1. Introduction
Health systems routinely collapse reasoning into state.
Interpretations are written as facts. Conclusions overwrite observations. Outputs of reasoning systems are treated as authoritative records rather than contingent results.
This collapse destroys accountability.
Reasoning is not state.
Reasoning is a consumer of state.
2. Reasoning Defined
Reasoning is the process by which meaning, inference, or decision is derived from canonical state.
Reasoning may be:
- Human
- Algorithmic
- Statistical
- Rule-based
- Learned
The method is irrelevant to the boundary.
Reasoning consumes canonical state.
Reasoning does not produce canonical state.
3. Reasoning Does Not Assert Truth
Canonical truth is established by valid assertion under authority.
Reasoning proposes conclusions. It does not assert facts.
A conclusion may be useful, persuasive, or correct. It remains an interpretation unless explicitly asserted under authority as new canonical state.
Systems that treat reasoning outputs as truth collapse interpretation into substrate.
4. Derived Outputs
Outputs of reasoning include:
- Assessments
- Classifications
- Scores
- Predictions
- Recommendations
- Decisions
These outputs are derived state.
Derived outputs:
- Are contextual
- Are revisable
- May expire
- May conflict
- May be wrong
They may not modify canonical state.
5. Human and Machine Parity
Human reasoning and machine reasoning are structurally equivalent at the substrate boundary.
Both:
- Consume canonical state
- Operate under assumptions
- Produce derived outputs
- Are subject to error
No reasoning process acquires canonical privilege by virtue of being human, institutional, or automated.
6. Feedback Without Rewrite
Reasoning may inform future assertions.
When new information arises, it is asserted as new canonical state under authority. The reasoning that motivated the assertion remains derived and inspectable.
Feedback that rewrites prior state collapses causality.
7. Safe Placement of Intelligence
Reasoning belongs downstream of canonical state.
Placing intelligence upstream introduces silent mutation, feedback loops, and self-reinforcing error.
Placing intelligence downstream preserves:
- Auditability
- Disagreement
- Correction
- Longitudinal integrity
Intelligence must be powerful and constrained.
8. Invariants
The following invariants hold:
- Reasoning consumes canonical state
- Reasoning outputs are derived
- Reasoning does not assert truth
- Reasoning cannot rewrite canonical state
- Human and machine reasoning hold no canonical privilege
- Derived outputs are contextual, revisable, and may be wrong
- Feedback from reasoning does not rewrite prior state
Violating any invariant collapses accountability.
9. Discussion
Reasoning is necessary and fallible.
Treating reasoning as substrate produces systems that cannot explain themselves, correct themselves, or be trusted under scale.
By confining reasoning to derived consumption, HealthMemory preserves truth while enabling unlimited intelligence downstream.