1. Who this is between
“EvalQA”, “we” is the company operating eval.qa. “You” is the person or organisation using it. Using the site, the qabit console or adapter, the evalqa CLI, or engaging us for a Baseline Assurance Sprint or the Continuous Assurance Platform means these terms apply, unless we have signed something with you that says otherwise — in which case that document wins.
2. What we provide
Independent technical verification of analytics agents that generate SQL against your warehouse: invariant synthesis, in-perimeter execution, adjudication by qualified reviewers, and signed assurance bundles; plus the qabit console and adapter. Availability targets and support commitments exist only where an order form says so.
3. Independence
The Independence Charter is incorporated into these terms. In particular: we take no success fee; we do not author your production prompts, semantic models or dbt models; and a confirmed Sev-1 finding cannot be redacted from a final bundle. You may decide who sees the bundle; you may not ask us to make it incomplete.
4. Your data stays yours
You own everything you give us access to and everything we produce about it — invariants, findings, bundles. You grant us a limited licence to process it solely to deliver the contracted service, as set out in the DPA. We do not train or fine-tune models on your data. The evidence mode named on your order form is your processing instruction.
5. What a bundle is, and is not
A signed assurance bundle is technical evidence supporting your own go/no-go decision and residual-risk acceptance. It is not a certification, a guarantee, a warranty of fitness, or legal advice. Your accountable executive retains full ownership of the deployment decision. Findings marked EXCLUDED_UNSUPPORTED or INDETERMINATE are exactly that. Domains marked PILOT or UNVALIDATED are tested where present but not benchmark-validated, and the bundle says so.
6. Fees
The Baseline Assurance Sprint ($15,000) is payable in full at signature by card or Stripe; kickoff is gated on receipt. The Continuous Assurance Platform ($48,000 per year) is billed annually upfront and recognised ratably. Scoped additions are as published on the pricing page or your order form. Prices exclude tax. Thirty days’ notice of a price change; never retroactive. There are no success fees of any kind.
7. What we ask you not to do
- Grant us access to data you do not have the right to expose to a processor, or that is barred from disclosure.
- Tamper with the runner, the nonce, the suite commitment or the manifest, or present an EAL-1 self-run as independently verified evidence. The runner adversary model exists to make this detectable; doing it anyway ends the engagement.
- Use the service to build a competing verification service, or scrape it.
- Attempt to identify individual reviewers, or contact them outside the platform.
- Probe or attack the service outside the disclosure policy.
8. The qabit adapter and console
While your account is active you may use the MIT-licensed adapter in your own systems under its licence, and the console under these terms. You are responsible for telling your own users that their interactions may be reviewed by people; we give you the wording. Pin an adapter version if you need change control.
9. Recall
If we find a flaw in our own method or scoring, the recall protocol applies: automated hold within 4 hours, written notice within 24 business hours, re-evaluation and RCA within 5 business days at no charge. You agree that a held record means what it says until re-signed.
10. What we do not promise
Verification reduces the chance that a consequential defect reaches production; it does not eliminate it. Sequential testing bounds a failure probability; it does not make the bound acceptable — that judgement is yours. Except where the law says otherwise, the service is provided as is, without implied warranties.
11. Liability
Neither of us is liable for indirect, incidental or consequential loss, or for lost profits or data. Our total liability is capped at the fees you paid us in the 12 months before the claim, with the market carve-outs an enterprise order form will specify (breach of confidentiality, IP infringement, gross negligence, wilful misconduct). Nothing here limits liability that cannot lawfully be limited.
12. Confidentiality
Each of us protects the other’s confidential information with at least reasonable care and uses it only for this relationship. Reviewers are individually bound by confidentiality agreements covering everything they see. Bundles are confidential to you; notary records contain nothing confidential.
13. Publicity
Neither of us names the other publicly without written permission. Design partners agree to be named and to a case study they approve line by line; nothing else is implied.
14. Ending it
Either party may terminate the platform at the end of a subscription year with 30 days’ notice. We may suspend for non-payment or a breach of section 7, with notice and a chance to fix it where fixing it is possible. On termination the exit covenant applies: your suites and bundles are already yours; Mode 2 diagnostics are deleted by cryptographic erasure.
15. Changes
We may update these terms; the date at the top moves and material changes are emailed to account holders at least 30 days ahead.
16. Law
Governed by the laws of the State of Delaware, USA, without regard to conflict-of-laws rules; we each submit to the courts of that state. Where your local consumer law gives you rights that cannot be waived, those still apply.