Reference architecture
Intelligence that can explain itself, respect authority, and learn under control.
Tangerine separates evidence, reasoning, authorization, and execution—then reconnects them through a closed decision loop.
Layered by responsibility
Six architectural layers. One accountable flow.
Decision surfaces, workflows, collaboration, and role-specific views.
Permissions, thresholds, approvals, constraints, and escalation.
Structured analysis, scenario comparison, confidence, and abstention.
Objects, relationships, provenance, validity, and decision history.
APIs, streams, connectors, documents, models, and enterprise services.
Identity, security, deployment boundaries, observability, and lifecycle.
Canonical loop
Decisions begin in reality and end in evidence.
A decision is not complete when an answer appears. It is complete when the action, authority, and observed outcome can be connected back to the original state.
Authority architecture
Capability is not permission.
Tangerine distinguishes what a person, workflow, or reasoning service can determine from what each is authorized to approve or execute.
Two loops
Run the business. Improve the intelligence. Keep the boundary visible.
Runtime loop
Decide inside the moment
Bind current state, evidence, constraints, authority, and action into a governed operating cycle.
Learning loop
Improve outside the moment
Review outcomes, assumptions, policies, models, and operating design before controlled release.
Delegation contract
Every delegated action has a defined envelope.
The contract states the objective, permitted actions, prohibited actions, evidence required, confidence threshold, escalation path, timeout, and recovery behavior.
Scope
Which objects, systems, time horizon, and consequences are in bounds?
Authority
What may be recommended, prepared, approved, or executed—and by whom?
Failure behavior
When must the service abstain, contain, escalate, expire, or recover?
Replay and observability
Reconstruct the decision—not just the system log.
A meaningful replay includes observed state, evidence versions, assumptions, policies, model and workflow versions, authority checks, human interventions, action, and outcome.
Explore trust architecture
Start with consequence
Architect the decision before automating the action.
Choose a decision where fragmented evidence, unclear authority, or slow coordination materially affects the outcome.
Start the conversation