TRUST

Receipts, not vibes.

Three things buyers should never have to take on faith: who owns the data, how secure the system is, and whether the accuracy numbers are real. Here they are in writing.

DATA RIGHTS — THE WHOLE POLICY

01  You own all derived data: transcripts, detection logs, classifications, delivery records, exports.
02  Everything exports in open formats, on demand, at any point in the relationship.
03  We never train models across customers on your data without your written opt-in.
04  On termination, your data is returned and deleted, with written certification.
05  This policy is incorporated into every contract — it is not marketing.

SECURITY POSTURE

Where we are, stated plainly

Security practices and runbooks are documented and available under NDA today. SOC 2 Type I observation is on the published roadmap with a target date, and this page updates as status changes — we'd rather show a date than imply a certificate.

// status board
security practices doc  AVAILABLE UNDER NDA
role-based access + audit logs  SHIPPED
soc 2 type i  IN PROGRESS — TARGET PUBLISHED IN CONTRACT PACK
pen test  SCHEDULED WITH TYPE I

HOW WE MEASURE

Every accuracy number has a methodology — or it doesn't get stated.

Hand-labeled ground truth

Classifier accuracy is validated against a human-labeled clip set — hundreds of clips, labeled before the model saw them — not against the model's own confidence.

Per-customer validation

Your accuracy is measured on your broadcasts during the pilot, against hand-checked games. You see your number before you sign an annual contract.

Human review where it counts

Invoice-grade reporting adds a review queue: every machine-flagged element passes a human before it's marked delivered. Auditable line by line, by design.

Ask the uncomfortable questions.

Bring your security questionnaire to the call. We'll answer it line by line.

Book a demo →