Verification
Is your AI actually correct — and can you prove it?
An AI verification audit is a fixed-price review where we build the answer key your system never had, plant controlled traps, grade the answers and deliver an evidence-backed defect ledger. Fixed price NOK 15,000–25,000, delivered in under a week.
How does the answer-key method work?
We hold no opinions about your AI. We measure it — against an answer key built from your own sources.
1. We build the answer key
The answer key your system never had: correct answers, correct figures, correct sources — derived from your own primary sources.
2. We plant traps
Controlled error sources and edge cases where we know what a correct system must answer — and what a failing one will do.
3. We grade the system
The system is run against the key. Every deviation is documented with page-and-quote evidence and countersigned by an independent model.
4. You get the defect ledger
Scorecard, failure catalogue, security posture and priced remediation slices — a decision basis, not an opinion.
What do you get?
- Scorecard with completion rate and failure catalogue
- Security posture of codebase and API surfaces
- Numeric correctness: can a wrong figure reach the user?
- Defect ledger with page-and-quote evidence per finding
- Independent countersignature on every finding
- Priced remediation slices (fixed price each)
Fixed price
15 000–25 000 NOK ex. VAT
- Scope and price agreed before start — no hourly billing
- Delivered in under a week
- Runs on codebase and synthetic data — normally no DPA required
- Liability governed by the Norwegian SSA-O framework
Voluntary technical quality audit — not statutory auditing under the Norwegian Auditors Act.
Frequently asked questions
What is an AI verification audit?
A fixed-price review where we build an answer key for your system, plant controlled traps, grade the answers and deliver a defect ledger with page-and-quote evidence. You learn where your AI is wrong — with documentation, not gut feeling.
What does it cost?
NOK 15,000–25,000 ex. VAT, fixed price agreed before start. The price depends on scope: number of answer surfaces, data sources and whether figures/amounts are involved. No hourly billing, no surprises.
What do I receive?
A scorecard with completion rate and failure-mode catalogue, security posture, numeric-correctness findings where relevant, and priced remediation slices. Every finding is countersigned by an independent model before delivery.
Can you audit an app built with AI tools (Lovable, Cursor, Bolt)?
Yes. We assess whether the app is real, secure and finishable: what actually works, which of the canonical failure patterns are present, and what it costs to make it production-ready.
Do you need access to personal data?
No. The audit is designed to run on the codebase and synthetic data, so a data processing agreement is normally not required. If we later remediate in systems holding real personal data, a DPA is signed first.
How do I know your findings are correct?
Every finding carries concrete evidence: file and line, page and quote, or a reproducible test. Findings are additionally countersigned by an independent model that re-derives the answer from the primary source. Findings without evidence are not delivered.
Is this statutory auditing?
No. This is a voluntary technical quality audit of AI systems — not statutory auditing under the Norwegian Auditors Act, and we do not act as auditors.
How long does it take?
Normally under one week from access to delivered scorecard. Scope is agreed in a short, free scoping call first.
Receipts, not claims.
Tell us what your AI must answer correctly — and we will show you whether it does.
Book a scoping call