{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp05_complete_proposal_portfolio20_20260803","cell_id":"invariant_mode_decomposition_design__information_theory","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"cand01_mode_targeted_predictive_telemetry_refresh","proposal_index":1,"version":0,"title":"Mode-Targeted Refresh Coding for Drift-Prone Predictive Telemetry","problem":"A predictive telemetry codec reconstructs several correlated sensor streams from prior decoded values. After packet loss or quantization error, the codec's predictor can carry reconstruction error into later frames. Per-stream error summaries can remain acceptable even while a particular combination of stream errors persists or grows, because the predictor acts on coupled directions rather than on streams independently. Bit-allocation and refresh policies that rank individual streams by variance, entropy, or current error can therefore neglect the joint error direction most likely to corrupt later reconstructions.","actors":["Telemetry codec engineer who defines the predictor and bit allocation","Network operator who sets the transmission budget and reliability policy","Receiving system that reconstructs the sensor state","Downstream application owner whose decisions depend on the reconstruction"],"observable_state":"Following bounded loss or quantization disturbances, reconstruction-error vectors repeatedly align with a similar cross-stream pattern and decay more slowly than their individual components suggest. The pattern appears in lagged error trajectories or structured reconstruction residuals, while average per-stream distortion remains within its ordinary range.","consequence":"A coupled reconstruction error can survive across frames, consume later correction capacity, and distort a downstream decision-relevant combination of measurements even though no individual stream is an obvious refresh priority.","affected_objective":"Maintain decision-relevant reconstruction fidelity and recovery from channel disturbances under a fixed telemetry bit budget.","intervention":"Define a local error-transition operator F for the deployed predictor, where e(t+1) is approximated by F e(t) after a small reconstruction disturbance. Decompose F, trace each mode back to its contributing sensor variables, and classify modes by persistence or growth. In replay, perturb modal coordinates and candidate refresh controls to rank modes by their effect on a predeclared downstream distortion measure. Reallocate a bounded portion of existing refresh and protection bits to transmit intra-coded correction coefficients for the highest-leverage weakly damped or growing modes: the encoder projects the current source state onto each selected joint direction, and the decoder uses the protected coefficient to reset its state along that direction. Retain ordinary coding for the remaining coordinates. Permit deployment only while held-out reconstruction residuals, modal conditioning, spectral separation, and operating-regime checks remain within declared tolerances.","structural_mapping":[{"archetype_element":"Transformation Scope","domain_realization":"The one-frame propagation of small encoder-decoder reconstruction errors under a fixed predictive codec and specified channel-loss regime."},{"archetype_element":"State-Vector Definition","domain_realization":"A synchronized vector of reconstruction errors across the telemetry streams, expressed in fixed units and evaluated at frame boundaries."},{"archetype_element":"Invariant Mode Basis","domain_realization":"Joint cross-stream error directions that the local transition approximately preserves."},{"archetype_element":"Modal Gain Spectrum","domain_realization":"Each mode's one-frame error multiplier, including sign or phase when oscillation is possible."},{"archetype_element":"Stable/Unstable Mode Partition","domain_realization":"Modes classified as decaying, marginal, oscillatory, or growing inside the declared disturbance and channel regime."},{"archetype_element":"Modal Intervention Map","domain_realization":"A mapping from protected refresh coefficients and bit allocations to reductions in particular modal error coordinates."},{"archetype_element":"Reconstruction Residual Check","domain_realization":"Held-out difference between actual codec error trajectories and trajectories reconstructed from the retained modal model, assessed for both magnitude and structure."},{"archetype_element":"Mode Drift Monitor","domain_realization":"Periodic comparison of mode directions, gains, ordering, and retained-to-discarded spectral separation after predictor, source, or channel changes."},{"archetype_element":"Interpretation Scope Contract","domain_realization":"The decomposition is treated as a local description of error propagation, not as a global channel model or evidence that a mode has semantic meaning."}],"mechanism_mapping":[{"mechanism_slug":"eigendecomposition_workflow","role":"Factor the explicit local error-transition matrix into joint error directions and their one-frame gains.","counterfactual_removal":"Without the decomposition, refresh decisions revert to named streams or surface errors and cannot identify a preserved coupled direction."},{"mechanism_slug":"modal_stability_analysis","role":"Classify whether each error mode decays, persists, oscillates, or grows within the fitted operating window.","counterfactual_removal":"Without stability classification, a currently small but weakly damped or growing mode has no principled priority over a large transient mode."},{"mechanism_slug":"modal_sensitivity_sweep","role":"Perturb modal coordinates and feasible refresh coefficients in replay to rank their influence on the downstream distortion measure and record cross-mode effects.","counterfactual_removal":"Without the sweep, allocation would follow eigenvalue magnitude alone and could protect a persistent mode that has little consequence while missing a lower-gain, decision-relevant mode."},{"mechanism_slug":"residual_reconstruction_test","role":"Test whether the retained modes reproduce held-out error propagation and whether omitted residuals contain consequential structure.","counterfactual_removal":"Without residual testing, an apparently compact modal account could hide discarded error behavior that invalidates the refresh policy."},{"mechanism_slug":"spectral_gap_monitor","role":"Detect narrowing separation, rotation, or reordering that makes the selected modal basis unsafe to reuse.","counterfactual_removal":"Without monitoring, a once-valid refresh basis could continue allocating bits after the source, predictor, or channel regime changes."},{"mechanism_slug":"spectral_decomposition_report","role":"Record variable loadings, conditioning, couplings, thresholds, and the local interpretation limits supplied to codec and operations reviewers.","counterfactual_removal":"Without an explicit scope contract, operators could treat approximate modes as independent, stationary, or causal facts."}],"causal_chain":["Predictive decoding couples current reconstruction errors across telemetry streams and carries them into the next frame.","Repeated disturbances reveal that some joint error directions are preserved or damped less strongly than individual-stream summaries indicate.","Decomposition of the local error-transition operator exposes those directions and assigns each a scalar propagation gain.","Stability analysis separates quickly decaying modes from weakly damped, marginal, oscillatory, or growing modes.","Outcome-based sensitivity testing identifies which persistent modes materially move the declared downstream distortion measure and which refresh controls can reduce them.","Protected refresh coefficients aligned with selected modes reset several coordinated error components with a single joint correction while the total bit budget remains fixed.","Held-out residual and drift checks determine whether the modal policy remains an adequate local representation; failure suspends the policy and restores the baseline allocation."],"baseline":"Use the existing predictive codec and allocate refresh or protection bits per stream according to current reconstruction error, variance, entropy contribution, or a fixed round-robin schedule, without modeling coupled error propagation.","nearest_rivals":["A rate-matched periodic full-state keyframe policy that resets every stream without modal selection","Per-stream adaptive refresh based on each stream's recent distortion or innovation magnitude","Direct end-to-end rate-distortion optimization over candidate bit allocations without an explicit modal representation","PCA-based transform coding that protects high-variance source directions rather than directions defined by error propagation"],"remaining_contrastive_claim":"The proposal's testable distinction is the allocation criterion: scarce refresh protection is assigned to joint directions selected by their propagation dynamics and decision sensitivity, rather than to individual-stream magnitude, source variance alone, or uniform reset timing. Whether that criterion improves the declared objective remains to be tested.","authority_safety":{"decision_authority":"The codec owner may authorize an offline replay and a shadow-mode comparison. Any live bit reallocation requires joint approval from the network operator and downstream application owner under the existing transmission budget.","authorized_first_step":"Fit and evaluate the modal error model on one bounded archived telemetry trace, then compare simulated mode-targeted refresh against rate-matched baselines on held-out disturbance segments without changing live transmissions.","excluded_actions":["Changing the live predictor, channel code, or transmission rate during the first evidence step","Dropping safety-critical streams or mandatory refreshes to fund modal coefficients","Treating a mode as causal or semantically meaningful solely from its loadings","Deploying outside the fitted disturbance, source, predictor, or channel regime","Continuing automated modal allocation after a residual, conditioning, gap, or drift limit is breached"],"halt_rollback":"If held-out residuals contain declared consequential structure, eigenvectors are too ill-conditioned for stable interpretation, selected modes drift beyond tolerance, the spectral gap is inadequate, or any protected stream loses required coverage, halt the modal policy and restore the unchanged baseline allocation table."},"negative_tests":{"strongest_counterevidence":"Across held-out loss and quantization disturbances, error trajectories do not repeatedly align with stable joint directions, or a rate-matched per-stream or periodic-keyframe policy matches the candidate on both downstream distortion and recovery behavior.","problem_falsifier":"A memoryless or stream-independent error model predicts held-out post-disturbance trajectories as adequately as the coupled transition model, with no persistent structured residual attributable to joint propagation.","intervention_falsifier":"A reproducible, decision-relevant persistent mode exists, but refreshing its coefficient under the same bit budget does not reduce the predeclared held-out distortion or recovery criterion relative to the strongest rate-matched rival.","risks":["The fitted operator may reflect the archived disturbance pattern rather than the deployed channel regime.","A non-normal or nearly defective transition matrix may yield fragile eigenvectors despite apparently distinct gains.","Mode-aligned coefficients may be expensive to quantize or protect, erasing their allocation advantage.","Correcting one mode may excite another through nonlinear predictor behavior or saturation.","Reallocating bits could reduce minimum protection for a low-variance but safety-critical stream.","Feedback or synchronization errors could cause encoder and decoder modal states to disagree."]},"next_evidence_step":"Using one archived multistream trace, inject a bounded, predeclared set of packet-loss and quantization disturbances into a deterministic codec replay. Fit F only on the training portion; estimate its modes, conditioning, gains, and sensitivity ranking; choose retained modes using declared stability, consequence, and residual criteria; and simulate protected modal refresh under exactly the baseline bit budget. On held-out segments, compare downstream distortion, recovery trajectory, per-stream minimum fidelity, residual structure, and basis stability against periodic keyframes, per-stream adaptive refresh, and direct rate-matched allocation. The result authorizes at most a shadow-mode test, not live control.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Not assessed because no prior proposal content was supplied in the sealed context; the candidate is characterized solely by its mode-targeted correction of coupled predictive-codec error propagation.","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}