Think of ESP360° as a decision-context layer beside telemetry.

Six architectural domains

Every record represents one decision moment decomposed into six orthogonal but interacting domains. This is the level at which the architecture can be evaluated without exposing its implementation.

1 · Context

External environment: weather, social visibility, financial stress, access barriers, technology failure, institutional trust. Context modulates how every other domain interacts. The same glucose reading in a private home and in a public meeting carries different decision weight.

2 · Cognitive / analytical

Active reasoning state: known facts, inferences, gaps, assumptions, competing interpretations, cognitive load. Expert decision-making involves holding several valid interpretations at once. Without this layer a decision record has no reasoning substrate.

3 · Layer activation

Which reasoning layers are active, dormant or in conflict. Layer conflicts — physiological urgency pulling against emotional state — produce the most clinically significant deviations.

4 · Confidence / uncertainty

Certainty across several dimensions: statistical, interpretive, ethical, experiential. Uncertainty is information. Systems that collapse it to zero produce confidently wrong outputs.

5 · Authority

Who holds the decision right, what override authority exists, what escalation is available. Human authority is preserved at every decision point; the architecture makes no autonomous recommendation.

6 · Temporal / consequence

Circadian phase, elapsed time, cumulative burden. States propagate forward — this morning's sleep debt changes this afternoon's confidence. Temporal carry-forward is what distinguishes chronic disease management from episodic care.

A working computational prototype

The current implementation is not a diagram-only concept. It is a functioning research prototype with hundreds of annotated records, hundreds of structured fields per record, tens of thousands of synchronized CGM observations and a large formula-derived inference surface. The engineering significance is that decision context has been made computable rather than left as free-text commentary.

The public site intentionally describes the architecture by function rather than by reconstructive recipe. At a high level the system separates source observations from derived and interpretive state, synchronizes those states to the moment of action, preserves temporal carry-forward and contradiction, compares trajectories, classifies deviation before attribution, and retains human review and provenance.

Conceptual data objects

A production implementation could separate patient state, event state, glucose state, decision state, narrative state, temporal carry-forward, contradiction records and provenance records. Database technology is an implementation choice, not the architecture itself.

Two properties are load-bearing and easy to lose in a naive schema. Contradiction preservation: when a person holds two incompatible beliefs at the same moment, both are stored, because the conflict is the signal. Temporal carry-forward: state does not reset at a record boundary, and it does not reset at midnight either — which is the same structural error abstract #47 measures in CGM segmentation.

Why provenance rather than inference

Inference answers "what probably explains this action?" Provenance answers "what was actually documented as known, believed, feared, uncertain or intended at the time?" Those are different machine-learning targets, and a model trained on the first cannot be audited against the second.

Where it sits

Beside telemetry, not inside it. Device data stays authoritative for what happened; the decision layer wraps it with what was being reasoned when it happened. Nothing in the design requires modifying the device pipeline, which is deliberate — an architecture that demands changes to the sensor stack will not be adopted.