MORPHZ / PREPRINT / 2026

Nondeterministic CognitiveSymbol Evaluation overStructured Context

The paper defines the Morphz computational model, its implementation boundary, and layered experiments: an LLM evaluates cognitive symbols over agent-owned Structured Context while a deterministic runtime retains authority over capabilities, versions, real observations, and committed state.

Computational model
Can an LLM be organized as a nondeterministic cognitive-symbol evaluator over durable Structured Context while real-world effects and authoritative state remain under runtime control?

Main contributions

01

Computational identity

The model is not merely a message completer. It is a nondeterministic cognitive evaluator constrained by input structure, operator semantics, and output contracts.

02

Result re-entry

Programs, cognitive objects, and evaluation results share a recursive representation, allowing results to enter bindings or become durable objects referenced by later work.

03

Authoritative commit

Candidate cognition is separated from reality. The runtime owns validation, capabilities, versions, transactions, scheduling, and recoverable execution.

Experiments and results

Layered experiments cover mechanism executability, cross-model applicability, experience transfer, and system-level external validity on complex terminal tasks. Protocols, run data, and failure samples are published with the research materials.

Read the technical introduction to the computational model