Independent AI evaluation

Prove your AI works.

Your AI feature is live. Sooner or later someone senior asks you to prove it gives the right answers, and right now you cannot. We test it against your own real cases and hand you the evidence in five working days.

INDEPENDENTEVIDENCE NOT OPINIONFIVE WORKING DAYSINDEPENDENTEVIDENCE NOT OPINIONFIVE WORKING DAYS
The gap

Nobody in the room is paid to find the fault.

01

Your cloud bills you when it runs

The architect your cloud provider assigned you is helpful, free, and earns nothing by telling you your AI is wrong. Every answer it gives, right or wrong, lands on your bill.

02

Your builder is marking their own homework

The team or agency that shipped it cannot be the one that signs it off. They are defending the thing they built, and everyone in the meeting knows it.

03

A chat window will not stand behind anything

You can ask an AI how to test your AI. You end up with your own opinion in a longer document, and nobody outside the building to point at.

What you get back

Evidence you can forward to the person asking.

A scorecard on your real cases

We build a test set from your own domain, not a generic benchmark, and score every single answer right, wrong, or actively dangerous.

The failure map

Where it breaks, what the breaks have in common, and which one a security reviewer or an angry customer finds first. Worst first, not alphabetical.

What each answer costs

The real cost per query today, and what the same system costs at ten times the traffic. Most teams have never once measured this.

How it runs

Access. Cases. Run. Report.

01

Access

Read only access to the system, plus one call to learn what a right answer actually looks like in your business. That definition is the whole job, and only you have it.

02

Cases

We build a real test set out of your own domain. Your documents, your edge cases, the questions your customers actually ask, including the awkward ones.

03

Run

Your system answers every case. We score the lot, then go back through the disagreements by hand, because that is where the interesting failures hide.

04

Report

You get the scorecard, the failure map and the cost, in a form you can send straight to whoever asked. Five working days from access.

The terms

Every number here is a term of the deal, not a boast.

0
working days from access to scorecard
0
well-architected pillars the report covers
0%
of the fee credited against the engagement
0
people in your chain paid to find the fault
Price

One number sits on the button. The other is earned.

Start here

The Proof Run

£2,000

The diagnosis. Buy it here, today.

  • A test set built from your own domain
  • Every answer scored right, wrong or dangerous
  • The failure map, worst first
  • Cost per answer, now and at ten times the traffic
  • Five working days from access
  • Credited in full against the Evidence Engine

The Evidence Engine

£15,000

The fix. Sold on the call, never off a button.

  • The eval harness built inside your own cloud account
  • Guardrails on the failures that actually matter
  • Monitoring, so a bad change is caught before release
  • Full handover, your team owns it after
  • Your Proof Run fee comes off the price

The slow route

Four weeks

An AWS consultancy staffs this with one or two consultants for under four weeks, then quotes. That is not our guess, it is what they print in their own job adverts.

  • Procurement first, findings later
  • Usually the same firm that would build the fix
  • You still write the test set

If the Proof Run finds nothing worth fixing, you do not pay. You keep the scorecard either way.

Find out before your customer does.

£2,000. Five working days. If it finds nothing worth fixing, you do not pay.