{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"invariant_mode_decomposition_design__art_aesthetics","trajectory_id":"R","attempt_index":0,"archetype_slug":"invariant_mode_decomposition_design","domain_slug":"art_aesthetics","decision":"CANDIDATE","problem_id":"runaway_coupled_visual_modes_in_feedback_art","causal_lever_id":"local_modal_control_of_visual_feedback","proposal":{"problem":"Feedback-based generative visual artworks can develop runaway motifs, oscillation, saturation, or compositional collapse even while individual parameter and average image metrics appear acceptable. Because rendering variables interact, artists tuning one parameter at a time cannot reliably identify the coupled direction causing the failure. HYPOTHESIS: locally stable visual-state modes exist within bounded operating regimes.","actors_substrate":["generative artists and creative technologists","curators and installation operators","viewers, including photosensitive viewers","feedback renderer, display hardware, and frame stream"],"observable_state":"Successive frames or versions represented by measurable color, luminance, spatial-frequency, motion, density, symmetry, and focal-distribution descriptors, plus the renderer parameters and local update rule.","consequence":"Unseen growing or weakly damped combinations can overwhelm intended focal hierarchy, produce visually monotonous collapse, or cross preset flicker and luminance safety limits.","affected_objective":"Preserve the artist's intended dynamic composition and variety while keeping exhibited output inside declared visual-safety and technical limits.","structural_mapping":[{"archetype_element":"Repeated coupled transformation","domain_realization":"The feedback renderer maps one frame-state and control vector to the next frame-state.","claim_kind":"CORPUS"},{"archetype_element":"Surface coordinates obscure invariant directions","domain_realization":"INFERENCE: separately named sliders and image descriptors may combine into recurring visual directions that the local renderer update amplifies or damps.","claim_kind":"INFERENCE"},{"archetype_element":"Scalar modal response","domain_realization":"Eigenvalue magnitude and phase estimate local growth, decay, persistence, or oscillation per visual mode.","claim_kind":"INFERENCE"},{"archetype_element":"Action-relevant mode selection","domain_realization":"Modes are retained by stability risk, aesthetic sensitivity, safety consequence, and out-of-sample reconstruction contribution—not magnitude alone.","claim_kind":"HYPOTHESIS"},{"archetype_element":"Residual and drift governance","domain_realization":"Frame reconstruction error, mode rotation, and spectral-gap erosion determine when the local modal account must be suspended and refit.","claim_kind":"CORPUS"}],"component_map":[{"component":"Transformation Scope","status":"adapted","domain_realization":"One fixed renderer build, display configuration, feedback rule, and bounded parameter neighborhood."},{"component":"State-Vector Definition","status":"adapted","domain_realization":"Standardized frame descriptors, selected latent controls, and one-step history needed to represent oscillation."},{"component":"Invariant Mode Basis","status":"direct","domain_realization":"Eigenvectors of the estimated one-step local state-transition Jacobian, traced back to visual descriptors."},{"component":"Modal Gain Spectrum","status":"direct","domain_realization":"Eigenvalues recording estimated one-step amplification, damping, and oscillation."},{"component":"Dominant Mode Selection Rule","status":"adapted","domain_realization":"Retain modes crossing preregistered stability, safety, aesthetic-sensitivity, or residual-reduction thresholds."},{"component":"Stable/Unstable Mode Partition","status":"direct","domain_realization":"Classify modes as decaying, marginal, growing, or oscillatory within the test window."},{"component":"Modal Intervention Map","status":"adapted","domain_realization":"Map permitted renderer-control changes to their projected effects on risky visual modes."},{"component":"Reconstruction Residual Check","status":"direct","domain_realization":"Compare held-out frame transitions with reconstructions from retained modes; inspect magnitude and structured leftovers."},{"component":"Mode Drift Monitor","status":"direct","domain_realization":"Track mode orientation, gain, ordering, and residuals during operation."},{"component":"Interpretation Scope Contract","status":"adapted","domain_realization":"Modes describe local render dynamics, not artistic meaning, beauty, or audience value."},{"component":"Mode-Coupling Register","status":"adapted","domain_realization":"Record near-degenerate, non-orthogonal, and perturbation-induced cross-effects."},{"component":"Local Linearization Window","status":"direct","domain_realization":"Declared bounds on controls, frame statistics, hardware, and elapsed time around the fitted operating point."},{"component":"Spectral Gap Threshold","status":"direct","domain_realization":"Minimum separation required before treating retained visual modes as individually stable targets."}],"mechanism_dispositions":[{"slug":"eigendecomposition_workflow","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Decompose the explicit estimated square one-step Jacobian and report conditioning.","counterfactual_removal":"Without it, growth and intervention cannot be expressed in invariant repeated-update directions."},{"slug":"modal_sensitivity_sweep","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Perturb permitted controls and modal coordinates to rank effects on intent proxies and safety limits.","counterfactual_removal":"The spectrum would not identify which controllable change affects consequential outcomes."},{"slug":"modal_stability_analysis","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Classify local frame-to-frame modes by growth, decay, and oscillation.","counterfactual_removal":"The proposal could not distinguish merely visible modes from runaway modes."},{"slug":"mode_shape_testing","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Empirical excitation is partly supplied by the sensitivity sweep; a separate physical mode-identification workflow is unnecessary when renderer transitions are observable.","counterfactual_removal":"No material change; empirical validation remains in the bounded perturbation test."},{"slug":"network_spectral_centrality_analysis","disposition":"incompatible","contribution_type":"NONE","adaptation_or_rejection":"The target is a dynamical visual-state operator, not node-importance ranking.","counterfactual_removal":"No change."},{"slug":"power_iteration_probe","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Dominant-only estimation would hide secondary unstable or safety-relevant modes; the pilot uses a manageable full Jacobian.","counterfactual_removal":"No change."},{"slug":"principal_component_analysis","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"High variance is not repeated-update gain or aesthetic consequence.","counterfactual_removal":"No change."},{"slug":"reduced_order_model","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"The first test diagnoses and controls the existing renderer; it does not require a surrogate simulator.","counterfactual_removal":"No change to the causal test."},{"slug":"residual_reconstruction_test","disposition":"selected_load_bearing","contribution_type":"TEST_DESIGN","adaptation_or_rejection":"Test retained modes on held-out frame transitions and reject structured residuals.","counterfactual_removal":"A misleading low-order modal explanation could pass unchecked."},{"slug":"singular_value_decomposition","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"SVD may diagnose transient amplification but its paired singular directions are not invariant repeated-update modes; reserve it for a non-normality audit.","counterfactual_removal":"The primary eigenmode chain remains intact."},{"slug":"spectral_decomposition_report","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Document descriptor loadings, conditioning, couplings, residuals, and prohibited semantic interpretations.","counterfactual_removal":"Operators could mistake local dynamical modes for objective aesthetic meanings."},{"slug":"spectral_gap_monitor","disposition":"selected_supporting","contribution_type":"OPERATIONAL","adaptation_or_rejection":"Monitor retained/dropped separation, mode rotation, and residual growth during exhibition.","counterfactual_removal":"The controller could continue after its modal targets lose identity or validity."}],"causal_chain":["Controlled frame-state perturbations estimate a local one-step renderer transformation.","Decomposition reveals coupled directions and their local gains.","Stability and outcome sensitivity identify growing modes that threaten aesthetic intent or safety constraints.","The intervention map selects bounded control changes projected to damp those modes.","Held-out residual tests check whether retained modes explain consequential transitions.","Gap, drift, and residual monitors suspend control when the local modal account ceases to hold."],"baseline":"Artists or operators tune one slider at a time from sampled renders while watching aggregate color, luminance, motion, and flicker metrics.","nearest_rival":"A black-box optimizer adjusts renderer controls directly against aggregate aesthetic and safety scores without an interpretable modal model.","authority_safety":{"affected_parties":["artist whose intent may be altered","installation operators","viewers, especially photosensitive viewers","venue responsible for display safety"],"decision_authority":"The artist approves aesthetic objectives and permitted controls; the venue safety lead sets non-waivable display limits and may halt exhibition.","authorized_first_step":"Offline, replay one recorded work segment and apply small reversible perturbations inside artist-approved controls; compare modal damping against baseline tuning on held-out seeds, with no public exposure.","excluded_actions":["autonomous live changes outside approved control bounds","treating modes as measures of beauty or artistic worth","overriding flicker, luminance, accessibility, or hardware limits","training on identifiable audience-response data without consent"],"halt_rollback":"Stop perturbations and restore the recorded baseline if any safety limit is approached, residual exceeds its preregistered budget, the spectral gap falls below threshold, modes rotate beyond tolerance, or the artist judges intent materially distorted."}},"negative_tests":{"strongest_counterevidence":"The renderer may be strongly nonlinear, deliberately nonstationary, or observer-dependent, so a local Jacobian can become obsolete before it enables useful control; apparently dominant modes may also be artifacts of chosen descriptors.","analogy_break":"Mechanical or mathematical modes have operator-defined identity, whereas aesthetic qualities depend on representation, context, and viewers. A visual eigenmode therefore cannot be equated with meaning, merit, or a universally perceived Gestalt.","failure_condition":"The approach fails if no operating window yields repeatable modes, acceptable conditioning, a sufficient spectral gap, and held-out residuals below the consequence-weighted budget.","problem_falsifier":"Under repeated bounded trials, coordinate-level changes independently predict failures and no repeatable coupled growing, marginal, or oscillatory direction improves prediction beyond ordinary metrics.","intervention_falsifier":"Repeatable risky modes exist, but mode-targeted bounded controls do not outperform one-slider tuning or the black-box rival on held-out intent proxies, safety margins, recovery time, and unintended cross-effects.","risks":["Descriptor choices may encode the analyst's taste as if it were structural fact.","Near-degenerate or non-normal modes may swap, mix, or exhibit transient growth missed by eigenvalues.","Automated damping may sterilize intentional instability or surprise.","Offline safety metrics may not capture all viewer responses.","Monitoring latency could allow a rapid excursion before halt."]},"null_rationale":null,"classification":{"candidate_kind":"MECHANISM_COMPOSITION","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 is limited to feedback-based generative art with an explicit or estimable local transition; no novelty or empirical effectiveness is claimed."}