What is live
No figures that go stale. The capability catalogue lives elsewhere; this is the maturity an investor needs so as not to buy a shadow.
| Layer | Status |
|---|---|
| On-prem product: install, institute, commission, deliver with a seal | [V] in operation, closed beta |
| Persistent agent identity, private corpus, Task Queue, bilingual Console | [V] |
| Authentication (OIDC) and roles on the backend, not just on buttons | [V] |
| Governance as a tray: approve means apply | [V] built; generic enforcement is off by default [P] |
| Selection ladder (fitness): lower and middle rungs | [V] measured; routing rungs in shadow [P]; recursive self-improvement not started [D] |
| Ontology catalog (RDF / JSON-LD, closed vocabulary, SHACL) | [V] deployed as a surface distinct from the live graph; [P] no in-catalog tenancy, no automatic sync |
| Host federation (audited envelopes, trust policies) | [V] transport; [P] envelopes are not cryptographically signed |
| SaaS multi-tenancy | [D] last by decision |
Self-improving, precisely. Variation (the agent changes skills and memory) already exists in the runtime. Measured selection is the ladder: today it recommends; it does not act on live routing. Saying “the system improves itself” skips a rung that is not there.
Two failure modes already found
- A governance layer that records everything and blocks nothing is theatre. The commitment is that gates which must bite are tested biting, with a negative control.
- Selling a conformity that does not exist. Naming a log after a chained audit engine, without chaining it, would lie in the most expensive direction. Gaps stay gaps in public until they are green in the code.