Provider-neutral contracts
Identity, decisions and scene contracts are provider-neutral, so continuity survives provider changes.
How it works
A request becomes responsible work when identity, context, evidence, authority and action stay connected.
Humans+ turns interaction into a governed continuity loop. It does not treat each response as an isolated generation.
The model
Five grouped stages, one continuous loop
Understand
What happens: the interaction is received and bound to identity, scope, consent and authority. Why it matters: context is not re-explained. What it will not do: it will not act without established authority or consent.
Orient
What happens: the Digital Partner's identity and current AMS state are loaded. Why it matters: responses are situated, not cold. What it will not do: it will not treat internal state as external truth or emotion.
Ground
What happens: CKPE retrieves authorised evidence and memory; chronology and contradictions are accounted for; scene readiness is assessed. Why it matters: reasoning is grounded, not guessed. What it will not do: it will not treat a negative retrieval as proof beyond its certified scope.
Reason
What happens: a bounded scene is composed and routed to an approved model for reasoning. Why it matters: a scene is not a prompt—it is a governed contextual unit. What it will not do: it will not let a model vendor become the Partner's identity.
Deliver & learn
What happens: output or action is validated against policy and quality gates, delivered, then reflected on to create reviewable memory candidates. Why it matters: growth is governed and attributable. What it will not do: it will not silently drift or act without approval.
The complete continuity loop
Sixteen logical stages, one governed system
These are logical boundaries, not necessarily sequential model calls. Deterministic rules, small models and parallel preparation are preferred where appropriate.
Receive interaction
From web, voice, messaging, email, API or another authorised channel.
Establish identity, tenant, authority, consent, disclosure
Boundaries are set before anything else happens.
Resolve service, task, participants and Partner
Who and what is responsible for this work.
Load enduring identity and AMS state
Continuity and current internal state.
Retrieve authorised evidence via CKPE
Account for chronology and contradictions.
Determine scene readiness
Clarify or escalate if the context is not good enough.
Compose a bounded scene
Render it for an approved model.
Route reasoning to an approved substrate
Form a response, plan or decision package.
Update shared understanding
Without inventing agreement.
Validate against policy and quality
Authority, evidence and quality controls.
Deliver or execute an authorised action
Only authorised actions execute.
Reflect and create memory candidates
Update internal state and reviewable memory.
Model independence
The model is a substrate, not the identity
Model selection is governed by capability, sensitivity, latency, cost, availability, quality and behavioural compatibility. The Partner's identity and decision record stay independent of the model used for any step.
Governed routing
Approved models are routed by capability, policy, risk, cost and availability.
Promotion & rollback
Behavioural compatibility, identity stability and rollback readiness are checked before a model is promoted.
Microcopy: "The Partner's identity and decision record remain independent of the model used for this step."
See how continuity becomes accountable work.
Explore trust by design, or talk to the team about your organisation.