{"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":"cand04_modal_voice_command_codebook","proposal_index":4,"version":0,"title":"Modal Codebook Redesign for Relayed Voice Commands","problem":"A control service transmits a small vocabulary of spoken commands through noisy radio, automatic transcription, and occasional human relay. Existing audits score each command separately, but every decoding or relay step applies a coupled confusion transformation across the whole command alphabet. A contrast involving several commands can persist, alternate, or decay slowly across repeated relays even when each command's one-hop accuracy appears acceptable. Renaming the command with the worst individual error rate may therefore leave the structurally persistent and operationally consequential confusion direction intact.","actors":["Command issuer who selects a message from the approved vocabulary","Radio or transcription system that carries and decodes the message","Receiving operator who interprets and may relay the decoded command","Communications-protocol engineer who maintains the codebook","Safety authority responsible for command and confirmation policy"],"observable_state":"In controlled relay trials, decoded-command distributions repeatedly align with the same cross-command error contrast after successive transmission and interpretation steps. The estimated confusion operator has a nontrivial mode with gain near the persistence boundary or an oscillatory sign or phase, and held-out substitution errors retain that mode's pattern despite acceptable coordinate-level accuracy summaries.","consequence":"A hazardous command substitution can remain likely through relay or read-back stages because the communication process preserves the underlying confusion pattern rather than independently correcting each mistaken command.","affected_objective":"Reduce consequential command substitutions across bounded noisy relay conditions while preserving transmission time, vocabulary coverage, and minimum per-command intelligibility.","intervention":"Estimate a square command-confusion operator C from controlled transmissions of the fixed vocabulary through the specified radio, transcription, and human-relay path. Decompose C into invariant error directions and scalar gains, excluding the trivial probability-conservation mode from intervention ranking. Classify nontrivial modes as rapidly decaying, slowly decaying, persistent, or oscillatory under repeated relay. In simulation and controlled trials, perturb candidate codeword assignments, acoustic forms, and short confirmation symbols to measure their effect on action-specific substitution loss and on other modes. Redesign only the smallest approved subset of command tokens and confirmation prompts needed to damp the highest-consequence persistent modes: commands loading with opposing signs on a hazardous mode receive more distinguishable signal forms or a mode-specific disambiguation check. Validate the revised codebook on held-out speakers, noise conditions, relay depths, and transcription versions; retain the existing codebook unless residual, conditioning, minimum-accuracy, and drift requirements pass.","structural_mapping":[{"archetype_element":"Transformation Scope","domain_realization":"One complete encode, transmit, decode, and optional human-interpretation step over a fixed command alphabet under declared channel and operating conditions."},{"archetype_element":"State-Vector Definition","domain_realization":"A probability or signed-error vector over the approved command classes at a relay boundary."},{"archetype_element":"Invariant Mode Basis","domain_realization":"Cross-command error contrasts whose direction is approximately preserved when the confusion operator is applied."},{"archetype_element":"Modal Gain Spectrum","domain_realization":"Each nontrivial eigenvalue describes how strongly its command-confusion pattern decays, persists, or alternates over another relay."},{"archetype_element":"Stable/Unstable Mode Partition","domain_realization":"Nontrivial modes are classified as rapidly decaying, slowly decaying, marginally persistent, or oscillatory within the tested stationary regime."},{"archetype_element":"Dominant Mode Selection Rule","domain_realization":"A mode is actionable only when its persistence and command-substitution consequence exceed declared thresholds and its eigenvector is sufficiently conditioned for interpretation."},{"archetype_element":"Modal Intervention Map","domain_realization":"Candidate token changes and confirmation symbols are mapped to changes in specific modal gains and to command-level safety, latency, and intelligibility outcomes."},{"archetype_element":"Mode-Coupling Register","domain_realization":"The protocol records when damping one confusion mode rotates, strengthens, or creates another, including near-degenerate and non-normal interactions."},{"archetype_element":"Reconstruction Residual Check","domain_realization":"Observed held-out relay distributions are compared with modal predictions, and unexplained substitutions are inspected for consequential structure."},{"archetype_element":"Mode Drift Monitor","domain_realization":"The confusion basis and spectrum are re-estimated after vocabulary, radio, transcription, workforce, acoustic-environment, or relay-procedure changes."},{"archetype_element":"Interpretation Scope Contract","domain_realization":"Modes describe bounded communication behavior and are not treated as cognitive traits, speaker deficiencies, causal explanations, or guarantees outside the tested relay conditions."}],"mechanism_mapping":[{"mechanism_slug":"eigendecomposition_workflow","role":"Compute the complete invariant-mode basis and gain spectrum of the explicit square command-confusion operator.","counterfactual_removal":"Without eigendecomposition, the audit remains organized by individual commands and cannot identify an error contrast preserved across successive relays."},{"mechanism_slug":"modal_stability_analysis","role":"Interpret each nontrivial gain relative to the relay persistence boundary and classify slowly decaying, persistent, and oscillatory confusion patterns.","counterfactual_removal":"Without stability classification, a small present error that survives repeated relay cannot be distinguished from a larger error that self-corrects after one step."},{"mechanism_slug":"modal_sensitivity_sweep","role":"Perturb codeword assignments and confirmation controls to rank their effects on consequential substitutions, modal gains, latency, and cross-mode behavior.","counterfactual_removal":"Without the sweep, redesign would follow eigenvalue magnitude or acoustic intuition alone and could target a persistent but harmless mode or displace error into another hazardous mode."},{"mechanism_slug":"residual_reconstruction_test","role":"Compare held-out multi-relay outcomes with modal reconstructions and inspect omitted residuals for structured command substitutions.","counterfactual_removal":"Without residual testing, a compact modal model could appear adequate while leaving a low-frequency but consequential substitution outside the retained modes."},{"mechanism_slug":"spectral_gap_monitor","role":"Track whether the governed confusion modes remain separated and directionally stable after channel, vocabulary, or decoder changes.","counterfactual_removal":"Without monitoring, the protocol could retain redesigned tokens after the targeted mode rotates, merges with another mode, or ceases to be identifiable."},{"mechanism_slug":"spectral_decomposition_report","role":"Document command loadings, gains, conditioning, intervention sensitivities, couplings, tested conditions, and prohibited interpretations for protocol approval.","counterfactual_removal":"Without a bounded report, operators could interpret tentative mathematical modes as independent error causes or deploy token changes beyond the evaluated channel."}],"causal_chain":["An intended command enters a bounded alphabet and passes through a noisy encoding-decoding and interpretation step.","That step redistributes probability across all command classes according to a coupled confusion operator rather than producing independent per-command errors.","Repeated relay preserves some cross-command error directions more strongly than coordinate-level accuracy summaries reveal.","Eigendecomposition exposes those directions, while stability analysis identifies which nontrivial patterns decay slowly, persist, or alternate.","Outcome sensitivity distinguishes operationally consequential modes from mathematically prominent but harmless ones and identifies token or confirmation changes with leverage over them.","Selected command tokens or checks are redesigned to reduce the gain of the hazardous confusion modes at the source-codebook level.","Held-out relay trials test whether the redesigned alphabet actually damps those patterns without violating minimum intelligibility or latency requirements.","Residual and drift checks withdraw the modal interpretation when substitutions no longer follow the fitted operator."],"baseline":"Maintain the existing command vocabulary and review commands independently using one-hop accuracy, overall word-error rate, operator reports, or ad hoc renaming of the single most-confused token, with a uniform read-back rule where required.","nearest_rivals":["Universal closed-loop confirmation or read-back for every command","Pairwise phonetic-distance maximization across the complete command vocabulary","Per-command renaming based on individual confusion rates","Direct end-to-end optimization of expected command-substitution cost without an interpretable modal model","Replacement of automatic transcription with human-only decoding or a higher-bandwidth digital command interface"],"remaining_contrastive_claim":"The candidate's testable distinction is that codewords and checks are selected to damp consequential invariant patterns of a repeated confusion transformation, rather than to improve commands independently, maximize pairwise distance alone, or confirm every message uniformly. Whether this criterion improves held-out relay behavior under matched vocabulary, latency, and confirmation budgets remains to be tested.","authority_safety":{"decision_authority":"The communications-protocol engineer may conduct offline analysis and controlled trials. The safety authority alone may approve a vocabulary or confirmation-policy change, with operational owners responsible for training and cutover approval.","authorized_first_step":"Estimate and analyze the confusion operator using a frozen copy of the current vocabulary, approved recordings or consenting test participants, simulated channel impairments, and no live operational traffic.","excluded_actions":["Changing a live command token, read-back requirement, or transcription configuration during the first evidence step","Removing an existing mandatory confirmation or safety check to satisfy the latency budget","Testing ambiguous commands during real incidents or operational control activity","Interpreting modal loadings as evidence about an individual's competence, language ability, or protected characteristics","Deploying a codebook that lacks version signaling, operator training, backward compatibility, and an explicit cutover plan","Using the decomposition when eigenvectors are ill-conditioned, the operator is materially nonstationary, or held-out residuals remain consequential"],"halt_rollback":"Retain or restore the existing vocabulary and confirmation policy if any revised token lowers a mandatory per-command floor, increases a declared hazardous substitution, creates incompatible codebook versions, breaches the latency limit, produces structured held-out residuals, or targets a mode that is not stable across approved speakers and channel conditions."},"negative_tests":{"strongest_counterevidence":"A rate-matched universal confirmation rule, pairwise code-distance design, or direct substitution-cost optimizer performs at least as well on every declared held-out safety and latency criterion, while the estimated invariant modes are poorly conditioned or fail to persist across speakers, channels, or relay depths.","problem_falsifier":"Multi-relay outcomes are adequately explained by independent one-hop command errors, with no reproducible nontrivial confusion direction that persists, alternates, or predicts consequential substitutions beyond coordinate-level measures.","intervention_falsifier":"A stable and consequential confusion mode is identified, but redesigning its high-loading tokens or adding its disambiguation check does not reduce the predeclared held-out substitution criterion relative to the strongest budget-matched rival.","risks":["Sparse observations for rare hazardous commands may make the confusion operator unreliable.","A non-normal or nearly defective operator may yield fragile eigenvectors and misleading command loadings.","A token change that damps one mode may create a new confusion pattern elsewhere in the alphabet.","Simulated noise, recorded speech, or consenting test participants may not represent the authorized operational regime.","Different accents, speech patterns, equipment, or transcription versions may rotate the modes and create unequal performance.","Mixed old and new codebooks during transition could introduce errors not represented in either fitted operator.","Optimizing repeated relay behavior may be irrelevant where a reliable closed-loop digital acknowledgement already prevents action on an ambiguous command."]},"next_evidence_step":"Freeze one command vocabulary, relay procedure, transcription version, channel-impairment set, latency budget, command-specific minimums, and substitution-cost table. Use a bounded controlled corpus divided by speaker, noise condition, and relay sequence into fitting and held-out partitions. Estimate C only from the fitting partition, compute its complete spectrum and eigenvector conditioning, exclude the conservation mode, and preselect one bounded token-and-confirmation redesign using declared persistence, consequence, coupling, and residual rules. On held-out trials, compare multi-relay substitution patterns, minimum per-command accuracy, latency, residual structure, modal gain, and basis stability against the unchanged vocabulary, universal confirmation, per-command renaming, pairwise-distance design, and direct substitution-cost optimization. The result may authorize only a non-operational shadow trial.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 addressed temporal numerical reconstruction-error propagation in a predictive telemetry codec and transmitted modal refresh coefficients. Proposal 2 addressed protected-information leakage through correlated analytics outputs and filtered high-gain disclosure directions. Proposal 3 addressed weak recoverability directions in distributed scientific-array storage and changed persistent parity projections and placement. This proposal addresses semantic substitution across a discrete spoken-command alphabet and changes the codebook and confirmation structure to damp persistent modes of a stochastic confusion operator. Its communicated object, actors, consequence, transformation, intervention, evidence, and adoption setting are distinct from each of proposals 1, 2, and 3, and it requires none of their telemetry, privacy-release, or archival-storage systems.","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}