Have We Already Done This One? Asking Your Own Tuning Archive

Every busy tuning shop is sitting on a question it cannot ask: have we already done this one? The files are there — years of projects on a workstation, an external drive, a NAS in the corner. The answer is in them. There has never been a way to ask.

The job you already finished

A read lands. Same engine you have seen a hundred times, and something about it is familiar. Maybe you did this exact ECU eighteen months ago for a different customer. Maybe you did something close enough that the hard thinking is already done. But “familiar” is not a search, so you tune it again from scratch — carefully, properly, and for the second time.

We measured how often that happens, on a real workshop archive of 167 delivered jobs. One job in six had already been finished by that same shop. Twenty-nine files out of 167 were work the shop already owned and could not find.

That is not a tidiness problem. That is a day a week.

Why searching does not work

The obvious answer is to search by filename or checksum, and it fails immediately. The same physical ECU read with a different tool produces different bytes — different length, different padding, different checksum. Two reads of one car look like two unrelated files to anything comparing hashes.

So the shop falls back on memory and folder names. golf_7_stage1_final_v2_REAL_final. You know the folder.

What the Relay actually does

The AI Relay runs on the machine that holds your projects. It reads them where they are and answers your portal. Your files stay on your machine until you click. Nothing is uploaded to run a search; the Relay indexes locally and the portal asks it questions.

Instead of hashing whole files, it reads the identity the ECU itself carries — the software and hardware numbers written at fixed positions inside the binary. Two reads of one car disagree about almost everything and agree about that.

The result, on the same 167 jobs:

• 167 of 167 files found their own work back.
• 40 of 40 unrelated files made zero false claims — it says nothing rather than something wrong.
• Files sharing an identity agreed on 72.2% of their bytes (median 86%), against a 21.4% control group of files that merely looked similar.
• One identity linked nine files no checksum could connect. Sixteen out of sixteen were answered by identity alone.
• Where the identity matched fully, a 2 MB read sat inside an 8 MB read as one intact block — nine chunks out of nine.

It needs no map definitions and no damos. It reads what is in the file.

What you see in the portal

A file arrives. Before you open anything, the job tells you whether this archive has seen it before, what was done, and how confident that link is. If the answer is “we did this in March”, you open the March project. If the answer is “nothing here”, it says so plainly — which is worth as much, because a wrong suggestion costs more than no suggestion.

The control group is the important half of that. Forty files that had never been touched by the shop produced no matches at all. A system that guesses is a system you stop trusting after the second bad guess.

An honest description of the limits

This finds your own work. It is not a shortcut around tuning, it does not write a calibration, and it does not tell you a file is safe. It answers one question — have we been here before, and where are the notes — and it answers it without your archive leaving the building.

It also needs setting up. Indexing years of projects on your own hardware is a real piece of work, and it is different in every shop: different tools, different folder habits, different volume. That is why there is no buy button on this one.

Talk to us about it

The AI Relay is available on request. If you have an archive worth asking questions of, tell us roughly how big it is and what you tune with, and we will tell you honestly whether it will help you — including if the answer is no.

Talk to us about the AI Relay →