Representation Dynamics - Minimal Research Release

This record provides a minimal, curated research package for the Representation Dynamics research programme, documenting the development and empirical evaluation of representation-based approaches to semantic distinction in natural language processing.

The package brings together selected reproducibility materials from REP-DYN-P8, REP-DYN-P23, and REP-DYN-B, together with the accompanying research article Representation Dynamics and Semantic Distinction. It is intentionally limited to materials necessary to document the research trajectory and its empirical development in natural language processing.

The materials include the frozen P8 multilingual evidence set and associated representation-based measurements, the final Paper 1 closure package with frozen N=540 results, and the public core of Benchmark A. The research uses sentence-transformer representations and quantitative measures of representational separation to investigate semantic distinction across linguistic data.

The accompanying materials document methodology, frozen results, model specifications, provenance, licensing, and reproducibility boundaries. Private research-workspace contents, raw model responses, case-level data, prompts, adjudication materials, credentials, internal reports, logs, backups, and other privileged materials are deliberately excluded from this release.

The package is released under CC BY-NC 4.0 for non-commercial use. Third-party datasets and resources remain subject to their original licenses. The record is intended as a controlled research archive documenting the development and empirical grounding of the Representation Dynamics programme.