Measurable human oversight. Priced for the buyers who need it.
Three tiers for patent-pending oversight infrastructure that helps teams evidence human review for NIST AI RMF, Colorado SB 26-189, SEC AI guidance, and EU AI Act Article 14 conversations. The score is identical across every tier; verification rigor, decision volume, and procurement-readiness differ.
- DRS plus JIS scored on every decision
- Signed Methodology Annex (PDF)
- HMAC-chained audit log export
- Self-Reported assurance level
- Email support during business hours
- Production status monitoring
- Everything in Starter
- Audited assurance level
- Quarterly Methodology Annex refresh
- Big-4 compatible artifact bundle
- Slack plus email priority support
- Negotiated uptime and support terms
- Everything in Audited
- Custom cognition probes and escalation policies
- Named-auditor coordination
- Custom MSA with negotiated SLA and indemnity caps
- Dedicated technical contact plus monthly review call
- Data processing agreement and residency review
Shape the evidence layer before your market asks for it.
$10,000 one-time. 30 days. One AI workflow. 10 seats. Closes 2026-06-05. The pilot produces an oversight evidence packet and a buyer-facing summary. Starter is a separate post-pilot purchase, with no Year-2 lock. The cohort is founder-reviewed: a brief fit call confirms cohort fit before the checkout link is shared.
What each plan includes.
| Capability | Starter | Audited | Enterprise |
|---|---|---|---|
| Assurance level | Self-Reported | Audited | Audited |
| Decisions per month | 100,000 | 500,000 | Custom |
| Domains | 1 | 3 | All 6 |
| Methodology Annex refresh | At signing | Quarterly | Quarterly |
| Deployment | Hosted | Hosted | Self hosted |
| Custom MSA and SLA | No | Negotiated | Yes |
Frequently asked.
Is this a consulting engagement or a product?
A product. The first five pilots receive in-person founder installation; after that, organizations connect a surface and the scoring runs autonomously.
What is the difference between Self-Reported and Audited?
The score is the same. The verification rigor is different. Self-Reported is the buyer signing the data declaration. Audited is a Big-4 firm or accreditation body co-signing the data feed and the assurance. Same FICO analogy: a self-pulled score versus a lender-pulled report.
Can operators see that they are being measured?
Yes. The Specification section 7.5 requires disclosure at onboarding. Operators know maintenance interactions exist; they cannot identify which specific interactions are maintenance events. This satisfies regulatory transparency without compromising the measurement.
Does COHESION handle the data, or does it stay in our system?
Behavioral telemetry flows to the COHESION scoring API for computation. Records are stored under your organization account in Cloudflare D1, retained for 24 months, encrypted at rest, GDPR compliant. You retain ownership.
What is the Founding Design Partner cohort?
$10,000 one-time. 30 days. One AI workflow. 10 seats. Closes 2026-06-05. The pilot produces an oversight evidence packet and buyer-facing summary. Starter is a separate post-pilot purchase, with no Year-2 lock. The cohort is fit-call gated: see the Design Partner page.