Genie gets better at your estate by working it — on your hardware, in your rack. And every deployment feeds one shared brain that every deployment reads, so a practice that installs on a Monday is not starting from nothing.
Method goes up. Method comes down. Nothing else moves in either direction.
The line that is easy to get wrong“Try the print spooler first” is method. “When the hostname matches this pattern, try the print spooler” is an estate wearing method’s clothes. So a pattern has to hold on cases from practices that did not contribute it, or it is not distributed at all.
Understood means the line can be rebuilt
Every product here says it parses your logs. Usually that means it pulled four fields out and dropped the rest.
Take the structure. Rebuild the line. Compare.
And the remainder is a list, not a percentage99.7% can be true while hiding the only shapes that mattered - because what a platform fails to understand is disproportionately the new. A firmware nobody has seen. A device installed on Tuesday. A tool that has never run here before. So we publish the list instead.
The half nobody collects
Every product records the fix
A resolved ticket. A runbook. A knowledge base article.
None records the four hundred
That a thing was tried four hundred times and worked twelve. So every practice rediscovers the same dead ends, separately, for ever.
And when you overrule it
An expert just labelled an example. It is never shown to anyone as a score — a technician who thinks their disagreements are counted stops recording them.
Nothing enters on its own word
A brain every deployment feeds is a brain every deployment can poison. A compromised MSP does not send an alert — it sends contributions, correctly signed.
Guard
What it means
Three rooms
Intake, proving, production. A production reader has no database grant on intake, so a bug cannot reach it.
Practices, not deployments
Forty deployments of one practice count as one. Independence is proven from billing and tenure, not declared.
Held-out cases
A pattern must improve results on evidence its contributor never saw. If any class gets worse it does not ship.
Small first
A bounded slice of the fleet, measured against those who did not take it. The rollout halts itself if the numbers fall.
Withdrawable, always
Getting a pattern into four hundred deployments is easy. Getting it back out is the part nobody builds.
And the attack nobody discusses is the slow oneA sequence of contributions each individually harmless, whose aggregate moves a measure over months. The defence catches the sequence rather than the item.
It is part of the product, not a tier
You can decline telemetry — the ten fields and a signature are opt-in and off by default, and that does not change. What you cannot opt out of is the software getting better at its own job, and we would rather say so here than bury it in a policy.
And no deployment depends on it. If the shared brain is unreachable, your platform runs exactly as it ran yesterday.
The boundary was written first, and it still holds
We wrote what may never leave a deployment before we wrote what may — and published it before there was anything to publish it about. A boundary set in advance is one you can hold us to, and not one we get to redraw the moment it becomes inconvenient.
She is awake right now, and five estates are talking to her. The brain has a name and an address — genie.dream-genie.ai — and she is not a nightly export or a batch job. She is in continuous conversation with every platform that feeds her, taking in what worked and what turned out to be a dead end, and handing back method as fast as she earns it.
And every exchange proves the shape again. Method crosses. Evidence does not. Not because we filter it on the way out, but because the payload has ten fields and a signature and there is no field for a hostname to travel in. She checks it on every bundle, in both directions, every time one moves — the guarantee is not a policy, it is the shape of the thing.
He runs your estate. She is what he reads. Five estates have made her quick; the sixth makes her sharper, and so does yours — every practice that joins is a practice the rest stop rediscovering the same dead ends behind.
Including if the answer is smallThe claim is that a deployment installed today is not a beginner — and we measure it the only way that means anything: the same estate, the same faults, scored twice, once with the shared brain and once without. We publish the difference and nothing else. If it turns out to be a week rather than a leap, this page will say a week. Every vendor in this market claims their AI learns; we are the only one telling you what the number would have to be to disappoint us.
