Prove an owned model wins on your task, before you pay to build one.
$6–9k · one week, fixed scope · you keep the proof
We test a model you’d own against whatever does the job today, on a sample of your real data, and give you a straight go/no-go, plus the test that proves it, yours to keep whichever way it goes.
Free, about two minutes, no email to see your verdict. Or book a free call.
Not ready for a Pilot? Start with a $1,500 Scan.
A one-day read on a sample of your data: is an owned model worth proving on your task at all? It’s the low-cost way in, and the fee comes off your Pilot if you go ahead. The free fit check points you to the right first step.
A verdict you can act on, and the test that proves it, yours to keep.
The same test behind our public benchmark, run on a sample of your task and your data. Whichever way it goes, you walk away owning the proof.
A straight go/no-go
Does a model you’d own match or beat what you use today, by how much, and why? A clear yes or no, measured on a sample of your real data, not a generic leaderboard. If it won’t win, you hear that plainly.
The test that proves it, yours to keep
Built from your data and your definition of right, with the pass mark agreed up front. Re-run it on the next model, catch the day quality slips, and show the result to whoever needs convincing, your board, your auditor, the team that has to trust it.
What it costs to own, when cost is the point
For cost-driven cases, the all-in cost of running it yourself at your volume, and where it breaks even against your current bill. The number that says whether owning it pays for itself.
The working model itself, not a write-up
You keep the actual model that produced the result, run on your sample and handed over, whatever shape the task called for. The thing, not a slide about it.
One week, fixed scope, three steps.
Agree the task and the bar
Under NDA, we settle the one task, what “right” means, and the mark it has to clear, before any work starts. You share a sample of your data, or we agree how to capture one. Nothing leaves your environment if it can’t.
Run the head-to-head
We build the test from your data, then run a model you’d own against what you use today, on your sample, the same way our public benchmark does it.
Verdict and handover
You get the go/no-go with the margin and the reason, the cost number where cost is the point, and everything handed over outright, the test and the working model both yours. Use it to scope a build with us, take it elsewhere, or run it yourself. If the answer was no, you still own the proof that says so.
Common questions
The fit check (two minutes, free) tells you whether an owned model is plausibly worth pursuing. The Pilot proves it, real numbers on a sample of your data, plus the test that produced them. Fit check first, then the cheap Scan to confirm, then the Pilot settles it.
Then it says so, with the numbers and the reason, and you keep the test either way. A $6–9k no is far cheaper than a six-figure build that quietly underperforms.
You can, if you already know an owned model wins. But proving it first is cheap insurance: the Scan fee credits toward the Pilot, and the Pilot scopes the Build, so it doesn’t cost you time, it saves it.
Prove it before you spend on a build.
Score your taskPrefer email? phil@baseweight.co · We work under NDA.