{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"invariant_mode_decomposition_design__neuroscience","trajectory_id":"R","attempt_index":0,"archetype_slug":"invariant_mode_decomposition_design","domain_slug":"neuroscience","decision":"CANDIDATE","problem_id":"distributed_preictal_instability_missed_by_channelwise_monitoring","causal_lever_id":"unstable_neural_state_mode_suppression","proposal":{"problem":"HYPOTHESIS: In focal-epilepsy monitoring, channel-wise thresholds can miss a distributed, weakly damped neural-state direction that grows before a seizure; stimulation chosen by contact-level activity may therefore arrive late or target the wrong combination of sites.","actors_substrate":["patients undergoing clinically indicated intracranial monitoring","coupled neural populations sampled by electrodes","epilepsy clinicians and clinical investigators","recorded electrophysiology and stimulation logs"],"observable_state":"A time-windowed state vector of prespecified features from all retained electrodes, with an estimated local transition operator, its modal coordinates and gains, reconstruction residual, spectral gap, and mode drift.","consequence":"Missed or late warning, ineffective stimulation, and avoidable exposure of neural tissue to stimulation.","affected_objective":"Earlier recognition and safer suppression of seizure-generating dynamics without increasing false alarms or stimulation burden.","structural_mapping":[{"archetype_element":"coupled transformation","domain_realization":"INFERENCE: A locally estimated transition matrix maps one multielectrode neural-state window to the next.","claim_kind":"INFERENCE"},{"archetype_element":"hidden invariant directions","domain_realization":"HYPOTHESIS: Combinations of electrode features, rather than single channels, contain persistent or growing preictal dynamics.","claim_kind":"HYPOTHESIS"},{"archetype_element":"modal scalar response","domain_realization":"Eigenvalue-derived persistence or growth classifies candidate neural modes within the fitted window.","claim_kind":"CORPUS"},{"archetype_element":"action-relevant mode selection","domain_realization":"Modes are retained by held-out seizure association, stimulation-response sensitivity, spectral separation, and residual tolerance—not magnitude alone.","claim_kind":"HYPOTHESIS"},{"archetype_element":"modal intervention map","domain_realization":"INFERENCE: Existing stimulation events can estimate which contact patterns perturb the selected mode and with what cross-mode effects.","claim_kind":"INFERENCE"},{"archetype_element":"bounded interpretation","domain_realization":"The model is descriptive and locally predictive unless perturbation evidence supports a causal interpretation; residual, drift, and regime limits remain visible.","claim_kind":"CORPUS"}],"component_map":[{"component":"Transformation Scope","status":"adapted","domain_realization":"Patient- and regime-specific short-horizon neural transition operator."},{"component":"State-Vector Definition","status":"adapted","domain_realization":"Prespecified multielectrode features, scaling, lag, exclusions, and missing-data rules."},{"component":"Invariant Mode Basis","status":"direct","domain_realization":"Eigenvectors of the fitted local transition operator."},{"component":"Modal Gain Spectrum","status":"direct","domain_realization":"Eigenvalues with uncertainty estimates."},{"component":"Dominant Mode Selection Rule","status":"adapted","domain_realization":"Retain modes meeting predictive or stimulation-leverage criteria while satisfying residual and gap gates."},{"component":"Stable/Unstable Mode Partition","status":"direct","domain_realization":"Decaying, marginal, oscillatory, and growing modes within the local window."},{"component":"Modal Intervention Map","status":"adapted","domain_realization":"Estimated effects of recorded contact-level pulses on modal coordinates."},{"component":"Reconstruction Residual Check","status":"direct","domain_realization":"Held-out state and clinically consequential residual error."},{"component":"Mode Drift Monitor","status":"direct","domain_realization":"Track mode rotation, gain changes, reordering, and residual growth."},{"component":"Interpretation Scope Contract","status":"direct","domain_realization":"Patient, state, time horizon, preprocessing, and noncausal-use limits."},{"component":"Mode-Coupling Register","status":"adapted","domain_realization":"Cross-mode effects, near-degeneracy, and non-normal sensitivity."},{"component":"Local Linearization Window","status":"direct","domain_realization":"Explicit amplitude, behavioral-state, and prediction-horizon bounds."},{"component":"Spectral Gap Threshold","status":"direct","domain_realization":"Prespecified separation and mode-identity thresholds required for actionability."}],"mechanism_dispositions":[{"slug":"eigendecomposition_workflow","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Decompose the estimated local neural transition operator into a full basis and gain spectrum.","counterfactual_removal":"Without it, growing directions cannot be separated from electrode coordinates."},{"slug":"modal_sensitivity_sweep","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Use recorded stimulation perturbations to rank modal leverage and register cross-effects.","counterfactual_removal":"The analysis could describe instability but could not nominate an outcome-relevant control direction."},{"slug":"modal_stability_analysis","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Classify temporal modes only inside the fitted neural regime.","counterfactual_removal":"The proposed preictal lever loses its growing-versus-decaying distinction."},{"slug":"mode_shape_testing","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Empirical perturbation identification is redundant because the candidate starts with an estimated explicit operator; unobserved excitation would also confound absence with invisibility.","counterfactual_removal":"No change; perturbation evidence is handled by the sensitivity sweep."},{"slug":"network_spectral_centrality_analysis","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Node importance in a connectivity eigenvector does not establish temporal instability or stimulation leverage.","counterfactual_removal":"No change to the causal chain."},{"slug":"power_iteration_probe","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Dominant-only recovery is inadequate when near-leading modes, coupling, and residual reconstruction must be checked.","counterfactual_removal":"No change; full decomposition supplies the required modes."},{"slug":"principal_component_analysis","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Variance directions are not necessarily dynamical growth directions.","counterfactual_removal":"No change; it remains only a possible rival feature reduction."},{"slug":"reduced_order_model","disposition":"selected_supporting","contribution_type":"OPERATIONAL","adaptation_or_rejection":"Project candidate interventions into retained modal coordinates for fast offline replay within the validated window.","counterfactual_removal":"The test remains possible with the full model but loses efficient bounded intervention simulation."},{"slug":"residual_reconstruction_test","disposition":"selected_load_bearing","contribution_type":"TEST_DESIGN","adaptation_or_rejection":"Gate mode count and validity using held-out error and structured residuals.","counterfactual_removal":"A compact modal story could pass despite omitting clinically consequential dynamics."},{"slug":"singular_value_decomposition","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Useful for input-output amplification, but singular directions do not supply the required temporal stability verdict.","counterfactual_removal":"No change to this eigenmode-based test."},{"slug":"spectral_decomposition_report","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Record interpretation limits, uncertainty, coupling, and prohibited causal readings.","counterfactual_removal":"Mathematical results remain, but misuse risk rises and governance becomes inadequate."},{"slug":"spectral_gap_monitor","disposition":"selected_load_bearing","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Hard-gate use when retained modes lose separation or identity.","counterfactual_removal":"A drifting or near-degenerate basis could continue guiding decisions after its warrant disappears."}],"causal_chain":["Coupled neural activity is represented by a locally valid transition operator.","Decomposition exposes a distributed mode whose estimated gain is weakly damped or growing.","Held-out records test whether that mode changes before seizures beyond channel-wise signals.","Recorded stimulation responses identify contact combinations that reduce its coordinate without amplifying harmful modes.","Residual, gap, coupling, and drift gates restrict use to regimes where the mapping remains credible.","HYPOTHESIS: Suppressing that mode would reduce seizure progression more reliably than contact-level targeting."],"baseline":"Ordinary baseline: clinician-selected contacts plus per-channel amplitude, band-power, or event thresholds, without an explicit multivariate transition-mode model.","nearest_rival":"A multichannel nonlinear seizure predictor that forecasts onset accurately but selects stimulation from feature importance rather than a dynamical modal intervention map.","authority_safety":{"affected_parties":["monitored patients","caregivers","clinical staff","people underrepresented in the retrospective data"],"decision_authority":"For the first test, data custodians, investigators, and the ethics/IRB authority; any prospective stimulation additionally requires patient consent and the treating epilepsy team's authority.","authorized_first_step":"A single-site retrospective replay using existing intracranial recordings and clinically delivered stimulation logs, with prespecified patient-level splits, no care changes, comparison against baseline and nearest rival, and subgroup reporting.","excluded_actions":["autonomous stimulation","altering clinical treatment","withholding standard alarms or care","extrapolating across patients or recording regimes without validation","treating descriptive modes as anatomical seizure sources"],"halt_rollback":"Halt analysis claims if held-out residual, false-alarm burden, spectral-gap, mode-drift, or subgroup-performance limits fail; retire the basis and revert to the ordinary clinical baseline. Any later prospective protocol must stop stimulation on adverse events or predefined worsening."}},"negative_tests":{"strongest_counterevidence":"A well-calibrated channel-wise or nonlinear multichannel rival matches or exceeds held-out seizure warning and stimulation-response prediction, while modal gain and mode-directed pulse overlap add no independent information.","analogy_break":"Neural dynamics may be too nonlinear, nonstationary, partially observed, or non-normal for eigenmodes to evolve independently; electrodes may also miss the causal tissue, so a fitted mode can be a measurement artifact.","failure_condition":"No reproducible local window simultaneously provides mode identity, adequate spectral separation, acceptable residuals, and useful advance warning.","problem_falsifier":"If seizure onset and post-stimulation evolution are explained equally well by individual-channel states, with no stable distributed direction adding held-out value, the stated hidden-mode problem is absent.","intervention_falsifier":"On held-out recorded stimulation events, pulses aligned with the proposed suppressive contact pattern do not reduce the candidate mode or improve subsequent seizure-progression metrics versus matched rival, random-pattern, and no-pulse counterfactual estimates.","risks":["Retrospective confounding from clinician-selected stimulation","False reassurance from locally stable but globally unsafe dynamics","Mode swapping or ill-conditioned eigenvectors","Artifact-driven modes from referencing, movement, or preprocessing","Excess false alarms and unnecessary future stimulation","Poor transfer across patients, states, or electrode coverage"]},"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":0.74,"generator_notes":"Closed-book structural inference from the supplied packet; empirical neuroscience claims are hypotheses or inferences, and prior art was not searched."}