Cognitive Model Organism: The Unboxing of Thinking Process (Data and Reproduction Workflow)

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This repository contains the revised preprint manuscript (Version 3, revision 1), Supporting Information, and the complete frozen reproduction and analysis archive for "Cognitive Model Organism: The Unboxing of Thinking Process." We establish the Hortus Regis Alti (HRA) Experimental and Observational Protocol for studying a frozen language model (DeepSeek-R1-Distill-Qwen-7B, revision 916b56a44061fd5cd7d6a8fb632557ed4f724f60, INT8 linear layers with BF16 elsewhere) as a cognitive model organism. Across two experimental stages (18 Stage 1 conditions and 26 Stage 2 conditions across four observation configurations of one Bayesian decision problem), the archived corpus comprises 176 complete native-token trajectories containing 2,205,963 generated tokens. Eight paired 16K/32K baseline executions regenerate identical overlapping prefixes across 93,016 native-token positions. Complete mechanical extraction and contextual reading cover all 176 trajectories, with 1,926 location-verified evidence anchors mapped to character offsets and native-token coordinates. Included files:- Preprint_Not_Peer_Reviewed.pdf & Supporting_Information.pdf: Version 3, revision 1 manuscript and supplement (use these top-level PDFs for the latest text and figures; the internal docs/ folder inside the sealed reproduction ZIP retains the original frozen v3 PDFs to preserve archive byte identity).- CMO_Reproduction_v3_HRA_20261030.zip (SHA-256: 9daf88aab04b312a880b50811ea77227c086705763477bce094c3402a2e8b9d9): Outer reproduction distribution with integrity-checked launchers and embedded sealed scientific supplement CMO_FullCorpus_Analysis_Supplement_20261030_r2.zip (SHA-256: fcdf7e11a83e8c088fe23a8cde6fd940c299c0a1f22bbac30d16435c6fd475fd), containing all 176 raw JSONL records, native tokenizer, semantic ledgers, and a zero-dependency Python 3.11+ deterministic CPU recomputation pipeline.