CIRRA · Architecture
The scientific architecture.
CIRRA's architecture is organised around a single principle: biological context determines the question, evidence shapes belief, and experimental reality determines what the system should learn next.
Architecture concepts are research directions. Current capability is documented separately.
Illustration — A lattice of linked claims, mechanisms and evidence — illustrative of the CIRRA architecture.
Reasoning loop
Context → hypothesise → arbitrate → rank → validate → learn.
The canonical reasoning loop that organises every component of the architecture.
- Context
- Hypothesise
- Arbitrate
- Rank
- Validate
- Learn
Architecture concepts
Components under development.
These concepts organise the research architecture. They are not validated production capabilities.
ClaimBundle
A structured representation of a scientific claim, its evidence, and its competing interpretations.
CIRRA Lattice
A graph of related claims, mechanisms, and evidence that supports traversal and arbitration.
Context Fabric
The contextual layer that conditions how evidence is interpreted — disease, cohort, modality, prior belief.
Knowledge Node Mesh
A persistent substrate of validated knowledge that compounds across engagements.
Scientific Arbiter
The component that weighs competing mechanisms and produces a calibrated belief with uncertainty.
Deliberative Mechanism Search
The process of generating and evaluating candidate mechanisms against available evidence.
CIRRA-RANK
The current explainable ranking system with provenance and uncertainty for circRNA candidates.
Active Validation
The process of selecting experiments that would most reduce uncertainty in the current belief state.
Memory Hierarchy
Layered memory across engagements, cohorts, and the institutional knowledge base.
Evolution Plane
The substrate on which the system updates its beliefs as experimental reality arrives.
Learning Stack
The accumulation of validated learning trajectories across the institution.
Integrity Membrane
The governance layer that ensures provenance, auditability, and human review across every claim.
Evidence states
How CIRRA represents what it knows.
CIRRA distinguishes observed from inferred. It is allowed to abstain.
This is part of the architecture, not a marketing claim — every output carries its evidence state, and uncertainty is visible by default.
Observed
Directly supported by source data or experiment.
Inferred
Model or system inference from available evidence.
Supported
Evidence aligns with the hypothesis.
Contradicted
Evidence challenges the hypothesis.
Tested negative
Experiment did not support the expected outcome.
Unknown
No appropriate validation available.
Abstain
The system should not make a confident judgement.
Collaborate on CIRRA research.
We work with selected research groups on circRNA discovery and evidence integration.