How do we know it works?
Last night golden checks passed (95%) against a demo company whose true numbers are known.
What runs every night
Golden checks
A fictitious company with a known story — founding, a shutdown, a relaunch — is loaded like any customer. Chat, joins, forecasts, drivers, scenarios, documents and watchers must recover the true numbers. A drop of ten points alerts the team.
An AI judge
A sample of real answers is scored for groundedness, helpfulness and safety.
Receipts
Every model call is logged with its feature, model, tokens and estimated cost — shown to you under Usage, per month.
Questions
What happens when a nightly check fails?
The suite is scored against the company’s known answers, and a drop of ten points alerts the team.
Can I see what the AI costs me?
Yes. Every model call is logged with its feature, model, tokens and estimated cost, and the month’s total is under Usage in your workspace.