{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp05_complete_proposal_portfolio20_20260803","cell_id":"invariant_mode_decomposition_design__literature_literary_theory","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"cand_modal_ambiguity_checkpoint_001","proposal_index":1,"version":0,"title":"Modal Ambiguity Checkpoint for Unreliable-Narrator Revision","problem":"During successive revisions of a novel with an intentionally unreliable narrator, editors may improve individually visible features such as causal clarity, chronology, consistency, and lexical accessibility while repeatedly amplifying a coupled direction that makes the narrator appear more reliable. Passage-by-passage review can miss this transformation because no single edit or feature necessarily leaves the author's intended ambiguity range.","actors":["Author, who defines the intended range of narratorial ambiguity and retains final artistic authority","Developmental editor, who proposes and groups revisions","Editorial analyst or trained coders, who represent revision states and estimate the transformation","Blinded readers, who assess narrator reliability and reconstructability without seeing the modal analysis"],"observable_state":"For each sampled passage at each revision checkpoint, coders record a state vector comprising causal-link explicitness, narrator certainty, retained internal contradiction, temporal-order signaling, lexical-register distance between narrator and implied author, and focalization leakage. Consecutive versions provide paired state vectors. Within a declared manuscript phase, a locally fitted transition matrix represents how an editorial round maps centered pre-revision states to post-revision states. Blinded-reader ratings of perceived narrator reliability, productive uncertainty, and event-sequence reconstructability provide separate outcome measures.","consequence":"A manuscript may leave the author's declared ambiguity envelope through the cumulative action of mutually reinforcing edits even though each coordinate-level change passes ordinary review. The resulting version can foreclose intended competing interpretations or make uncertainty seem like an error that later revisions continue to remove.","affected_objective":"Keep narrator reliability and interpretive ambiguity within an author-declared envelope while permitting improvements in reader reconstruction of the underlying events.","intervention":"Insert a reversible modal checkpoint before the author accepts a substantial editorial batch. Estimate the within-phase revision transformation, decompose it into coupled directions and gains, and identify any persistent or growing mode associated with movement outside the declared ambiguity envelope. Project the proposed edit batch into those modal coordinates, then perform an edit-level sensitivity sweep by temporarily including or excluding edits. The editor prepares one alternative batch that reduces excitation of the concerning mode while preserving targeted clarity gains. A report shows original-variable loadings, reader-outcome sensitivity, residual error, coupled modes, and the local validity window. The author chooses among the original batch, the alternative, or neither; no manuscript change is automatic.","structural_mapping":[{"archetype_element":"Transformation Scope","domain_realization":"The transformation is one editorial round applied to passage-level narratological state vectors within a single declared phase of one manuscript."},{"archetype_element":"State-Vector Definition","domain_realization":"Coordinates operationalize causal explicitness, narrator certainty, contradiction retention, chronology signaling, narrator–implied-author register distance, and focalization leakage."},{"archetype_element":"Invariant Mode Basis","domain_realization":"Eigenvectors of the fitted revision matrix represent combinations of textual features that approximately retain their direction across editorial rounds."},{"archetype_element":"Modal Gain Spectrum","domain_realization":"Each eigenvalue estimates whether its associated feature combination is damped, preserved, reversed, oscillatory, or amplified by another comparable round."},{"archetype_element":"Dominant Mode Selection Rule","domain_realization":"A mode becomes action-relevant only when it is sufficiently stable under resampling, predicts a blinded-reader outcome tied to the author's ambiguity envelope, and contributes materially beyond coordinate-level rivals on held-out transitions."},{"archetype_element":"Stable/Unstable Mode Partition","domain_realization":"Modes are classified relative to persistence or growth under repeated rounds, with any classification restricted to the fitted manuscript phase."},{"archetype_element":"Modal Intervention Map","domain_realization":"Including or excluding proposed edits reveals which edit combinations excite or damp each action-relevant mode."},{"archetype_element":"Reconstruction Residual Check","domain_realization":"Held-out post-revision states are reconstructed from retained modes; large or narratologically structured residuals prevent reliance on the reduced representation."},{"archetype_element":"Mode-Coupling Register","domain_realization":"Near-degenerate, non-orthogonal, or sensitivity-linked modes are recorded so their literary interpretations are not treated as independent."},{"archetype_element":"Interpretation Scope Contract","domain_realization":"The analysis is descriptive of one manuscript and revision phase, does not establish authorial intention or universal narratological laws, and expires after a change in editor, revision purpose, manuscript section, or rating protocol."},{"archetype_element":"Mode Drift Monitor","domain_realization":"At later checkpoints, mode direction, gain, reader association, and retained-versus-discarded spectral separation are compared with the fitted values before reuse."}],"mechanism_mapping":[{"mechanism_slug":"eigendecomposition_workflow","role":"Decompose the explicit, locally fitted revision transition matrix into a complete set of approximate invariant feature combinations and their gains.","counterfactual_removal":"Without it, the checkpoint can compare individual features but cannot identify the coupled directions repeatedly preserved or amplified by editorial rounds."},{"mechanism_slug":"modal_stability_analysis","role":"Classify the extracted modes as decaying, persistent, oscillatory, or growing within the declared revision phase.","counterfactual_removal":"Without it, a mode's current prominence cannot be distinguished from its tendency to compound across successive rounds."},{"mechanism_slug":"modal_sensitivity_sweep","role":"Temporarily include and exclude proposed edits to rank their leverage on reader outcomes and record cross-mode effects.","counterfactual_removal":"Without it, the analysis identifies a concerning mode but cannot translate that diagnosis into a reversible alternative edit batch."},{"mechanism_slug":"residual_reconstruction_test","role":"Test whether retained modes reproduce held-out revision changes and whether omitted residuals contain structured narratological information.","counterfactual_removal":"Without it, a compact modal account could conceal low-amplitude but artistically decisive changes."},{"mechanism_slug":"spectral_decomposition_report","role":"Bind each mode to original textual coordinates, coupling warnings, reader associations, and explicit limits on interpretation.","counterfactual_removal":"Without it, provisional mathematical directions could be reified as authoritative literary categories or prescriptions."},{"mechanism_slug":"spectral_gap_monitor","role":"Check whether retained modes remain separated and directionally stable at later revision checkpoints.","counterfactual_removal":"Without it, an obsolete basis could continue guiding editorial choices after the revision process changes."}],"causal_chain":["Successive editorial rounds alter several narratological features together rather than independently.","A locally repeatable revision process preserves or amplifies some combinations of those features more strongly than others.","Coordinate-level review distributes that movement across individually defensible edits, making the cumulative coupled direction difficult to see.","Modal decomposition exposes the coupled directions and estimates their persistence or gain.","Association with blinded-reader outcomes distinguishes modes relevant to the author's ambiguity envelope from merely high-amplitude modes.","Edit-level sensitivity testing identifies a reversible combination of proposed edits that changes excitation of the relevant mode.","Author review of the original and alternative batches permits a decision informed by coupled transformation, residuals, and interpretive limits.","Reconstruction and drift checks withdraw the modal guidance when the approximation stops representing the revision process."],"baseline":"The baseline is ordinary developmental editing: tracked changes are assessed passage by passage against an authorial brief, with informal full-manuscript rereading after a batch. Review may note narrator consistency, chronology, or ambiguity separately but does not estimate a repeated transformation over their combinations.","nearest_rivals":["A narratological checklist comparing each draft with the authorial brief; it is interpretable but evaluates named coordinates rather than directions created by their joint evolution.","Close reading of successive versions; it can identify subtle textual consequences but supplies no explicit gain, stability boundary, or test of whether the same coupled movement recurs across rounds.","Blinded reader comparison of complete versions; it directly measures reception but detects an endpoint difference without isolating which edit combinations accumulated leverage.","A feature-by-feature revision dashboard; it monitors aggregate coordinate changes but can miss a consequential mode whose individual loadings are each modest."],"remaining_contrastive_claim":"The proposal is distinguishable from its rivals only if a stable coupled revision direction predicts held-out textual or reader-state change better than the same variables considered separately, and if projecting proposed edits onto that direction yields actionable information not already supplied by close reading, checklists, or whole-version reader comparison.","authority_safety":{"decision_authority":"The author retains authority over intended ambiguity and every manuscript change. The editor controls only which reversible alternatives to propose. Analysts and coders may describe fitted modes but may not assign artistic intent or require compliance.","authorized_first_step":"Conduct a retrospective, read-only analysis of already-created revision snapshots and prepare a nonbinding report plus simulated alternative edit packets for author review.","excluded_actions":["Automatically accepting, rejecting, or rewriting manuscript text","Treating a modal label as evidence of authorial intention or literary value","Training the model across other authors or manuscripts without separate permission and scope definition","Sharing manuscript text, reader responses, or inferred editorial tendencies outside the authorized team","Using the analysis to score the author, editor, or individual coder","Continuing to use a basis after residual, drift, or spectral-separation thresholds fail"],"halt_rollback":"Halt if coder agreement is below the preregistered threshold, held-out residuals exceed tolerance, modes are unstable under resampling, the retained spectral gap is inadequate, blinded-reader associations reverse across subsets, or the author withdraws consent. Discard the recommendation, revert to the unmodified manuscript snapshot, and return to ordinary editorial review; preserve only audit material the author has authorized."},"negative_tests":{"strongest_counterevidence":"The apparent reliability-amplifying mode corresponds to an explicit, contemporaneously documented artistic decision by the author, and ordinary feature-level review predicts both the textual transition and blinded-reader response equally well or better.","problem_falsifier":"Across consecutive snapshots, no coupled direction remains stable under coder substitution, passage resampling, and temporal holdout, or any apparent mode disappears after accounting for the individual state variables and planned revision goals.","intervention_falsifier":"On a subsequent consented revision batch, the modal checkpoint neither forecasts movement in blinded-reader reliability ratings nor produces an alternative edit packet that better satisfies the author's ambiguity envelope and reconstructability constraint than the baseline process.","risks":["Operationalized features may flatten context-dependent literary effects.","Coder judgments may import a preferred narratological theory into the state vectors.","A mathematically named mode may acquire unjustified authority in editorial discussion.","A small or shifting spectral gap may make mode identity unstable.","Local linear behavior may fail across chapters, revision purposes, or major structural rewrites.","Optimizing the measured ambiguity envelope may displace unmeasured aesthetic qualities.","Knowledge of the analysis may cause editors to game coordinates rather than read the prose.","Reader samples may confuse demographic or interpretive differences with transformation effects."]},"next_evidence_step":"With author consent, preregister the six textual coordinates, three reader outcomes, agreement threshold, residual tolerance, mode-stability criterion, and comparison models. Use 30 passages from four already-existing consecutive snapshots of one manuscript. Have two blinded coders score all passage versions, fit the transition matrix using the first two transitions, and test reconstruction and reader-outcome predictions on the third transition. Compare the modal model with coordinate-wise regression, the existing editorial checklist, and blinded whole-version ratings. Limit any sensitivity exercise to reversible simulated inclusion or exclusion of edits in six passages; do not alter the working manuscript.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Not assessed against prior proposals; this sealed generation used only the supplied archetype, mechanisms, domain card, and required schema.","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}