Boilerplate
EvalQA provides independent technical verification, validation and continuous assurance for enterprise analytics agents that write SQL against cloud data warehouses — Snowflake Cortex Agents, Databricks Genie and custom Text-to-SQL pipelines. It compiles deterministic relational invariants from a customer’s own semantic layer, runs adversarial challenge suites inside the customer’s perimeter, and produces a signed assurance bundle that a third party can verify. EvalQA issues technical evidence, not statutory certification.
The position, in one paragraph
Every platform in this category sells tooling you run on yourself. IEEE 1012-2024 defines independent verification along technical, managerial and financial axes, and self-assessment fails all three — a vendor paid on compute consumption cannot give your audit committee independent assurance of its own model. EvalQA is the second party. The differentiator is not detection capability; it is structural independence plus evidence a third party can check without trusting either side.
Claims we will not make
- That we certify anything. We do not, and we publish why there is no badge.
- That we have customers we do not have. There are none yet and the page says so.
- Any statistic without an n, an interval and a date.
- Urgency based on the lapsed August 2026 EU AI Act date. Annex III high-risk deferred to 2 December 2027; Article 50 transparency remains live. claim: external verification required
Things worth asking us about
- Why almost nobody in this category publishes judge–human agreement statistics.
- Why we published nine ways our own method fails.
- The kill rule: under 20% incremental finding yield across five customers and we say the product should not exist. Detail.
- What OpenAI closing its Evals platform on 30 November 2026 means for eval portability. Our position.
Assets
A logo kit, wordmark and headshots do not currently exist as published assets. planning Rather than link a file that is not there, we would ask you to email for what you need until this section is real.
Contact
[email protected] with subject [PRESS]. We will tell you when something is a plan rather than a shipped capability, without being asked.