{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"invariant_mode_decomposition_design__human_computer_interaction","trajectory_id":"R","attempt_index":0,"archetype_slug":"invariant_mode_decomposition_design","domain_slug":"human_computer_interaction","decision":"CANDIDATE","problem_id":"hidden_coupled_interaction_breakdown_modes","causal_lever_id":"modal_control_of_interaction_state_transitions","proposal":{"problem":"HYPOTHESIS: In complex digital workflows, recurring breakdowns can arise from coupled patterns of retries, backtracking, errors, delays, help-seeking, and handoffs. Per-event and average usability metrics may miss a persistent or growing combination until abandonment or operator escalation becomes visible.","actors_substrate":["People completing a bounded digital workflow","Accessibility-dependent and novice users","Product, UX-research, analytics, and support teams","Versioned interface and its event-transition telemetry"],"observable_state":"For one workflow and interface version, a time-indexed vector of aggregate rates for task states, actions, validation errors, retries, backtracks, help use, latency bands, handoffs, completion, and abandonment; sensitive attributes are excluded from the vector.","consequence":"HYPOTHESIS: A weakly damped interaction-breakdown mode persists across steps while aggregate completion and isolated event rates appear acceptable, delaying diagnosis and encouraging symptom-level interface changes.","affected_objective":"Reliable, efficient, accessible task completion without avoidable abandonment, repeated error, or support escalation.","structural_mapping":[{"archetype_element":"A coupled transformation","domain_realization":"An estimated local operator maps the workflow-state vector at one step or interval to the next.","claim_kind":"HYPOTHESIS"},{"archetype_element":"Invariant directions rather than visible coordinates","domain_realization":"Modes are traceable combinations of retries, errors, backtracks, latency, help, and handoffs that approximately retain shape under the operator.","claim_kind":"HYPOTHESIS"},{"archetype_element":"Scalar modal response","domain_realization":"Eigenvalue magnitude and phase represent estimated persistence, damping, growth, or oscillation within the declared window.","claim_kind":"INFERENCE"},{"archetype_element":"Action relevance","domain_realization":"Controlled perturbations rank interface changes by their effect on completion and breakdown modes, not by mode magnitude alone.","claim_kind":"HYPOTHESIS"},{"archetype_element":"Residual accountability","domain_realization":"Held-out reconstruction and outcome residuals retain behavior omitted by selected modes.","claim_kind":"INFERENCE"},{"archetype_element":"Governed local validity","domain_realization":"Version, task, cohort, and time bounds plus gap and direction drift determine when the basis must be retired.","claim_kind":"INFERENCE"}],"component_map":[{"component":"Transformation Scope","status":"adapted","domain_realization":"One versioned workflow; a locally estimated next-step interaction operator."},{"component":"State-Vector Definition","status":"adapted","domain_realization":"Aggregated workflow-state and interaction-event rates with documented scaling and exclusions."},{"component":"Invariant Mode Basis","status":"direct","domain_realization":"Eigenvectors of the estimated operator, mapped back to named interaction variables."},{"component":"Modal Gain Spectrum","status":"direct","domain_realization":"Eigenvalues quantify within-window damping, persistence, growth, or oscillation."},{"component":"Dominant Mode Selection Rule","status":"adapted","domain_realization":"Retain modes exceeding preregistered persistence, outcome-sensitivity, or risk thresholds."},{"component":"Stable/Unstable Mode Partition","status":"adapted","domain_realization":"Classify rapidly damped, weakly damped, growing, marginal, and oscillatory interaction modes."},{"component":"Modal Intervention Map","status":"adapted","domain_realization":"Map prototype changes to modal coordinates and task-success effects."},{"component":"Reconstruction Residual Check","status":"direct","domain_realization":"Measure held-out state and outcome error plus structured residuals."},{"component":"Mode Drift Monitor","status":"direct","domain_realization":"Track eigenvalue, direction, and ordering changes after releases or population shifts."},{"component":"Interpretation Scope Contract","status":"direct","domain_realization":"State that modes are local aggregate descriptions, not user traits or proof of causation."},{"component":"Mode-Coupling Register","status":"direct","domain_realization":"Record near-degeneracy, non-normality, and cross-effects under perturbation."},{"component":"Local Linearization Window","status":"direct","domain_realization":"Bound validity by workflow, version, task mix, cohort definition, and calendar period."},{"component":"Spectral Gap Threshold","status":"direct","domain_realization":"Require a preregistered retained-versus-omitted separation and direction stability."}],"mechanism_dispositions":[{"slug":"eigendecomposition_workflow","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Decompose the explicitly estimated square interaction-transition operator.","counterfactual_removal":"No invariant directions or gains remain, so the proposal becomes ordinary multivariate monitoring."},{"slug":"modal_sensitivity_sweep","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Perturb prototype controls offline and in a bounded trial; log outcome and cross-mode effects.","counterfactual_removal":"Modes cannot be converted into an intervention-leverage ranking."},{"slug":"modal_stability_analysis","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Classify interaction modes only within the fitted local window.","counterfactual_removal":"Persistent or growing breakdown directions cannot be distinguished from transient variation."},{"slug":"mode_shape_testing","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Physical excitation and sensor placement do not directly fit human interaction; bounded prototype tests supply empirical validation instead.","counterfactual_removal":"No change; operator estimation and prototype validation remain."},{"slug":"network_spectral_centrality_analysis","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"State-node centrality ranks nodes, not coupled temporal breakdown directions.","counterfactual_removal":"No change to the causal chain."},{"slug":"power_iteration_probe","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Dominant-only recovery could hide near-degenerate harmful modes; the bounded operator permits fuller decomposition.","counterfactual_removal":"No change; full decomposition supplies the selection rule and gap."},{"slug":"principal_component_analysis","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Variance directions do not identify next-step propagation.","counterfactual_removal":"No change; covariance is not the proposed transformation."},{"slug":"reduced_order_model","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"A runnable surrogate is unnecessary for the first bounded test.","counterfactual_removal":"The decomposition, perturbation, and prototype test remain executable."},{"slug":"residual_reconstruction_test","disposition":"selected_load_bearing","contribution_type":"TEST_DESIGN","adaptation_or_rejection":"Use held-out reconstruction and outcome residuals to choose mode count and expose omitted structure.","counterfactual_removal":"There is no fidelity gate against a misleading low-order explanation."},{"slug":"singular_value_decomposition","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"SVD may diagnose non-normal amplification but its paired directions are not invariant temporal modes; it is not the primary mechanism.","counterfactual_removal":"No change unless eigendecomposition proves ill-conditioned, which is a halt condition."},{"slug":"spectral_decomposition_report","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Document variable loadings, uncertainty, coupling, residuals, and prohibited person-level interpretations.","counterfactual_removal":"The mathematics remains, but safe organizational interpretation is materially weakened."},{"slug":"spectral_gap_monitor","disposition":"selected_load_bearing","contribution_type":"OPERATIONAL","adaptation_or_rejection":"Re-estimate retained/omitted separation and direction drift after releases.","counterfactual_removal":"A fixed basis could remain operational after its simplifying assumption fails."}],"causal_chain":["Estimate a bounded next-step operator from version-consistent aggregate interaction traces.","Extract modes and gains; reject ill-conditioned, untraceable, or gapless decompositions.","Identify weakly damped, growing, or oscillatory combinations associated with task failure on held-out data.","Use sensitivity sweeps to identify prototype changes that reduce the harmful mode without worsening other modes or accessibility outcomes.","Compare the modal prototype with the ordinary baseline and nearest rival in a bounded test.","Deploy no further if residual, drift, harm, or performance gates fail."],"baseline":"Ordinary practice: funnel conversion, mean task time, error counts, and per-event threshold dashboards followed by local screen-level fixes.","nearest_rival":"Process mining or sequence clustering that finds common paths and loops but does not estimate invariant directions, modal gains, or control sensitivity.","authority_safety":{"affected_parties":["Workflow users","People using assistive technologies","Support staff","Product and operations teams"],"decision_authority":"A cross-functional product owner with UX-research, accessibility, privacy, analytics, and operational review; the model alone has no release authority.","authorized_first_step":"Retrospective, aggregate analysis of one workflow/version, followed by an accessible prototype test; preregister comparison metrics, residual tolerance, spectral-gap threshold, and subgroup harm limits.","excluded_actions":["Person-level scoring or labeling","Inferring disability, intent, competence, or vulnerability","Automatic denial, friction, personalization, or support withdrawal","Unreviewed production rollout","Pooling incompatible interface versions"],"halt_rollback":"Halt if modes are ill-conditioned, untraceable, unstable across resamples, gapless, residual-heavy, or associated with subgroup harm. Roll back the prototype and return to baseline monitoring."}},"negative_tests":{"strongest_counterevidence":"Held-out modes rotate materially across resamples or releases, add no prediction beyond funnel/process-mining features, or mainly encode instrumentation, task mix, or cohort changes.","analogy_break":"People adapt strategically and socially; an interface release changes both behavior and measurement. The interaction process is not an autonomous stationary linear system, so modes may be only short-lived local summaries.","failure_condition":"No stable local operator, usable spectral separation, traceable mode, acceptable residual, or ethically testable intervention exists.","problem_falsifier":"Across matched versions and tasks, failures are explained adequately by one directly observable variable or existing sequence metrics, with no repeatable coupled propagation pattern or incremental held-out value from modes.","intervention_falsifier":"Against baseline and process mining, the mode-targeted prototype fails to reduce preregistered harmful-mode persistence and task failure, or worsens completion, accessibility, error, or support outcomes beyond limits.","risks":["Reifying aggregate modes as user types","Telemetry and consent violations","Instrumentation artifacts masquerading as modes","Accessibility harms hidden by aggregate improvement","Non-normal or near-degenerate modes producing unstable interpretations","Interface updates invalidating the fitted operator"]},"null_rationale":null,"classification":{"candidate_kind":"TESTABLE_CONJECTURE","prior_art_status":"UNSEARCHED","evidence_maturity":"HYPOTHESIS"},"revision_change_log":{"revision_kind":"ORIGINAL","prior_problem_id":null,"prior_causal_lever_id":null,"problem_changed":false,"causal_lever_changed":false,"conceptual_changes":[],"operational_changes":[],"repairs_addressed":[]},"confidence":0.78,"generator_notes":"Closed-book structural transfer. Candidate depends on demonstrating a locally stable, explicit interaction-transition operator; otherwise classify the fit as failed."}