no offer: there's someone taking the temperature. You don't answer with a meeting, you note who they are and since when they've been looking. If they come back with something specific, there's already a history.
Four unopened emails
Demonstration · invented data
A proposal that never got a yes
Sherk writes the proposal. Before it goes out, it passes through Urrutia, whose only job is to knock it down. Not a reviewer: the worst reasonable buyer, who sees the final document only — never the price that was discussed, or why.
Who actsSherk— writes it and rewrites it all five timesUrrutia— reads it cold and knocks it downPinky— stops the rounds and decides to send it anyway
Wouldn't sign
Pass 1«You're telling me what you do. You haven't told me what's wrong with me.»
Wouldn't sign
Pass 2 — the price was cut by 25%«Still doesn't convince me, and now I'm also wondering why it was more expensive yesterday.»
Wouldn't sign
Pass 3«You promise to talk to every person and then you say you'll do twenty interviews. I have thirty-four. Who aren't you talking to?»
Wouldn't sign
Pass 4 — the previous point was fixed«Now it adds up. But whoever signs the diagnosis is the same one who gets paid to deliver it, and nobody told me that.»
Wouldn't sign
Pass 5«The conflict of interest is better explained. I still think phase two is unnecessary.»
It went out without the yes, with the weak flank accepted knowingly. Chasing a favourable verdict round after round doesn't converge: every fix moves the document and creates a new expectation, and the next challenge lands exactly there.
That's why the stopping rule is set before the next disappointment, not after. And what the process bought wasn't a correction: it was going out knowing exactly where you can be hit.
Pass 1 of 5
And one for each of the others
Fourteen more specialists, fourteen more situations. Same invented companies, same warning: none of this happened. What is real is the rule applied in each one — the same rule written on their card.
«A partner at Grupo Berenguer wants to set up a separate company to sell the routing tool that was built in-house. He says he made it, in his own time, and the company never paid a cent for it.»
What it did
Didn't get into whether that's fair. Asked one thing: where he learned what he knows to build it. The answer was the company's own delivery operation — routes, time windows, incidents, three years of watching them.
What turned up
That the problem doesn't arrive when someone complains, but at the first serious review of the company. A buyer looking at the group will find a tool running on in-house knowledge, invoiced from outside, with no paper governing it. Themis didn't say it couldn't be done: it said it isn't written down today, and named who has to write it — a real lawyer, not itself.
When the person using the tool owns the company, «personal use» isn't a clean scenario.
«We need a growth hacker. Someone who brings us clients through digital marketing and also knows data.»
What it did
Before writing the profile, sent Pax to look at what that role actually is: what someone like that does daily at companies the size of Nordeste Logística, which tools they touch and what they answer for.
What turned up
That inside that request there were two different roles that almost never live in the same person: the one who captures and the one who measures. The profiles doing both either come from companies with fifteen marketing people holding them up, or do both adequately. Nolan wrote the two the research produced and said which to hire first. The other can wait six months; that one can't.
Before defining a new role, you research what that role looks like in the real world.
«Automate the monthly margin tracking for Valdeluz. We already have the close in Excel, it just needs connecting.»
What it did
Traced where every input came from. Twelve came from the ERP. One came from the very close report the dashboard was scoring itself against.
What turned up
That the system would work perfectly for months — while it rebuilt months already closed. And it would fall over the first month it had to be produced from scratch, which is exactly the month someone needs it to decide something. Didn't fix it quietly: flagged it on the dashboard itself, with its own line.
If an input came from the very report you're scoring yourself against, you say so.
«We're about to sign a small consultancy for the rollout at Clínica Sanvitores. Take a look at them first, just in case.»
What it did
Two searches, not one. The first by the company name: clean, nothing. The second by the portrait — sector, size, city and situation, without the name in front.
What turned up
A consultancy of the same size, the same city and the same niche that had walked away from a half-finished rollout at a private hospital fourteen months earlier. Different registered name, the same two surnames on the board. Nothing came up by the name. It came up by the portrait, which is the search nobody asks for.
Verifying an invented name is two searches: by the name, and by the portrait.
«Get me the list of heads of people in the ceramics sector in Castellón and a message for each one.»
What it did
Prepared all forty. And sent none of them: left them as drafts, which is the only thing it knows how to do.
What turned up
In the human read-through beforehand, three came up that shouldn't go out. One was going to a company that's already a client through another line — it would have landed as if we didn't know who they were. Another to someone who left that role five months ago. The third to a company in insolvency: the message talked about growth. Thirty-seven went out.
Nothing gets sent. Every prospecting message leaves as a draft for a person to read first.
«An HR software brand wants to sponsor six posts. They pay above rate and want to close this week.»
What it did
Before looking at the amount, cross-checked the name against the company's client list. The brand wasn't a client. But it competed head-on with one that is, and in the same account.
What turned up
That the decision wasn't its to make. A sponsorship that brushes against a client stops being a personal-brand operation, even if the money comes in elsewhere and even if the two things are legally independent. What reached Manu wasn't a yes or a no: it was what he gains and what he risks in each direction.
Nothing that touches the company passes through here. And if the sponsor is also its client or competitor, the decision isn't made here.
«The Valdeluz report runs to two and a half pages. Make it fit in two.»
What it did
Adjusted it. And then counted: rows before, rows after.
What turned up
That the comfortable fix — lowering the container height and hiding the overflow — left the page perfect and ate nine rows off the sites table. No error, no warning, no visible gap. The document looked better than before. It fitted in two pages the other way: by dropping a column nobody looked at, decided by a person.
«Here are the twenty-four illustrations for Berenguer's annual report. They came out really well.»
What it did
Didn't look at them one by one. Put all twenty-four together in a grid, which is where a batch's defect shows.
What turned up
The same face — young woman, dark hair, smiling at three-quarters — in nine of the twenty-four. None of them was wrong. Together, that face had become the lead character of the annual report of a company that never decided to have one. Didn't ask to change the nine: asked to change the pattern.
When an anonymous face repeats across a batch of generated images, it stops adding variety and becomes a mascot nobody decided to have.
The text that sounded like a machine without one odd word
How it came in
«This text on the website is well written but it sounds like AI and I don't know why. Fix it.»
What it did
Didn't change the vocabulary. Counted the cadence, by hand: standalone sentences placed for drama, slogans closing paragraphs, and the «it's not X, it's Y» construction.
What turned up
Seven «it's not X, it's Y» in eleven hundred words. Eleven one-line sentences placed to land a punch. The vocabulary was impeccable — not one of the words that give it away. What gave it away was the rhythm: every paragraph ended the same, as if breathing with the same lung. One survived. The one that had earned it.
What gives an AI away isn't the vocabulary, it's the cadence.
One hundred and nine files, and the one that said all went well
How it came in
«Rename and reorganise Nordeste's 109 case files under the new convention.»
What it did
Ran the batch. And then counted: 109 expected, 109 at the destination. And still went to look at the disk.
What turned up
Nineteen files renamed halfway. The process had hit a character it didn't expect, done what it could and carried on. No error. No warning. The final summary said completed. Nineteen out of a hundred and nine is 17%, and without looking at the disk you find out three months later.
A batch never fails out loud: it breaks in 19 files out of 109 without a single signal.
«Have the AI generate the template for Sanvitores' monthly report. That way we have it in a minute.»
What it did
Asked one thing before writing the brief: how many times that template will be touched. The answer: every month, for at least two years.
What turned up
That generating it was cheap the first time and expensive the next twenty-four. Every small change meant asking for the whole thing again, checking nothing else had shifted, and reconciling the numbers again. The toll is paid on every round and doesn't show, because each single round looks fast. The template was built by hand, once.
You generate when the result is an image. You build by hand when it's a structure.
«For the cover of Berenguer's annual report I need a delivery depot at dawn. There's no photo and we're not doing a shoot.»
What it did
Generated it. Four versions, with the style reference from the rest of the report attached to each one so the batch wouldn't drift apart.
What turned up
In the first run, invented text on the sides of the vans — letters that aren't words, which at cover size read as dirt. It's the defect generation brings: it only shows up if someone goes looking for it. It came out right on the fourth.
Nova is the only one with no written rule, and the gap stays empty on purpose: it gets filled when there is real work behind it to hold it up.
«The LinkedIn post did really well. Forty thousand impressions, two hundred comments.»
What it did
Looked at the other half of the number: how many of those people ended up somewhere of ours.
What turned up
Three visits to the owned domain. Forty thousand people read the thesis, argued about it and learned it — inside LinkedIn. A month later nobody could find that post, and what was left of it was authority for someone else's search engine. The proposal wasn't to stop publishing elsewhere: it was for every piece to have its full version at home.
The authority you build in someone else's house stays in someone else's house.
«Review the access rules on Arán's file server before connecting it to the new office. It's fairly tight, there are forty rules.»
What it did
Didn't count forty. Sorted the forty from most permissive to least and read the first one.
What turned up
A rule with a wildcard, added four years earlier for a migration that lasted a weekend. With that one in there, the other thirty-nine are decoration: access comes in through the top one anyway. Nobody had removed it because nothing had ever failed. A long list feels like control, and that feeling is exactly the risk.
The risk in an allowlist isn't its length, it's its most permissive entry.
Whoever owns the context is the one who accelerates
The tool isn’t the advantage. Whoever owns the context, works it, and looks after the data and the content: that’s who really accelerates all of these systems. If you want to go deeper into how to cultivate your talent, the whole method is one page away.
Janus — the filter on what comes in
Pinky — decides this one is Janus's and not his own
Sherk — writes it and rewrites it all five times
Urrutia — reads it cold and knocks it down
