Governed projections. Not free-text answers.
Proteus turns your live evidence into typed objects — runbooks, incidents, support resolutions — each carrying its own provenance, confidence, gaps, and policy. Every field is traceable to the sources that produced it. Every action is gated by what the evidence actually supports.
Answers without evidence are a liability. Proteus objects ship with the signal you need to decide whether — and how — to act on them.
Each value is attached to the knowledge entries, retrieval hits, or Hermes observations that produced it. No floating claims. No "because the model said so."
Per-field confidence with named, documented bands — speculative, provisional, supported, corroborated, authoritative — and a calibration log so future labels can tune them.
Schema-required fields without corroborating sources become tracked gaps, not silently defaulted values. You see what the projection doesn't know, with suggested next actions.
Every object carries a policy decision: is this safe to display, act on, or automate? Synthetic evidence downgrades actions. Human review is requested when it should be.
Five stages, each leaving artifacts you can inspect. The viewer shows you every one.
Proteus is infrastructure. Your agents, dashboards, and automations read from it — and every read has receipts.
The viewer accepts a query, gathers evidence from your running Solstice stack, and returns a fully-annotated projection — data, confidence, provenance, gaps, policy, lineage — all in one screen.