{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"invariant_mode_decomposition_design__criminology_forensic","trajectory_id":"R","attempt_index":0,"archetype_slug":"invariant_mode_decomposition_design","domain_slug":"criminology_forensic","decision":"CANDIDATE","problem_id":"forensic_toxicology_coupled_qc_drift","causal_lever_id":"action_relevant_drift_mode_control","proposal":{"problem":"Forensic toxicology laboratories commonly monitor calibration, blank, control, retention-time, ion-ratio, and sensitivity indicators separately. Multiple sub-threshold changes may evolve together as a coupled degradation pattern, allowing unreliable positive or negative findings before any single indicator triggers corrective action.","actors_substrate":["forensic toxicology laboratory","quality-assurance staff","instrument and sample-preparation workflow","sequential calibrator, blank, and control measurements","case-result review process"],"observable_state":"Across successive analytical runs, several quality-control measurements co-move in repeatable combinations while remaining individually within limits; later reruns, control failures, or amended findings cluster after particular combinations begin growing.","consequence":"Latent coupled drift may increase erroneous, delayed, or unnecessarily repeated forensic findings, affecting defendants, victims, investigations, and courts.","affected_objective":"Reliable and timely forensic toxicology conclusions with traceable quality-control decisions.","structural_mapping":[{"archetype_element":"Many visible variables are acted on jointly by a repeated transformation.","domain_realization":"The analytical workflow carries an instrument-and-preparation state from one run to the next, jointly affecting several QC measurements.","claim_kind":"HYPOTHESIS"},{"archetype_element":"Surface coordinates can conceal a growing direction.","domain_realization":"Each QC indicator may remain within its limit even when their weighted combination persistently approaches a failure-associated state.","claim_kind":"HYPOTHESIS"},{"archetype_element":"Modes are ranked by response and consequence rather than visibility alone.","domain_realization":"Estimated drift modes are ranked by persistence, association with held-out QC failures, and sensitivity to authorized maintenance actions.","claim_kind":"INFERENCE"},{"archetype_element":"Residual and regime checks bound modal interpretation.","domain_realization":"Unexplained analyte-specific failures, mode rotation, and changes in instrument, method, reagent lot, or maintenance state remain visible and suspend use of the model.","claim_kind":"INFERENCE"}],"component_map":[{"component":"Transformation Scope","status":"adapted","domain_realization":"A regularized run-to-run transition operator for one instrument, method, and defined maintenance regime."},{"component":"State-Vector Definition","status":"direct","domain_realization":"Standardized calibration residuals, blank response, control recovery, retention-time shift, ion-ratio deviation, internal-standard response, and sensitivity indicators per run."},{"component":"Invariant Mode Basis","status":"adapted","domain_realization":"Approximately invariant combinations of QC variables estimated from the local transition operator."},{"component":"Modal Gain Spectrum","status":"direct","domain_realization":"Estimated persistence or amplification factor for every retained drift mode, with uncertainty and conditioning diagnostics."},{"component":"Dominant Mode Selection Rule","status":"direct","domain_realization":"Retain modes only when gain, held-out failure association, or reconstruction contribution exceeds preregistered thresholds."},{"component":"Stable/Unstable Mode Partition","status":"adapted","domain_realization":"Classify modes as decaying, persistent, oscillatory, or growing within the validated run regime."},{"component":"Modal Intervention Map","status":"adapted","domain_realization":"Map permitted responses—recalibration, cleaning, reagent replacement, maintenance inspection, or confirmatory rerun—to predicted changes in modal coordinates."},{"component":"Reconstruction Residual Check","status":"direct","domain_realization":"Reconstruct original QC trajectories and inspect held-out residual magnitude and structure, including analyte-specific residuals."},{"component":"Mode Drift Monitor","status":"direct","domain_realization":"Re-estimate mode direction, ordering, gain, and residual behavior on a fixed review schedule."},{"component":"Interpretation Scope Contract","status":"direct","domain_realization":"Modes are laboratory-process warnings, not proof that any case result is wrong and not evidence about a person's culpability."},{"component":"Mode-Coupling Register","status":"direct","domain_realization":"Record near-degenerate, non-orthogonal, or intervention-coupled modes that cannot be treated independently."},{"component":"Local Linearization Window","status":"adapted","domain_realization":"Validity is limited to the specified instrument, method, QC design, reagent and maintenance regime, and observed range of operating states."},{"component":"Spectral Gap Threshold","status":"direct","domain_realization":"A preregistered minimum separation between action-relevant and omitted modes gates simplified monitoring."}],"mechanism_dispositions":[{"slug":"eigendecomposition_workflow","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Decompose the estimated run-transition operator only if it is sufficiently stable, identifiable, and well-conditioned.","counterfactual_removal":"Without invariant directions and gains, the proposal collapses to coordinate-level QC monitoring."},{"slug":"modal_sensitivity_sweep","disposition":"selected_supporting","contribution_type":"TEST_DESIGN","adaptation_or_rejection":"In retrospective simulations, perturb modes and authorized maintenance inputs to rank outcome-relevant leverage and detect cross-effects.","counterfactual_removal":"Modes could still be detected, but the test could not distinguish actionable modes from merely prominent ones."},{"slug":"modal_stability_analysis","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Classify run-to-run modes relative to a discrete-time stability boundary within the validated regime.","counterfactual_removal":"The system could not distinguish harmless variation from persistent or growing degradation."},{"slug":"mode_shape_testing","disposition":"incompatible","contribution_type":"NONE","adaptation_or_rejection":"Physical excitation-response testing does not match an operational forensic assay workflow and could disrupt evidence processing.","counterfactual_removal":"No loss; modes are estimated from routine QC transitions."},{"slug":"network_spectral_centrality_analysis","disposition":"incompatible","contribution_type":"NONE","adaptation_or_rejection":"The relevant substrate is a multivariate process state, not a node-importance problem.","counterfactual_removal":"No loss; node rankings are not part of the causal chain."},{"slug":"power_iteration_probe","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Dominant-only estimation would conceal secondary or nearly tied hazardous modes in a modest state space.","counterfactual_removal":"No loss; full-spectrum estimation is required."},{"slug":"principal_component_analysis","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"PCA is the nearest rival but ranks variance rather than run-to-run amplification or persistence.","counterfactual_removal":"No loss to the proposed causal chain; PCA remains a benchmark."},{"slug":"reduced_order_model","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"A runnable process surrogate is unnecessary for the first warning-system test and risks unjustified extrapolation.","counterfactual_removal":"No loss; shadow monitoring can use modal scores directly."},{"slug":"residual_reconstruction_test","disposition":"selected_load_bearing","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Use held-out reconstruction error and residual structure as a hard gate on mode selection and deployment.","counterfactual_removal":"Low-variance analyte-specific failures could be hidden by the modal summary."},{"slug":"singular_value_decomposition","disposition":"selected_supporting","contribution_type":"TEST_DESIGN","adaptation_or_rejection":"Compare singular amplification with eigenmode results to expose non-normality and ill-conditioned eigenvectors; do not label singular vectors invariant modes.","counterfactual_removal":"The core analysis remains possible, but transient amplification and fragile eigenbases are easier to miss."},{"slug":"spectral_decomposition_report","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Document variable loadings, uncertainty, couplings, residuals, scope, and prohibited interpretations for laboratory review.","counterfactual_removal":"Technical outputs could be mistaken for case-level proof or causal attribution."},{"slug":"spectral_gap_monitor","disposition":"selected_supporting","contribution_type":"OPERATIONAL","adaptation_or_rejection":"Track gap, mode rotation, ordering, gains, and residuals; suspend alerts when thresholds fail.","counterfactual_removal":"A once-valid basis could remain in use after process change."}],"causal_chain":["A repeated laboratory workflow induces coupled run-to-run changes in the QC state.","A locally valid transition estimate exposes persistent or growing combinations hidden by separate limits.","Residual, conditioning, and spectral-gap gates remove unstable or incomplete modal summaries from use.","Sensitivity analysis identifies which permitted maintenance responses are plausibly able to damp action-relevant modes.","Shadow alerts prompt earlier technical review and confirmatory checks.","If the hypothesized modes predict failures out of sample, earlier review reduces exposure to unreliable findings without treating the alert as case evidence."],"baseline":"Independent QC limits and univariate control charts, followed by recalibration, rerun, or maintenance after a coordinate breaches its rule.","nearest_rival":"PCA-based multivariate statistical process control using high-variance component scores without an explicit run-transition or stability interpretation.","authority_safety":{"affected_parties":["people whose specimens or cases are analyzed","defendants and convicted persons","victims and families","laboratory analysts and managers","investigators, attorneys, and courts"],"decision_authority":"The laboratory director and quality manager may authorize a shadow pilot; validated technical procedures and applicable legal review govern any later operational use.","authorized_first_step":"Retrospectively fit the model on 30 consecutive runs from one instrument-method regime, preregister thresholds, and evaluate the next 10 runs in shadow mode against existing QC failures, reruns, and maintenance records.","excluded_actions":["altering, suppressing, or reporting a case result from a modal alert alone","using modal scores as evidence of guilt, innocence, intoxication, or analyst misconduct","changing validated acceptance criteria during the pilot","pooling instruments or methods outside the declared scope","delaying required action under existing QC rules"],"halt_rollback":"Stop and discard the pilot basis if conditioning is unacceptable, the spectral gap falls below threshold, held-out residuals exceed budget, modes rotate beyond tolerance, existing QC failures are missed, or false alerts exceed the preregistered limit; preserve ordinary QC as the sole operational system."}},"negative_tests":{"strongest_counterevidence":"Historical unreliable results may be explained almost entirely by isolated analyte, reagent-lot, contamination, transcription, or maintenance events already caught by existing rules, with no repeatable coupled transition mode out of sample.","analogy_break":"Unlike a fixed physical operator, laboratory behavior is stochastic, partly observed, changed by analyst responses, and subject to abrupt instrument, method, lot, and maintenance shifts; approximate modes may therefore lack stable identity or causal meaning.","failure_condition":"The transition estimate is ill-conditioned or regime-dependent, relevant eigenvectors are unstable, no usable spectral separation exists, or residuals retain safety-relevant structure.","problem_falsifier":"Across scoped historical and prospective-shadow runs, sub-threshold multivariate patterns do not reproducibly precede QC failures, reruns, maintenance events, or amended findings beyond individual indicators and ordinary covariates.","intervention_falsifier":"Even when a reproducible hazardous mode exists, authorized actions selected by the modal intervention map do not reduce its score or subsequent failure rate relative to existing response rules in a bounded prospective trial.","risks":["False alerts could increase reruns, delays, cost, or unnecessary maintenance.","A mode may encode benign batch composition or documentation practices rather than degradation.","Retrospective labels and maintenance actions may create confounding or leakage.","Aggregation could hide analyte-specific or rare safety-critical failures.","A technical warning could be overstated in legal proceedings.","Feedback from acting on alerts could invalidate the fitted transition."]},"null_rationale":null,"classification":{"candidate_kind":"MECHANISM_ADAPTATION","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":0.79,"generator_notes":"Closed-book structural transfer. Empirical existence, stability, predictive value, and controllability of forensic-laboratory drift modes remain hypotheses requiring the bounded shadow test."}