{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"invariant_mode_decomposition_design__gender_studies","trajectory_id":"R","attempt_index":0,"archetype_slug":"invariant_mode_decomposition_design","domain_slug":"gender_studies","decision":"CANDIDATE","problem_id":"coupled_gender_harm_escalation_blind_spot","causal_lever_id":"mode_targeted_procedural_prevention","proposal":{"problem":"Organizations monitoring harassment, misgendering, exclusion, complaint handling, retaliation, and attrition as separate indicators can miss recurring combinations that jointly escalate gender-related harm, especially when averages conceal trajectories concentrated among intersectionally marginalized groups.","actors_substrate":["workers, students, or members exposed to gender-related harms","transgender, nonbinary, queer, women, and intersectionally marginalized subgroups","complaint handlers, supervisors, and equity staff","deidentified longitudinal incident, climate, process, and retention records","institutional policy and support processes"],"observable_state":"Repeated deidentified measurements show several harm, response, and exit variables moving together; aggregate indicators remain stable while particular combinations persist or grow within bounded organizational regimes.","consequence":"Institutions react to isolated symptoms, overlook escalating configurations, and may direct prevention resources toward visible but low-leverage indicators.","affected_objective":"Earlier, more equitable prevention of gender-related harm without profiling individuals or treating descriptive statistical modes as identities or causes.","structural_mapping":[{"archetype_element":"coupled transformation","domain_realization":"A locally estimated transition operator maps one period's vector of harm, reporting, institutional-response, support-access, and exit indicators to the next period.","claim_kind":"HYPOTHESIS"},{"archetype_element":"invariant directions","domain_realization":"Approximately persistent combinations of indicators represent recurring organizational trajectories, not essential gender categories or individual types.","claim_kind":"INFERENCE"},{"archetype_element":"scalar modal response","domain_realization":"Estimated gain indicates whether a coupled trajectory tends to decay, persist, oscillate, or grow within the study window.","claim_kind":"INFERENCE"},{"archetype_element":"dominant-mode intervention","domain_realization":"Procedural changes are selected for their predicted effect on harmful indicator combinations rather than on one headline metric.","claim_kind":"HYPOTHESIS"},{"archetype_element":"residual visibility","domain_realization":"Holdout reconstruction errors and subgroup-patterned residuals preserve harms the retained modes fail to represent.","claim_kind":"INFERENCE"},{"archetype_element":"governed local validity","domain_realization":"Modes are retired after policy, population, reporting, or data-definition shifts invalidate the fitted regime.","claim_kind":"INFERENCE"}],"component_map":[{"component":"Transformation Scope","status":"adapted","domain_realization":"One-period transitions within a specified institution, policy regime, population, and observation interval."},{"component":"State-Vector Definition","status":"adapted","domain_realization":"Deidentified rates and process measures for harms, reporting, response delay, retaliation signals, support access, belonging, and exit, with missingness indicators."},{"component":"Invariant Mode Basis","status":"adapted","domain_realization":"Eigenvectors of the locally estimated transition operator, interpreted only as coupled indicator trajectories."},{"component":"Modal Gain Spectrum","status":"direct","domain_realization":"Eigenvalues with uncertainty intervals quantify estimated persistence, growth, decay, or oscillation."},{"component":"Dominant Mode Selection Rule","status":"adapted","domain_realization":"Retain modes only when holdout stability, harm relevance, subgroup residual limits, and uncertainty thresholds are satisfied."},{"component":"Stable/Unstable Mode Partition","status":"adapted","domain_realization":"Classify modes relative to a preregistered persistence boundary, with an indeterminate class for intervals crossing it."},{"component":"Modal Intervention Map","status":"adapted","domain_realization":"Map feasible procedural bundles—response deadlines, confidential support, anti-retaliation review, supervisor accountability—to predicted modal effects."},{"component":"Reconstruction Residual Check","status":"direct","domain_realization":"Measure total and subgroup-stratified holdout residual magnitude and structure."},{"component":"Mode Drift Monitor","status":"direct","domain_realization":"Re-estimate direction, ordering, and gain after scheduled intervals and institutional changes."},{"component":"Interpretation Scope Contract","status":"adapted","domain_realization":"Forbids reading modes as identities, individual risk scores, causal proof, or universally stable social structures."},{"component":"Mode-Coupling Register","status":"direct","domain_realization":"Records near-degenerate, non-orthogonal, and intervention-linked modes that cannot be acted on independently."},{"component":"Local Linearization Window","status":"adapted","domain_realization":"Validity is limited to the sampled institution, interval, variable definitions, and approximately unchanged policy regime."},{"component":"Spectral Gap Threshold","status":"direct","domain_realization":"A preregistered separation and bootstrap-stability threshold governs whether modal ranking is actionable."}],"mechanism_dispositions":[{"slug":"eigendecomposition_workflow","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Decompose the estimated square transition operator; report conditioning and uncertainty rather than claiming an exact social law.","counterfactual_removal":"Without it, coupled temporal directions and their gains are unavailable, collapsing the modal causal lever."},{"slug":"modal_sensitivity_sweep","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Perturb feasible procedural controls in the local model and record outcome and cross-mode responses.","counterfactual_removal":"Removing it leaves no defensible bridge from descriptive modes to intervention choice."},{"slug":"modal_stability_analysis","disposition":"selected_supporting","contribution_type":"OPERATIONAL","adaptation_or_rejection":"Classify growth or decay only inside the fitted institutional regime and propagate interval uncertainty.","counterfactual_removal":"Monitoring could not distinguish persistent or growing trajectories from transient variation."},{"slug":"mode_shape_testing","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Physical excitation-and-sensor testing does not transfer safely to people or institutions; observational validation is supplied by holdout data.","counterfactual_removal":"No proposed step changes."},{"slug":"network_spectral_centrality_analysis","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"The target is coupled indicator dynamics, not ranking people or network nodes; such ranking would add profiling risk.","counterfactual_removal":"No proposed step changes."},{"slug":"power_iteration_probe","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"A dominant-only estimate would hide secondary harmful modes and is unreliable near a small gap.","counterfactual_removal":"No proposed step changes."},{"slug":"principal_component_analysis","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Variance directions are not temporal invariant directions; high variance need not equal harmful persistence.","counterfactual_removal":"No proposed step changes."},{"slug":"reduced_order_model","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"A runnable surrogate adds extrapolation authority without being necessary for the bounded test.","counterfactual_removal":"No proposed step changes."},{"slug":"residual_reconstruction_test","disposition":"selected_load_bearing","contribution_type":"TEST_DESIGN","adaptation_or_rejection":"Evaluate retained modes on time-held-out data and inspect subgroup-structured residuals.","counterfactual_removal":"There would be no hard check that compression omits consequential harms."},{"slug":"singular_value_decomposition","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"SVD may diagnose conditioning, but its paired reconstruction directions do not supply the required temporal stability interpretation.","counterfactual_removal":"The primary eigen-based test remains executable."},{"slug":"spectral_decomposition_report","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Publish uncertainty, residuals, couplings, subgroup limits, prohibited interpretations, and retirement conditions with the results.","counterfactual_removal":"Decision makers could mistake descriptive modes for identities, causes, or individualized predictions."},{"slug":"spectral_gap_monitor","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Track bootstrapped gap and mode rotation; suspend use when either breaches threshold.","counterfactual_removal":"A once-valid simplification could continue after the dominant subspace becomes unstable."}],"causal_chain":["Separate dashboards obscure jointly evolving harm and response variables.","A bounded transition model estimates combinations that persist or amplify together.","Residual, uncertainty, conditioning, and gap tests determine whether any combination is decision-worthy.","Sensitivity analysis identifies a feasible procedural bundle predicted to damp a harmful mode without worsening registered coupled modes.","A consented, cluster-level pilot compares that bundle with ordinary practice while retaining all raw harm measures.","Reduced harmful-mode gain plus improved direct outcomes would support the intervention; disagreement triggers rejection or model revision."],"baseline":"Ordinary practice: separate annual climate summaries, complaint counts, response-time metrics, and case-by-case remediation, sometimes stratified but without modeling their joint temporal evolution.","nearest_rival":"Intersectionally stratified longitudinal regression or multilevel event-history analysis targeting prespecified outcomes without an invariant-mode decomposition.","authority_safety":{"affected_parties":["people experiencing gender-related harm","complainants and non-reporters","transgender, nonbinary, queer, women, and intersectionally marginalized groups","staff whose work processes change","advocates and support providers"],"decision_authority":"An accountable institutional equity body with affected-party representation, independent privacy review, and authority limited to aggregate procedural changes.","authorized_first_step":"Analyze deidentified historical and prospectively held-out aggregate data in shadow mode; only after review, test one reversible procedural bundle in two consenting units with a matched or stepped comparator, preregistered thresholds, and no individual scoring.","excluded_actions":["ranking, labeling, disciplining, or surveilling individuals","inferring gender identity or sexuality from behavior","reducing services because a mode appears stable","treating non-reporting as absence of harm","deploying outside the validated regime","suppressing raw or qualitative evidence that conflicts with the model"],"halt_rollback":"Halt and restore prior procedures if consent is withdrawn, privacy is breached, direct harm or subgroup disparity worsens beyond the preregistered bound, residuals become subgroup-structured, the spectral gap or mode stability fails, or staff use outputs for individual decisions."}},"negative_tests":{"strongest_counterevidence":"Reported co-movement may mainly reflect changing survey participation, reporting access, coding rules, or institutional shocks rather than a persistent harm-generating process; qualitative accounts may also show that mathematically grouped indicators have incompatible meanings.","analogy_break":"Unlike a fixed physical operator, gendered power relations are reflexive: measurement, disclosure, policy, and interpretation can change the process itself, so modes may lack stable identity and cannot be treated as natural social kinds.","failure_condition":"No adequately stable local transition operator, actionable spectral separation, acceptable subgroup residual, or ethically measurable state vector exists.","problem_falsifier":"On repeated data, separate indicators and prespecified intersectional models detect harmful escalation at least as early and reliably as coupled trajectories, with no reproducible residual pattern showing a missed joint process.","intervention_falsifier":"In the bounded pilot, the selected procedural bundle does not reduce the preregistered harmful-mode gain and direct harm outcomes relative to baseline, or it worsens any protected subgroup outcome despite apparently improving the modal metric.","risks":["reifying statistical modes as gender identities or universal structures","concealing rare severe harms through compression","selection bias from differential reporting and attrition","privacy leakage from small intersectional cells","policy gaming that improves measured modes without improving lived conditions","false causal confidence from an observational transition model","resource diversion from survivor-defined needs"]},"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":0.72,"generator_notes":"The fit depends on repeated, sufficiently comparable measurements and an explicit local transition model. The proposal treats modal structure as a fallible organizational diagnostic, not as a taxonomy of people or proof of causation."}