The technician's no

A technician is under a truck with a fault he cannot place. VEDA has put a list on the screen: check this first, then this, and here is why.

He reads the top line and does not believe it. He has seen this noise before on this model, and it was never that. So he ignores the list, checks the thing his hands are already near, and finds it. The truck is out of the bay by lunch.

On the way past the screen he taps no.

Nobody writes this up. No complaint is raised, no meeting is called, and by Friday nobody remembers it happened. The only trace left in the building is that one word.

Why the no is the most valuable thing in the system

It is easy to assume that a diagnostic tool is learning when it is right. It is not, really. A yes only confirms something it already believed.

A no is different. A no says: on this exact machine, with this history, your reasoning did not survive contact with the truck. And it says so at the moment a person with the fault in front of them made the call — which is the only moment anyone can say it honestly.

A system that hears nothing but yes cannot tell the difference between being right and never having been properly tested. That is why we built the whole thing around the disagreement rather than around the answer. The list is the cheap part. What a technician does when the list is wrong is the expensive part, and until now nobody has been collecting it.

This is also why we do not want the tool to sound certain. Certainty invites two responses from a busy workshop: follow it blindly, or stop reading it. Both destroy the record. A ranked list with reasons attached invites the third response, which is the useful one — an argument.

How the loop actually runs

In plain terms, there are four steps, and they repeat all day:

• VEDA proposes. It reads the signals on the machine's own wiring and the repair history of that model, and puts up a ranked list with the reasoning attached.
• The technician decides. They fix the truck. VEDA has no hands and no vote.
• The verdict goes in. Yes or no, and what it actually turned out to be. This takes a moment, not a form.
• That verdict is the only thing VEDA learns from. Not forums, not the manual, not its own guesses. Confirmed outcomes from people who opened the machine.

That is a narrower diet than most systems run on, and deliberately so. It is the only diet that tells the model something true about itself. Over a year of ordinary work it is what turns three technicians' instinct into something the fourth one can use in their first week.

Where all of that ends up

Every yes and every no is written to the record on the box in the corner of your workshop. It does not go to us, and it does not go to a datacentre. If you are an OEM, it does not leave the site that generated it.

Which means the thing that makes VEDA better over time is the same thing you would never want travelling — how your fleet actually fails, and who worked it out. It stays in the building, and it stays yours.

And the decision stays with the technician. VEDA's job is to get a better starting point in front of them faster. It is not to make the call, and we are not trying to build the version that does.

VEDA · drag to turn