Skip to main content

Capability Graph

The Capability Graph answers the second question: what can this identity actually do? Not what its résumé claims, not what its title implies — what it has demonstrably done, backed by evidence, attached to its Passport.

The model

Field What it is
Capability node A specific, named capability in a domain (invoice.review, contract.liability-assessment)
Evidence pointers Verifiable references to demonstrated work — completed tasks, reviewed outputs, attestations
Confidence Explicit uncertainty, not a hidden score — the graph says how sure it is and why
Freshness Capabilities decay; the graph tracks when evidence was last produced
Visibility policy The holder controls who sees what — capability is opt-in and consent-governed

Not a leaderboard

The graph is designed for revelation, not exclusion. There is no global ranking, no opaque score, no surveillance dataset dressed up as skills tracking. Evidence is cryptographically verifiable, uncertainty is explicit, and the holder owns the record. When routing needs to know "who is qualified to review this contract?", the graph answers with evidence — it does not sort people into winners and losers.

Why routing depends on it

HumanOS routes work on capability, not job title. A task declares what it requires; the graph identifies which identities — human or agent — meet the bar. Cost is considered only among the qualified. This is the difference between "send it to whoever is free" and "send it to someone who can actually do it."

The same graph serves both directions:

  • Production work generates capability evidence as a side effect — every governed task completed under HumanOS strengthens the record.
  • Training through Academy feeds the identical graph. A capability earned in training is the same object production routing reads. No parallel résumé.

Work with the graph

Next in the ladder: Delegations →