Reasoning Traces: Representation and Retrieval of Transformational Structure in Decision Processes

Fuente: Zenodo
Salvato in:
Dettagli Bibliografici
Autore principale: Teichner, Steven
Natura: Recurso digital
Pubblicazione: Zenodo 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866901886871797760
author Teichner, Steven
author_facet Teichner, Steven
contents <p>This work introduces <em>reasoning traces</em> as a structured representation of reasoning processes, enabling reasoning to be stored, compared, and retrieved based on transformation structure rather than textual similarity.</p> <p>A reasoning trace models reasoning as a transition between an initial state and a resulting state through an explicit transformation descriptor, supported by discrete reasoning units and relational links. This representation abstracts away from surface language and captures the functional structure of reasoning, allowing semantically distinct inputs to be mapped to shared transformation patterns.</p> <p>The paper presents a minimal, reproducible protocol in which a standard language model is used as an extraction mechanism to segment reasoning, identify transformations, and construct structured reasoning trace objects. The purpose of this protocol is to demonstrate that unstructured reasoning can be systematically converted into a structured, transformation-based representation using widely available interfaces, without specialized infrastructure.</p> <p>Across examples drawn from legal, business, and policy contexts, the demonstration shows that reasoning processes can be normalized into reusable transformation patterns and retrieved based on how reasoning operates rather than how it is expressed. This suggests a shift from text-centric retrieval to reasoning-centric retrieval, in which queries operate over transformations rather than surface representations.</p> <p>This work is intentionally scoped as a representation-level contribution. It establishes the plausibility of modeling reasoning as structured transformations and demonstrates that such representations support cross-domain comparison and retrieval in principle. Questions of large-scale validation, consistency, and system-level implementation are identified as directions for future work.</p> <p>The approach provides a foundation for auditable, explainable, and reusable reasoning systems, and may be extended through formal schemas, validation mechanisms, and graph-based storage for large-scale reasoning analysis.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19591087
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Reasoning Traces: Representation and Retrieval of Transformational Structure in Decision Processes
Teichner, Steven
Auditable AI
Explainable AI
AI reasoning
Knowledge infrastructure
Reproducibility
Traceability
Schema-driven systems
Human-AI collaboration
Decision support systems
AI governance
<p>This work introduces <em>reasoning traces</em> as a structured representation of reasoning processes, enabling reasoning to be stored, compared, and retrieved based on transformation structure rather than textual similarity.</p> <p>A reasoning trace models reasoning as a transition between an initial state and a resulting state through an explicit transformation descriptor, supported by discrete reasoning units and relational links. This representation abstracts away from surface language and captures the functional structure of reasoning, allowing semantically distinct inputs to be mapped to shared transformation patterns.</p> <p>The paper presents a minimal, reproducible protocol in which a standard language model is used as an extraction mechanism to segment reasoning, identify transformations, and construct structured reasoning trace objects. The purpose of this protocol is to demonstrate that unstructured reasoning can be systematically converted into a structured, transformation-based representation using widely available interfaces, without specialized infrastructure.</p> <p>Across examples drawn from legal, business, and policy contexts, the demonstration shows that reasoning processes can be normalized into reusable transformation patterns and retrieved based on how reasoning operates rather than how it is expressed. This suggests a shift from text-centric retrieval to reasoning-centric retrieval, in which queries operate over transformations rather than surface representations.</p> <p>This work is intentionally scoped as a representation-level contribution. It establishes the plausibility of modeling reasoning as structured transformations and demonstrates that such representations support cross-domain comparison and retrieval in principle. Questions of large-scale validation, consistency, and system-level implementation are identified as directions for future work.</p> <p>The approach provides a foundation for auditable, explainable, and reusable reasoning systems, and may be extended through formal schemas, validation mechanisms, and graph-based storage for large-scale reasoning analysis.</p>
title Reasoning Traces: Representation and Retrieval of Transformational Structure in Decision Processes
topic Auditable AI
Explainable AI
AI reasoning
Knowledge infrastructure
Reproducibility
Traceability
Schema-driven systems
Human-AI collaboration
Decision support systems
AI governance
url https://doi.org/10.5281/zenodo.19591087