History-structured forecasting of rewarded give-up behavior in a rodent metacognition task: data and code

This record provides the de-identified trial-level data, model outputs, audit documentation, and executable code for History-structured forecasting of rewarded give-up behavior in a rodent metacognition task. The central result concerns predictive information carried by subject-specific sequential behavioral history beyond current-trial context. The pinned code is the source of record for the primary M5+Changepoint benchmark and the structuring-cause validation. It also includes the all-23 robustness inputs, trivial behavioral baselines, a fail-closed verifier, and an explicit account of analyses retained as archival or exploratory. Bin Yin conceived history-structured forecasting and its application to these animal-behavior records. Human researchers designed and conducted the experiment, trained and tested the animals, collected the primary data, and maintained the original records before the AI-assisted computational phase. Post-collection AI-assisted tools were used under author direction for bounded implementation, reconciliation, checking, and language refinement. Tool outputs were treated as unverified until checked against source records, executable code, manifests, and independent reruns. Quantitative figures and tables were rendered from validated data by transparent scripts, not directly by AI systems. The apparatus image is an original experimental photograph, and Figure 1 is a scripted schematic grounded in documented hardware and procedures; neither is synthetic imagery. The human authors reviewed and approved all outputs and accept responsibility for the work. Data are licensed CC BY 4.0. Code is licensed BSD 3-Clause. The concept DOI is 10.5281/zenodo.21228501.