ZynoviqWhen someone asks how your AI was tested, have the answer.
Auditors, risk committees, insurers and government buyers are starting to ask who tested an AI system, against what, and when. Every Zynoviq evaluation produces a versioned record and a report written for that reader. It is evidence of testing, not a certification of your AI.
See a sample report.
Every result traceable to its source
A score on its own proves nothing. The record ties every number to the system, the benchmark version, the run and the professional judgment behind it.
| Field | What is recorded |
|---|---|
| System | Product, version and configuration, as declared by the customer |
| Benchmark | Name, version and valid-as-of date of the law it reflects |
| Run | Date, environment, method (who ran it, how answers were marked) and the number of cases run |
| Result | Scores by capability, critical failures and the outcome of every case |
| Basis | The certified answer, legal authority and professional review behind each judgment |
| History | Earlier runs on the same system, with regressions and recoveries between them |
Written for the person who has to rely on it
The report is written for a reader who was not in the room: an auditor, a risk committee, an insurer, a procurement panel. It contains, in this order:
- Scope and limits: what was tested, what was not, and in which environment.
- Method and marking rules: how cases were built and certified; rule-based marking for structured fields, qualified professionals for free text.
- Benchmark version and valid-as-of date.
- Results, critical failures first, then scores by capability and case-level outcomes.
- The customer's response to each critical failure, labelled as supplied by the customer.
- The testing record above, including history.
- Independence statement and dual-role disclosure: ground truth set by qualified professionals independently of the AI vendor; Zynoviq's Enterprise AI division disclosed.
Review of the report format with practising audit professionals before launch: Planned.
Who reads it
- Statutory and internal auditors, asking how an AI that touches the books was checked.
- Risk and audit committees, deciding what the AI may do without a person.
- Professional-indemnity insurers, asking what testing sits behind AI-assisted advice.
- Government buyers, at acceptance and again at audit. For government AI vendors.
- Procurement, comparing vendors who all say their AI is accurate. For buyers of AI.
Findings, not certificates
| It is | It is not |
|---|---|
| A record of which system was tested, against which benchmark version, when, and with what result | A certificate, approval, badge or assurance of the AI system |
| Findings against ground truth set by qualified professionals independently of the AI vendor | An independent evaluation, audit or certification of the system |
| Every failure with the correct answer and the rule behind it | A legal or tax opinion on your affairs |
| A report written so a third party can read it without us | A statement that the system is safe, compliant or fit for every use |
| A history of re-runs that shows whether quality moved | A guarantee about how the system will behave in production |
"Independent" means only that the expected answers are set by qualified professionals independently of the vendor of the AI being tested. Conflict-of-interest policy.
Our dual role, disclosed
Zynoviq Solutions Private Limited also runs an Enterprise AI division that builds AI products. Our evaluation work is delivered by separate teams; we never test, grade or produce data for the division's products; our reports never recommend them; and if your system includes one, we tell you in writing before we test. Every report states this. How we stay separate.
Evidence — FAQ
Is this a certification?+
No. It is evidence that a named system was tested against a named benchmark version on a date, with the result and the basis for every judgment. We certify our answers, not your system. What we do not do.
Can our auditor rely on it?+
That is for your auditor to decide. The record is written so an auditor, risk committee or buyer can read it without us: scope, method, marking rules, benchmark version, results and the basis for each judgment are all stated. We are not your statutory auditor and the record is not an audit.
What if our system fails some cases?+
The report lists critical failures first, each with the correct answer and the rule. You may add your response to each one, which appears in the report labelled as supplied by the customer. A later re-run shows whether the failure was fixed.
Do you have a stake in the result?+
We are paid for the testing, not for the result, and we never test products of Zynoviq's own Enterprise AI division. That dual role is disclosed in every report. How we stay separate.