Xchool
The token is a row of evidence.
Education still invoices the room. This invoices the row. One accepted record that a named person can do a specific thing. Eval, harness, proof.
01 Eval
Specs
Closed SKUs. What Done is. What artifact counts. Who may attest. The price.
02 Harness
This org
Their floor, language, time, templates, constraints. Forward deployed. Not a course.
03 Evidence
UGC
What a human produced to show they can. Not AI. Accepted, it bills. Failed attempts are free.
Who pays is not who sits in the room.
The unit only works if these four seats stay distinct. VCs watch. They do not buy. The budget is already inside the org.
Pays
Enablement, Academy, Transformation. A P&L sponsor.
Project budget, not the content catalog. Measured on ramp, error, Copilot with no ROI. Training Magazine: 21% have final purchase. Sell them, not the recommenders.
Uses
Customiser + learner
The customiser locks twelve SKUs on their files. The learner submits UGC from live work. Neither writes the cheque.
Attests
Line manager, QA, examiner
Already signs off informally. Now the signature is a metered record. Not L&D. Not the model. Not HRBP.
Learns
A named person in a named role
AE, RM, claims, ops lead. Rejected submissions do not bill. Same idea as Fin: you do not pay for the miss.
Not a client list. Stripe, Palantir, Intercom are the unit analog (meters, forward deployed, pay per accepted outcome). AXA is the trigger: Copilot plus an upskilling program, July 2026. Academies are where the customiser already sits. 89% of US orgs already have an LMS. LMS purchase intent is 14%. Do not sell a library. US training $102.8B. External products $16B. $874 per learner, hours down to 40. OECD: practice with GPT looked 48% to 127% better, then 17% worse with the tool off. Anti: LMS replacement, a 5,000-seat literacy course, SCORM completions, people impacted.