{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"invariant_mode_decomposition_design__psychology","arm":"RETRIEVAL_FIRST","candidate_id":"C_H2_mode_adaptive_exposure_pacing_v0","hypothesis_id":"H2","version":0,"title":"Mode-adaptive exposure pacing","problem":"Exposure hierarchies may advance or repeat trials using mean fear reduction even when fear, approach, avoidance, arousal, and safety behavior jointly exhibit growing, oscillatory, or marginal dynamics. The scalar rule can therefore escalate difficulty while avoidance-related dynamics are unstable or repeat an exposure merely because harmless arousal persists.","actors":["Patient receiving exposure therapy for an anxiety disorder","Treating clinician who selects exposure difficulty and retains treatment authority","Clinical researcher or analyst who estimates the trial-transition model and produces pacing recommendations"],"observable_state":"Between completed exposure trials, record a prespecified patient-level state vector comprising fear, approach behavior, avoidance behavior, arousal, and safety behavior, together with the difficulty delivered. Across trials within the tested range, estimate the transition of that vector and observe dominant mode shapes, their gains and stability classes, their sensitivity to difficulty changes, reconstruction residuals, and whether the mode estimates remain identifiable and stable.","consequence":"Fear-only pacing can produce premature escalation when fear falls while an avoidance or safety-behavior mode remains growing or weakly damped, or unnecessary repetition when persistent arousal belongs to a stable mode that is not sustaining avoidance. The proposed clinical benefit is fewer such pacing errors; the prespecified comparative claim is at least 20% greater reduction in safety behavior across four sessions than fear-score pacing without increasing dropout by more than 5 percentage points.","affected_objective":"Select the next exposure difficulty so that avoidance and safety behavior decline without unnecessary repetitions, premature escalation, or an unacceptable increase in dropout.","intervention":"At the boundary between one completed exposure trial and selection of the next, estimate a local multivariate transition operator over fear, approach, avoidance, arousal, and safety behavior within the tested difficulty range. Recover dominant response modes, classify each as decaying, growing, oscillatory, or marginal, and perturb the fitted difficulty input to estimate outcome sensitivity. Present the clinician with an increase, hold, or reduction recommendation based on the stability and sensitivity of the dominant action-relevant mode, while displaying residual error, mode coupling, and the valid exposure range. The clinician may accept, modify, or reject the recommendation.","structural_mapping":[{"archetype_element":"Transformation Scope","domain_realization":"The patient-specific, trial-to-trial transition from the post-trial response state and delivered difficulty to the next observed response state, restricted to a prespecified exposure range and four-session evaluation window."},{"archetype_element":"State-Vector Definition","domain_realization":"A prespecified vector of fear, approach, avoidance, arousal, and safety behavior measured after each exposure trial."},{"archetype_element":"Invariant Mode Basis","domain_realization":"Approximately invariant combinations of the five response variables recovered from deliberately varied, clinician-approved graded trials."},{"archetype_element":"Modal Gain Spectrum","domain_realization":"Estimated trial-to-trial gain for each recovered response mode."},{"archetype_element":"Stable/Unstable Mode Partition","domain_realization":"Classification of modes as decaying, growing, oscillatory, or marginal within the fitted local regime."},{"archetype_element":"Modal Intervention Map","domain_realization":"Estimated effect of increasing, holding, or reducing exposure difficulty on each mode and on subsequent avoidance and safety behavior."},{"archetype_element":"Dominant Mode Selection Rule","domain_realization":"Prioritize a mode only when it is identifiable and has the greatest prespecified combination of persistence or growth, clinical consequence, and sensitivity to difficulty."},{"archetype_element":"Reconstruction Residual Check","domain_realization":"Reconstruct the observed five-variable trial response from retained modes and withhold modal pacing when out-of-sample residual magnitude or structure exceeds a prespecified tolerance."},{"archetype_element":"Interpretation Scope Contract","domain_realization":"Treat modes as local descriptive control coordinates within the tested patient, measures, sessions, and difficulty range—not as diagnoses, enduring traits, or proof of psychological causes."},{"archetype_element":"Mode-Coupling Register","domain_realization":"Report cross-effects, near-degenerate modes, and unstable mode identities that make a single-mode recommendation unreliable."}],"mechanism_mapping":[{"mechanism_slug":"mode_shape_testing","role":"Clinician-approved variation in graded exposure difficulty supplies empirical excitation from which the patient's response modes are estimated, with the tested conditions attached as the interpretation boundary.","counterfactual_removal":"Without empirical mode-shape recovery, the controller would have no patient-specific multivariate basis and would collapse toward a generic fear threshold or undifferentiated learned response rule."},{"mechanism_slug":"modal_stability_analysis","role":"The estimated gain of each response mode is interpreted to distinguish decaying, growing, oscillatory, and marginal dynamics before recommending escalation, holding, or regression.","counterfactual_removal":"Without stability classification, the system could identify correlated response patterns but could not distinguish a harmless persistent pattern from a growing avoidance-related pattern, eliminating the proposed pacing lever."},{"mechanism_slug":"modal_sensitivity_sweep","role":"Small in-model perturbations of difficulty estimate which identifiable modes and outcomes are actually responsive to the pacing decision and reveal cross-mode effects.","counterfactual_removal":"Without sensitivity estimation, the largest or least stable mode might be prioritized even when changing exposure difficulty cannot beneficially influence it, reducing the proposal to descriptive monitoring."}],"causal_chain":["Repeated clinician-approved graded trials generate within-patient variation in difficulty and multivariate responses.","That variation supports estimation of a local trial-to-trial transition operator and approximately invariant response modes.","Modal gains distinguish response combinations that decay from those that grow, oscillate, or remain marginal within the tested range.","A sensitivity sweep identifies whether changing difficulty is expected to move the clinically consequential dominant mode and exposes cross-mode effects.","The clinician uses the bounded recommendation to increase, hold, or reduce the next trial's difficulty while retaining override authority.","If the model is valid, fewer escalations occur during unstable avoidance or safety-behavior dynamics and fewer repetitions occur solely because stable arousal persists, improving safety-behavior reduction without an unacceptable dropout increase."],"baseline":"Clinician-managed exposure pacing in which the next difficulty is selected primarily from a scalar fear or anxiety score, including a fear-reduction threshold, while ordinary clinical judgment and stopping procedures remain available.","nearest_rivals":["Biofeedback-controlled VR exposure titration (C1, C2): combines physiological and behavioral or motion feedback to adjust exposure intensity or duration, but the located disclosures use thresholds or learned outputs rather than the full explicit modal stability-and-sensitivity rule.","Physiology-driven reinforcement-learning VRET (C3, C4, C6): individualizes exposure content from physiological response toward an anxiety target, but does not jointly decompose fear, approach, avoidance, arousal, and safety behavior into decision modes.","Idiographic system identification for adaptive behavioral interventions (C5): deliberately excites and estimates within-person dynamics for model-based intervention, but the located guidance is domain-general and does not demonstrate exposure pacing by eigenmode stability and outcome sensitivity.","Avoidance- and safety-behavior-informed exposure practice (C7): recognizes clinically important processes beyond fear reduction, but supplies treatment targets rather than a multivariate modal estimator and next-difficulty rule."],"remaining_contrastive_claim":"Between exposure trials, difficulty is selected by estimating a multivariate transition operator over fear, approach, avoidance, arousal, and safety behavior, decomposing it into dominant modes, and using each mode's stability class and intervention sensitivity—rather than a scalar anxiety target, physiological threshold, or undifferentiated learned policy—to advance, hold, or regress.","authority_safety":{"decision_authority":"The treating clinician retains sole authority over exposure content, difficulty, timing, continuation, and termination; the patient retains the ability to decline or stop according to the governing clinical protocol. The model is decision support only.","authorized_first_step":"Use completed-trial data in shadow mode to estimate modes and generate nonbinding increase, hold, or reduce recommendations that are hidden from pacing decisions until identifiability, residual, drift, and safety criteria have been reviewed.","excluded_actions":["Autonomous selection or delivery of exposure difficulty","Overriding a clinician's decision or a patient's request to pause or stop","Initiating exposure outside the clinician-approved hierarchy or tested difficulty range","Using modal estimates to diagnose a disorder or infer enduring traits or psychological causes","Suppressing ordinary clinical monitoring, adverse-event procedures, or stopping rules","Escalating when the model is unidentifiable, residuals exceed tolerance, modes are near-degenerate, or the state lies outside the fitted regime","Withholding indicated care because the model cannot produce a recommendation"],"halt_rollback":"Immediately revert to clinician-managed baseline pacing and suspend model recommendations when prespecified distress or adverse-event criteria are met, the patient or clinician stops participation, required state measures are missing, residual or drift limits fail, modes lose identifiability, or the proposed next difficulty lies outside the approved range. Preserve the last clinician-approved hierarchy state for review rather than automatically compensating with a larger subsequent change."},"negative_tests":{"strongest_counterevidence":"The searched record shows substantial collision: patents and adaptive VRET systems already perform response-dependent exposure titration using multimodal physiological and behavioral data, temporal machine learning, and reinforcement learning (C1-C4, C6), while domain-general behavioral-control guidance already combines deliberate excitation, idiographic system identification, and model-based dosing (C5). The residual distinction is therefore only the explicit joint invariant-mode stability-and-sensitivity decision rule, not adaptive, personalized, multimodal, or model-based exposure pacing generally.","problem_falsifier":"The nominated problem is falsified for this setting if trial-level avoidance, safety behavior, approach, and arousal add no reproducible pacing-relevant information beyond the scalar fear rule, or if fear-based pacing does not measurably produce either premature escalation during unstable avoidance dynamics or unnecessary repetition during clinically harmless arousal persistence.","intervention_falsifier":"The intervention is falsified if the multivariate modes cannot be identified reproducibly within the available trials, fail out-of-sample reconstruction or drift criteria, do not yield stable and sensitivity-dependent decisions distinct from fear-score pacing, or fail the four-session comparative criterion of at least 20% greater safety-behavior reduction without more than a 5-percentage-point increase in dropout.","risks":["Too few trials or insufficient difficulty variation may make the transition operator and dominant modes unidentifiable.","Near-degenerate, coupled, or drifting modes may produce unstable recommendations.","Measurement error or incompatible scaling across fear, behavior, and arousal variables may manufacture apparent modes.","A local linear model may fail after a large difficulty change or outside the tested exposure range.","Additional measurement and model-driven pacing could burden patients or distract clinicians.","Optimizing a fitted mode could worsen an unmodeled response represented in the residual.","Clinicians may overread descriptive modal coordinates as causal psychological constructs.","Automated recommendations could encourage escalation despite information not represented in the state vector."]},"next_evidence_step":"Conduct a bounded shadow-mode feasibility study using four sessions of clinician-directed graded exposure per participant. Before analysis, fix the five-variable state definition, scaling, local transition form, minimum excitation and identifiability requirements, dominant-mode rule, residual and drift tolerances, sensitivity perturbation, and halt criteria. Fit only on preceding trials, issue hidden increase/hold/reduce recommendations for each next trial, and compare them with actual fear-score-based decisions and subsequent safety behavior without changing care. Proceed to a clinician-supervised randomized comparison only if modes are reproducible, out-of-sample residuals meet tolerance, recommendations remain within the approved hierarchy, and the data support estimating the stated safety-behavior and dropout endpoints.","prior_art_status":"SEARCHED_BOUNDED","revision_record":{"parent_version":null,"progress_targets_addressed":["Expanded the terse hypothesis into explicit actors, observable state, consequence, intervention, mappings, causal chain, comparator, authority, safety exclusions, falsifiers, and a bounded evidence step.","Preserved the independent prior-art record's substantial-collision assessment and narrowed the proposal to its exact residual contrastive claim.","Made the initial evaluation nonbinding and shadow-mode before any treatment-changing comparison."],"conceptual_changes":["Defined the modes as local, approximate control coordinates rather than causal psychological constructs.","Separated modal stability from intervention sensitivity so an unstable but uncontrollable mode does not automatically determine pacing.","Restricted the innovation claim to explicit invariant-mode decomposition and classification for next-trial selection."],"operational_changes":["Specified the five-variable trial state and the decision boundary between completed trials.","Added identifiability, residual, drift, coupling, range, halt, and rollback gates.","Assigned final treatment authority to the clinician and limited the first study to hidden recommendations."],"evidence_changes":["Named the four closest prior-art families and their remaining differences using only sources C1-C7.","Converted the four-session outcome claim into an explicit intervention falsifier.","Defined a prospective, preceding-trials-only shadow evaluation as the first evidence step."],"claim_changes":["Disclaimed novelty in adaptive exposure pacing, multimodal sensing, personalization, temporal modeling, deliberate excitation, and automated titration generally.","Retained only the bounded-search distinction that stability and intervention sensitivity of explicit multivariate invariant modes jointly select the next exposure difficulty.","Kept the world-novelty boundary unresolved pending broader patent, bibliographic, non-English, product, and terminology searches."]}}