{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"invariant_mode_decomposition_design__psychology","arm":"PROPOSAL_FIRST","candidate_id":"imd-psychology-coupled-distress-mode-001","hypothesis_id":null,"version":0,"title":"Coupled-State Mode Monitoring for Voluntary Stress Self-Management","problem":"In repeated self-reports from adults managing work-related stress, tension, rumination, perceived control, task initiation, and social withdrawal may evolve as coupled combinations. Monitoring each variable separately can miss a persistent combination that is still modest in every coordinate but is amplified across successive observation windows. Coordinate-level alerts may therefore arrive late or recommend an action aimed at the most visible symptom rather than at the evolving combination.","actors":["Consenting adults participating in a voluntary stress self-management study","Licensed clinicians responsible for participant safety and interpretation","Behavioral-science researchers estimating and validating the transition model","Institutional ethics and data-governance reviewers"],"observable_state":"For each participant, a time-indexed state vector comprising brief ratings of tension, rumination, perceived control, task initiation, social withdrawal, and fatigue, plus subsequent measures of distress and planned-task completion. The focal observable is the participant's projection onto an estimated coupled mode and whether that coordinate grows, decays, or persists across consecutive observations.","consequence":"A weakly damped or growing distress-avoidance mode can accumulate before any single symptom crosses its alert threshold, potentially preceding avoidable distress escalation, withdrawal, or failure to begin intended tasks. Conversely, an invalid modal model could generate burdensome or misleading prompts.","affected_objective":"Detect and test action-relevant within-person patterns early enough to support participant-controlled, low-risk coping choices while minimizing false alerts, assessment burden, and unsupported clinical interpretation.","intervention":"Estimate a bounded, within-person local transition operator from repeated state vectors; decompose it into coupled modes; identify modes that are persistent and prospectively associated with the stated outcomes; and map only participant- and clinician-approved low-risk actions into those modal coordinates. Candidate actions might include a brief grounding exercise, initiating one predefined two-minute task step, or contacting a chosen support person. No action is deployed from the model until observational validation and a separate micro-randomized test show that the action changes the targeted mode or outcome without unacceptable burden.","structural_mapping":[{"archetype_element":"Transformation Scope","domain_realization":"The local mapping from one scheduled within-person psychological state vector to the next under an approximately stable daily routine."},{"archetype_element":"State-Vector Definition","domain_realization":"Standardized tension, rumination, perceived control, task initiation, social withdrawal, and fatigue measurements collected with fixed wording and timing."},{"archetype_element":"Invariant Mode Basis","domain_realization":"Weighted combinations of the measured variables that approximately preserve their direction under the fitted local transition."},{"archetype_element":"Modal Gain Spectrum","domain_realization":"Estimated eigenvalues indicating how strongly each coupled psychological-state direction grows, decays, persists, or oscillates between observations."},{"archetype_element":"Dominant Mode Selection Rule","domain_realization":"Retain a mode only if it is temporally reproducible, separated sufficiently from omitted modes, improves held-out outcome prediction, and has acceptable reconstruction residuals."},{"archetype_element":"Stable/Unstable Mode Partition","domain_realization":"Classify modes as decaying, marginal, oscillatory, or growing only within the participant, schedule, and observation range used for estimation."},{"archetype_element":"Modal Intervention Map","domain_realization":"Represent each approved action as a candidate input vector and estimate which action is expected to reduce the targeted mode rather than merely one visible symptom."},{"archetype_element":"Reconstruction Residual Check","domain_realization":"Reconstruct held-out state transitions from retained modes and inspect both error magnitude and structured residuals, including outcome-relevant low-variance patterns."},{"archetype_element":"Mode Drift Monitor","domain_realization":"Re-estimate mode direction and gain across rolling windows and suspend interpretation when modes rotate, reorder, or lose separation."},{"archetype_element":"Interpretation Scope Contract","domain_realization":"Treat modes as local dynamical summaries, not diagnoses, latent traits, or proof that their component variables share a single psychological cause."}],"mechanism_mapping":[{"mechanism_slug":"eigendecomposition_workflow","role":"Factor each explicit fitted transition matrix into a complete mode basis and gain spectrum, subject to conditioning checks.","counterfactual_removal":"Without it, the design reverts to separate-variable trajectories and cannot identify invariant coupled directions or their gains."},{"mechanism_slug":"modal_stability_analysis","role":"Classify each estimated mode by its repeated-transition behavior within the declared local window.","counterfactual_removal":"Without it, modes can be described but cannot be prioritized according to whether they decay, persist, oscillate, or grow."},{"mechanism_slug":"modal_sensitivity_sweep","role":"Perturb candidate action inputs in the fitted model to rank their predicted influence on the target outcome and record cross-effects on other modes.","counterfactual_removal":"Without it, the largest mode could be mistaken for the most actionable mode, and intervention spillovers would remain hidden."},{"mechanism_slug":"residual_reconstruction_test","role":"Determine whether retained modes reproduce held-out transitions and whether omitted residual structure carries safety- or outcome-relevant information.","counterfactual_removal":"Without it, compression could conceal meaningful behavior while presenting an unjustified tidy modal account."},{"mechanism_slug":"spectral_gap_monitor","role":"Track separation and directional drift of retained modes across time windows before any live use.","counterfactual_removal":"Without it, a once-stable basis could continue driving decisions after mode identity becomes ambiguous or the local regime changes."},{"mechanism_slug":"spectral_decomposition_report","role":"Document variable loadings, gains, coupling, residuals, operating window, and prohibited interpretations for clinician and ethics review.","counterfactual_removal":"Without it, numerical modes could be communicated as authoritative psychological constructs without their uncertainty and scope limits."}],"causal_chain":["Repeated observations define a multivariate within-person state and an explicitly bounded next-state transformation.","The fitted transformation may reveal coupled directions that are less visible than changes in individual questionnaire items.","Decomposition assigns each direction a scalar gain, making a persistent or growing distress-avoidance combination distinguishable from decaying fluctuations.","Held-out reconstruction, spectral separation, and drift checks determine whether that modal description is stable enough for further testing.","A sensitivity sweep maps approved actions to predicted changes in the target mode while exposing effects on other modes.","A micro-randomized comparison then tests whether an action selected for modal alignment actually reduces the next-window mode coordinate or the prespecified outcome relative to suitable controls.","Only if that causal test succeeds may a participant-controlled prompt be offered; reducing the target coordinate is then expected to interrupt its accumulation and lower subsequent distress or task disruption within the validated window."],"baseline":"Monitor each self-report item independently, alert when a symptom exceeds a fixed threshold, and let the participant or clinician select a familiar coping action using item-level interpretation and ordinary judgment.","nearest_rivals":["A supervised model that directly predicts next-window distress or task noncompletion from the raw state vector and selects whichever prompt performs best, without requiring invariant modes.","A multivariate symptom-score threshold that combines items into a fixed weighted total but does not model how coupled directions evolve over time.","Clinician-formulated individualized trigger plans based on interviews and observed patterns rather than intensive longitudinal estimation."],"remaining_contrastive_claim":"The proposal's surviving contrast is narrowly structural: if a reproducible coupled direction has interpretable gain and acceptable residuals, it supplies a dynamic intervention target and explicit model-validity monitors that a fixed symptom threshold lacks. It does not claim better outcomes than direct prediction or clinician formulation; those rivals must be compared empirically, and the modal design should be rejected if it adds no decision-relevant information.","authority_safety":{"decision_authority":"The participant retains authority over whether to record data, view an interpretation, or use an offered coping action. A licensed clinician determines whether any candidate action is clinically appropriate, and the ethics reviewer authorizes study progression; the model has no diagnostic, medication, crisis, employment, or eligibility authority.","authorized_first_step":"Conduct an ethics-approved observational feasibility study and locked retrospective analysis of consented data, producing participant-level transition estimates, residual checks, and a nonclinical decomposition report without issuing live recommendations.","excluded_actions":["Diagnosing a mental disorder or assigning a psychological trait from a mode","Starting, stopping, or changing medication or psychotherapy","Sending a prompt without prospective consent or an immediate opt-out","Using results for employment, insurance, discipline, eligibility, or involuntary monitoring","Interpreting a growing mode as imminent self-harm or substituting it for established crisis assessment","Deploying an action outside the participant, measurement schedule, or operating range in which the model was validated"],"halt_rollback":"Stop estimation or live testing for a participant if distress rises beyond the approved safety threshold, crisis indicators appear, missingness makes the state unreliable, residual error exceeds the locked tolerance, the retained spectral gap falls below its threshold, mode direction drifts beyond tolerance, or prompts create unacceptable burden. Disable model-driven prompts, revert to the participant's existing clinician-approved plan, preserve an audit record, and require fresh clinical and ethics review before resumption."},"negative_tests":{"strongest_counterevidence":"A preregistered direct predictor using the same measurements matches or exceeds held-out prediction and intervention performance while the purported modes are unstable across windows, poorly separated, or leave structured residuals related to the outcome.","problem_falsifier":"The inferred transition is adequately explained by one directly observable variable, or no coupled mode remains directionally stable and action-relevant on held-out observations; in that case the claimed hidden-mode problem is absent for this setting.","intervention_falsifier":"In a sufficiently powered micro-randomized comparison, modal-aligned actions do not change the targeted next-window mode coordinate or prespecified outcome more than a generic approved action, an item-threshold-selected action, or no prompt, or they cause greater burden or adverse effects.","risks":["Sparse, noisy, or reactive self-reports can produce spurious modes.","Irregular observation intervals can make estimated gains incomparable.","Nonlinear or rapidly changing dynamics can invalidate the local linear approximation.","Near-degenerate or non-normal modes can be numerically unstable and hard to interpret.","Actions chosen from observational associations may not causally control the mode.","Frequent assessment or prompting may increase rumination, burden, or disengagement.","Psychological labels attached to mathematical modes may stigmatize participants or falsely imply diagnosis.","Averages across participants may conceal incompatible individual dynamics."]},"next_evidence_step":"Run one preregistered observational pilot with at most 24 consenting adults over 28 days, using no model-driven intervention. Fit participant-level regularized transition operators on days 1–14 and evaluate days 15–28. Before analysis, lock acceptable missingness, conditioning, held-out reconstruction error, mode-direction similarity, spectral-gap, and incremental-prediction criteria. Compare the modal model against independent item thresholds, a fixed composite score, and a direct regularized next-outcome predictor. Advance only those participant models meeting every validity criterion to the design of a separately reviewed micro-randomized test.","prior_art_status":"UNSEARCHED","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}