Pricing that follows your cohort.
Start with a fixed-price pilot on your own scripts. Scale per script. Go sovereign when the data must never leave.
Pilot
Prove it on your own scripts
- ✓One unit, calibrated on dozens of your human-marked scripts
- ✓Full pipeline run on a real cohort
- ✓Side-by-side calibration report
- ✓Fixed scope, fixed price
Centre
Most centresMarking as a service, per script
- ✓Per-script pricing across your live units
- ✓Official marksheets, referrals & resubmissions included
- ✓Platform access: queue, reviews, reports, records
- ✓Assessor-in-the-loop modes at no extra step
- ✓New brief years as configuration
Sovereign
Your infrastructure, your boundary
- ✓Fully on-premise or private in-region cloud (UK · EU · Gulf)
- ✓Everything in Centre, deployed under your control
- ✓Custom units, gates and integrations
- ✓Named engineering support
Every plan includes: deterministic marking with written reasoning · verbatim evidence on every decision · the official marksheet template · the complete audit record · no learner data to third-party AI services.
Fair questions.
Why don't you list prices?+
Because the honest answer depends on script volume, unit count and deployment model. A pilot is fixed-price and quoted upfront; Centre pricing is per script and falls with volume; Sovereign is scoped per deployment. You'll have numbers after one call.
What does onboarding a unit involve?+
You bring the brief, the rubric and a calibration set of human-marked scripts — measured in dozens, not thousands. The unit is authored as configuration and gated by calibration before it marks anything live.
Do we still need assessors?+
Yes — by design. Sokros does the reading, checking and drafting; your qualified assessors sample, review and approve at the points you choose. Integrity outputs are always signals for a human, never automatic penalties.
What happens to our data?+
Learner data stays inside the deployment boundary. No third-party model APIs sit in the marking path, and retention and deletion follow your instruction.
Can a new brief year break last year's marking?+
No. Each brief year is a separate configuration version; both stay runnable and auditable. Past behaviour is never silently altered — releases are replay-tested byte for byte.