{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"invariant_mode_decomposition_design__economics_finance","arm":"PROPOSAL_FIRST","candidate_id":"cand_imdd_finance_001","hypothesis_id":null,"version":0,"title":"Mode-Aware Early Warning for Coupled Bank Liquidity Stress","problem":"A financial-system supervisor monitors each bank's reserve shortfall, payment delay, wholesale-funding spread, collateral encumbrance, and deposit outflow separately. Stress can nevertheless propagate through a repeatable combination of these variables across banks. Institution-level thresholds and system averages may remain acceptable while that combined direction becomes weakly damped or self-amplifying, causing supervisory attention to arrive only after recognizable funding or settlement disruption appears.","actors":["Financial-system liquidity supervisors","Payment-system operator","Supervised banks and their treasury desks","Bank depositors and payment customers indirectly exposed to disruption"],"observable_state":"At each daily or weekly interval, a state vector records standardized, aggregated bank-level liquidity gaps, queued-payment shares, short-term funding spreads, collateral encumbrance, deposit-flow rates, and bilateral funding exposures. The observable warning state is a persistent or growing projection onto a coupled mode even when no individual indicator breaches its ordinary threshold.","consequence":"Supervisors may investigate conspicuous institutions or isolated metrics while missing the cross-bank combination along which liquidity strain is propagating, delaying scenario escalation and preparedness review.","affected_objective":"Timely, traceable identification of system-level liquidity-stress directions while limiting false escalation and preserving institution-level due process.","intervention":"Build a shadow-mode dashboard from the locally estimated state-transition operator. Decompose the operator, classify growing and weakly damped modes, map each action-relevant mode back to contributing banks and variables, and use a sensitivity sweep to rank nonbinding supervisory responses such as requesting clarification, refining a stress scenario, or scheduling a liquidity-preparedness review. Continue ordinary dashboards in parallel, expose reconstruction residuals, and suspend modal interpretation when the spectral gap, conditioning, drift, or out-of-sample error violates preset limits.","structural_mapping":[{"archetype_element":"Transformation Scope","domain_realization":"A locally estimated one-period transition from the current cross-bank liquidity state to the next interval's state under a documented market and policy regime."},{"archetype_element":"State-Vector Definition","domain_realization":"Standardized liquidity gaps, payment queues, funding spreads, collateral encumbrance, deposit flows, and exposure measures for the included banks."},{"archetype_element":"Invariant Mode Basis","domain_realization":"Cross-bank combinations of liquidity variables that the estimated transition approximately preserves as directions."},{"archetype_element":"Modal Gain Spectrum","domain_realization":"Each mode's estimated persistence, damping, amplification, or oscillation over successive intervals."},{"archetype_element":"Stable/Unstable Mode Partition","domain_realization":"Modes classified as decaying, marginal, oscillatory, or growing within the declared local regime."},{"archetype_element":"Dominant Mode Selection Rule","domain_realization":"A mode becomes review-relevant only when its estimated gain, consequence-weighted sensitivity, and stability across resamples exceed preset thresholds."},{"archetype_element":"Modal Intervention Map","domain_realization":"A traceable mapping from each selected mode to contributing variables and to bounded supervisory information-gathering or scenario-review actions."},{"archetype_element":"Reconstruction Residual Check","domain_realization":"Out-of-sample comparison of observed state changes with changes reconstructed from retained modes, including inspection for structured residuals."},{"archetype_element":"Mode Drift Monitor","domain_realization":"Scheduled checks for rotation, reordering, gain changes, and loss of spectral separation."},{"archetype_element":"Interpretation Scope Contract","domain_realization":"The dashboard is an aggregate, local early-warning aid and not causal proof, a bank-viability verdict, or a basis for automatic enforcement."}],"mechanism_mapping":[{"mechanism_slug":"eigendecomposition_workflow","role":"Factor the documented local transition matrix into a complete set of candidate liquidity-stress directions and their scalar gains, subject to conditioning diagnostics.","counterfactual_removal":"Without the decomposition, the proposal reverts to coordinate-level indicators and cannot identify the combinations preserved or amplified by the transition."},{"mechanism_slug":"modal_stability_analysis","role":"Classify modes as growing, decaying, marginal, or oscillatory within the fitted operating window.","counterfactual_removal":"Without stability classification, the spectrum describes directions but cannot distinguish transient prominence from a direction that persists or grows under repeated transition."},{"mechanism_slug":"modal_sensitivity_sweep","role":"Perturb modal coordinates and candidate supervisory responses in the local model to rank which modes are consequential for payment disruption and which bounded responses address them with limited cross-effects.","counterfactual_removal":"Without the sweep, large-gain modes could be mistaken for decision-relevant modes, and cross-mode consequences of a proposed response would remain hidden."},{"mechanism_slug":"residual_reconstruction_test","role":"Determine whether retained modes reproduce held-out transitions within an explicit error budget and reveal structured behavior left outside the modal model.","counterfactual_removal":"Without residual testing, compression could discard a small but consequential liquidity pattern while still producing a persuasive spectrum."},{"mechanism_slug":"spectral_gap_monitor","role":"Track whether retained modes remain separated and identifiable as new observations arrive.","counterfactual_removal":"Without gap and drift monitoring, a once-useful basis could remain operational after modes rotate, merge, or swap identity."},{"mechanism_slug":"spectral_decomposition_report","role":"Document variable loadings, gains, couplings, residuals, permitted interpretations, and prohibited uses for supervisors who did not fit the model.","counterfactual_removal":"Without the report, tentative aggregate modes could be treated as independent causal facts or institution-level risk judgments."}],"causal_chain":["Coupled liquidity variables evolve through a locally estimable cross-bank transition rather than independently.","Decomposition exposes combinations that the transition approximately preserves and assigns each a gain.","Stability analysis identifies a growing or weakly damped combination before all of its constituent indicators breach separately.","Mapping the mode back to original variables makes its contributing observations and institutions inspectable.","Sensitivity analysis distinguishes consequential modes and bounded review actions from modes that are merely numerically prominent.","A supervisor can prioritize clarification or scenario review around the coupled direction while retaining ordinary controls and human judgment.","Residual, conditioning, gap, and drift checks determine whether continued modal use remains warranted."],"baseline":"Institution-by-institution threshold dashboards, systemwide averages, and fixed stress scenarios reviewed separately by metric; escalation occurs when a named bank or indicator crosses a predefined limit.","nearest_rivals":["Dynamic factor or principal-component monitoring of correlated liquidity indicators, which extracts common variation but does not directly classify amplification under an explicit transition.","Network centrality ranking of banks by funding exposures, which identifies structurally prominent institutions but does not identify multivariable growing directions.","Conventional multivariate liquidity stress testing, which evaluates specified shocks but may not discover the locally persistent combinations toward which observed dynamics are already moving."],"remaining_contrastive_claim":"The candidate's testable contrast is that supervisory review is prioritized by the estimated gain and consequence sensitivity of traceable cross-bank state-transition modes, with residual and mode-validity gates, rather than by indicator magnitude, variance explained, node prominence, or a fixed shock list alone.","authority_safety":{"decision_authority":"The designated financial-system liquidity supervisor retains authority to interpret alerts, request information, select scenarios, and initiate any formal supervisory process under existing rules; the model has no independent authority.","authorized_first_step":"Run a retrospective, access-controlled shadow analysis on previously collected aggregate supervisory data and compare its alerts with the ordinary dashboard without contacting banks or changing supervisory treatment.","excluded_actions":["Automatic enforcement, sanctions, capital changes, or liquidity requirements","Automatic public disclosure or institution-level labeling","Trading, market intervention, or payment throttling","Use of modal loadings as proof of causation, misconduct, or insolvency","Collection of new institution-level data outside existing authority","Suppression of ordinary monitoring because a retained-mode reconstruction appears adequate"],"halt_rollback":"Stop using the modal output and revert to the ordinary dashboard if held-out reconstruction exceeds its preset error budget, modes are ill-conditioned, the retained-versus-discarded spectral gap falls below threshold, mode identities drift beyond tolerance, or analysts cannot trace an alert to verified inputs. Preserve the shadow outputs for audit but remove them from the active review queue."},"negative_tests":{"strongest_counterevidence":"Across multiple held-out calm and stressed periods, modal alerts are unstable under reasonable variable scaling and sampling choices, arrive no earlier than ordinary thresholds, or concentrate on directions with no reproducible relation to payment disruption or liquidity escalation.","problem_falsifier":"The inferred problem is unsupported if consequential liquidity deterioration is adequately explained by one directly observable bank or indicator, or if no cross-variable direction persists long enough to provide a usable monitoring window.","intervention_falsifier":"The intervention is unsupported if the fitted transition cannot reconstruct held-out state changes within the declared budget, if action-relevant modes lack stable identities or a usable spectral gap, or if mode-based review prioritization fails a preregistered comparison with the ordinary baseline.","risks":["Spurious modes caused by scaling, missing data, reporting conventions, or simultaneous policy changes","Non-normal or ill-conditioned dynamics that make eigenvectors fragile despite apparently stable eigenvalues","Regime shifts that invalidate the local transition before scheduled re-estimation","Feedback if supervised institutions later infer and adapt to the alert logic","Stigmatization of banks whose variables load strongly on an aggregate mode","False reassurance from low total reconstruction error that hides a small consequential residual","Model complexity obscuring accountable supervisory reasoning"]},"next_evidence_step":"Using one frozen historical dataset already authorized for supervisory analysis, define the state vector and operating window in advance; fit the transition on an earlier segment; compute modes, conditioning, gains, sensitivity rankings, and retained-mode residuals; then evaluate on one untouched later segment whether mode identities and gaps remain stable, whether reconstruction stays within a preset budget, and whether shadow alerts precede or complement documented ordinary-dashboard escalations. No live decisions or external contacts occur in this step.","prior_art_status":"UNSEARCHED","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":["Initial proposal translating invariant-mode decomposition into coupled bank-liquidity early warning."],"operational_changes":["Bounded the first implementation to retrospective shadow analysis with explicit halt conditions."],"evidence_changes":[],"claim_changes":["Restricted the claim to a falsifiable prioritization contrast and made no claim of novelty, prevalence, demand, or effect size."]}}