Inputs
Context enters explicitly
Problems, constraints, model capabilities, roles, sources, and criteria are declared rather than assumed.
System architecture
Six layers. One inspectable process.
Multi-model connectivity is the foundation, not the final product. Each layer answers a distinct question and passes explicit state forward.
System map
The six-layer stack
L5
Deliberium WorkspaceMake the full process visible and steerable to humans.Input: Decision records, graphs, timelines · Output: Inspectable workspace state
L4
Decision IntelligenceCompare alternatives without hiding uncertainty.Input: Evidence, criteria, options · Output: Recommendation + provenance
L3
Evidence IntelligenceConnect claims to sources and counter-evidence.Input: Claims, sources, assumptions · Output: Confidence + uncertainty
L2
Deliberation EngineSupply roles, challenge, critique, and revision.Input: Participants, objectives, rubrics · Output: Revised positions + minority reports
L1
Thinking RuntimeControl progression through reasoning stages.Input: Problem, constraints, stage policy · Output: Stateful deliberation state
L0
Model FabricSelect participating intelligence.Input: Providers, models, capabilities · Output: Independent contributions
Layers are contracts
Progression
State moves deliberately
The runtime records where the process is, what has been challenged, and when a position changes.
Outputs
Decisions retain their trail
Recommendations connect to evidence, objections, unresolved questions, and provenance.