Every AI-native product eventually has the same conversation: what does a single unit of work actually cost? For Assertive, that unit is one governed scenario run. Here's the honest breakdown.
Three cost centres
A run consumes model tokens (planning, grounding, healing), browser minutes (headless Chromium in an isolated sandbox), and storage (traces, videos, screenshots retained for the report window). Everything else — dispatch, HMAC verification, changelog — is negligible at scale.
The healing tax
Self-healing selectors add cost only when they fire. A stable run costs a fraction of a cent in tokens. A run that re-grounds three selectors costs measurably more — and saves a human 15 minutes of triage. We surface both numbers in the unit-economics dashboard.
How we price
Tiers include a monthly run budget with transparent overage. Enterprise customers see a per-run p50/p95 cost line in the admin console — no black boxes. If your margins depend on QA being cheap, ours depend on being honest about the same numbers.