Conceptual framework, experiments, channels, and validated analysis.

(A) TRP-derived functional hypothesis relating the minimum, status-quo, and goal references to four candidate behavioral regions. The proposed correspondence organizes the test; it does not establish HMM modes as literal psychological reference points or direct measurements of subjective feeling. (B) Chronology of the three experiments; Study 1b reused Study 1a rats after approximately five months. (C) Multimodal channels and trial timing. Fecal boli are session-level validation, not an HMM emission. (D) Reconstruction keeps event zeros distinct from unrecorded modalities and culminates in a boundary-aware, theory-guided K=4 diagonal Gaussian HMM. Source data: reconstruction ledgers and analysis_constants.json.

Boundary-aware selection and the validated four-mode architecture.

(A) Subject-grouped five-fold discrete-aware held-out log probability for four diagonal variance floors; points are fold means and error bars are SE. The dashed line is the one-SE cutoff and blue marks the selected 0.01 floor. (B) Fitted raw-scale emission means, colored relative to each channel’s maximum. (C) Decoded occupancy pooled across 47,004 bins and the full experimental schedule; these proportions are descriptive, not normative recovery targets. (D) Fitted transition matrix. State labels follow the joint emission signature. Source data: model_selection_discrete_aware.csv, state_architecture.csv, and fitted_transition_matrix.csv.

Numerical and cross-study stability.

(A) Scores for 88 strictly valid full-data restarts grouped by likelihood basin; the star is selected seed 9 in the dominant basin. (B) Pairwise NMI among valid decoded partitions. (C) Fold-wise NMI after aligning each diagonal-floor partition to the 0.01 reference; horizontal lines indicate medians. (D) Cosine similarity between within-study empirical and pooled fitted standardized signatures. Source data: full_data_valid_restart_scores.csv, restart_NMI_matrix.csv, cross_floor_stability.csv, and cross_study_signature_similarity.csv.

Occupancy, autonomic validation, and exit routing beyond local occupancy.

(A) During conditioning, Exploratory occupancy declines as Fear/Danger rises; the joint trajectory does not separate associative learning from within-session familiarization. (B) Fear/Danger rises transiently and then declines during the first extinction session. Lines are subject-session means and bands are subject-bootstrap 95% intervals. (C) Negative-binomial GEE coefficients rescaled to a 10-percentage-point increase; bars are Mancl-DeRouen bias-reduced 95% confidence intervals. Occupancy coefficients share Quiescent as the omitted reference; mean freezing is from a separate model (17 rats, 130 sessions). (D) Observed-to-expected exit fractions under the ten-stratum local-order null; error bars transform the null 95% interval, and the dashed line denotes equality. Axis abbreviations are Fear = Fear/Danger, Q = Quiescent, and Goal = Goal-directed. Source data: within_session_decile_summary.csv, boli_gee_bias_reduced_sensitivity.csv, and transition_null_stratified_sensitivity.csv.