{"actors":["Literary scholars formulating claims about narrator knowledge or reliability","Research assistants annotating passages","A project lead adjudicating which textual evidence may support a claim"],"affected_objective":"Traceable attribution of textual evidence to narrative sources and defensible adjudication of claims about changes in narrator knowledge, alignment, and reliability.","arm":"COMMON_P1","authority_safety":{"authorized_first_step":"The project lead may authorize a non-public pilot on one public-domain novel using preselected passages and de-identified annotator records; the pilot may evaluate the annotation method but may not establish a canonical interpretation.","decision_authority":"The project lead selects the pilot text and makes provisional methodological decisions; annotators record observations and uncertainty but cannot settle interpretive disputes or alter publication, teaching, or assessment standards.","excluded_actions":["Do not treat the hybrid annotations as ground truth about the text.","Do not rank annotators, students, authors, or theoretical traditions from pilot results.","Do not automate acceptance or rejection of interpretive claims.","Do not publish, generalize, or incorporate the scheme into assessment criteria without separate review."],"halt_rollback":"Stop the pilot if source boundaries cannot be applied consistently, continuous dimensions invite unsupported precision, or the scheme suppresses material ambiguity. Preserve the raw passage notes, discard derived state assignments, and return to passage-specific prose annotations."},"baseline":"Passages are currently handled through chapter-level narrator labels plus prose notes. This makes major frame changes visible only when a reader records them, offers no stable rule for whether a local change resets attribution, and provides no structured way to distinguish gradual modulation within a voice from a change of narrative source.","candidate_id":"discrete_continuous_model_selection__literature_literary_theory__COMMON_P1","causal_chain":["Interpretive adjudication requires deciding which narrative source is responsible for each textual assertion and how that source's epistemic position changes.","Narrative discourse can change gradually within a source while also undergoing discrete switches through quotation, embedded documents, frame entry or exit, and explicit correction.","Chapter-level or smoothed representations interpolate across some source switches, allowing evidence from different attribution regimes to be combined.","Pure state labels instead force gradual changes in certainty, access, or alignment into arbitrary categories and create disputed cliffs between adjacent passages.","The hybrid protocol marks source-changing events while retaining continuous, dimension-specific traces within each resulting regime.","Explicit switching and uncertainty rules make missed events, artificial segmentations, and boundary cases inspectable against the text.","The project lead can then adjudicate a fixed interpretive claim using evidence whose source attribution and within-source modulation are separately visible."],"cell_id":"discrete_continuous_model_selection__literature_literary_theory","consequence":"Evidence may be attributed across incompatible narrative voices, gradual irony or estrangement may be mistaken for a categorical switch, and nearly identical passages may receive different evidentiary treatment solely because an analyst placed a boundary differently. The resulting interpretation is difficult to audit because source changes and within-source variation are conflated.","diversity_from_prior_proposals":"No prior proposals were inspected under runtime isolation. This candidate is internally differentiated by applying model-form selection to evidentiary attribution across narrative-source switches, rather than to genre classification, plot segmentation, or general representation choice.","experiment_id":"eoa_inverse_innovation_exp13_second_slot_policy60_20260806","intervention":"Introduce a hybrid narrative-source trace for adjudicating claims about narrator reliability. Annotators first mark only source-changing events with explicit boundaries: entry into or exit from a frame, direct quotation, inserted document, transfer to an embedded narrator, and explicit retraction or correction. Within each resulting source regime, they record separate continuous traces for epistemic access, expressed certainty, and evaluative alignment at paragraph resolution, increasing to clause resolution only around a proposed event. A discrete event resets source attribution; continuous values may vary within a regime but are not interpolated across a reset. Ambiguous boundaries receive competing annotations and an uncertainty note rather than a forced state. The representation is retained only if transition review shows that it preserves consequential source changes without inventing unsupported stages.","mechanism_mapping":[{"counterfactual_removal":"Without the combined model, the protocol must either smooth across source changes or replace gradual within-source modulation with categorical states.","mechanism_slug":"hybrid_discrete_continuous_model","role":"Combines discrete narrative-source resets with continuous traces of access, certainty, and alignment inside each source regime."},{"counterfactual_removal":"Without event-sensitive resolution, paragraph sampling can miss a clause-level quotation, correction, or frame transition that changes attribution.","mechanism_slug":"sampling_interval_choice","role":"Uses paragraph-level observation by default and clause-level observation around candidate source-changing events."},{"counterfactual_removal":"Without the audit, added annotations could appear systematic while still missing textual transitions or manufacturing boundaries that do not affect adjudication.","mechanism_slug":"transition_resolution_audit","role":"Checks proposed boundaries and smooth regions against textual markers, contested cases, and changes in evidentiary attribution."}],"nearest_rivals":["A pure discrete scheme assigning every passage to fixed narrator-reliability states","A single smoothed reliability or focalization trajectory across the text","Unstructured close-reading notes without explicit source-reset, sampling, or continuity rules"],"negative_tests":{"intervention_falsifier":"The intervention is falsified for the pilot task if the hybrid trace neither resolves nor makes more inspectable any preselected attribution dispute than the simpler rivals, while producing additional unsupported boundaries, unstable continuous traces, or different adjudications driven mainly by annotation order.","problem_falsifier":"The problem is falsified for the selected text and claims if narrative source remains fixed across the relevant passages, gradual variation is immaterial to evidence use, or adjudications remain identical when plausible event boundaries and within-source traces are removed.","risks":["Continuous traces may imply numerical objectivity where only situated interpretation exists.","Discrete event labels may reify theoretically contested distinctions among narrator, focalizer, quoted speaker, and implied author.","The annotation burden may redirect attention from unanticipated textual details.","Preselected dimensions may privilege one narratological framework and suppress alternatives.","Analysts may mistake agreement produced by shared instructions for independent textual support."],"strongest_counterevidence":"If independent annotators repeatedly disagree both on source-reset events and on the direction of within-source change, yet can justify their readings from the same textual marks, the text's productive ambiguity may be the relevant object rather than a transition structure suitable for hybrid modeling."},"next_evidence_step":"Run one counterbalanced pilot on 24 preselected passages from a single public-domain novel: eight expected source switches, eight expected gradual within-source shifts, and eight mixed or ambiguous cases. Two annotators independently apply the hybrid protocol and the two formal rivals to a fixed set of 12 interpretive claims. Record missed source-changing markers, unsupported boundaries, cross-boundary interpolation, annotator disagreement, time burden, and whether each representation changes the admissible evidence for a claim. End after this corpus and report only passage-level results and failure cases.","observable_state":"For each sampled clause or paragraph: the currently attributed narrative source; textual markers of quotation, document insertion, frame entry or exit, speaker transfer, correction, or retraction; annotator-valued direction and degree of epistemic access, expressed certainty, and evaluative alignment; the sampling resolution used; competing boundary placements; uncertainty notes; and the interpretive claims for which the passage is treated as evidence.","prior_art_status":"UNSEARCHED","problem":"A literary-research team adjudicating claims about an unreliable narrator represents narrative position either as a stable sequence of chapter-level states or as a smooth trajectory. The text, however, may contain discrete changes of narrative source alongside gradual shifts in certainty, knowledge, or alignment within a source. Smoothing can erase a quotation, embedded document, frame crossing, or correction that changes who is responsible for an assertion; rigid segmentation can turn gradual irony or distancing into arbitrary stages. Because the representation determines which passages may be aggregated as evidence, wrong granularity can alter the interpretation being adjudicated.","proposal_index":1,"remaining_contrastive_claim":"For passage-level adjudication of narrator-reliability claims, separating discrete source-attribution resets from continuous within-source modulation should expose a class of evidentiary errors that neither fixed narrator states nor a single smooth reliability trajectory can represent; the bounded pilot must determine whether those differences survive independent annotation and affect the admissibility of evidence.","revision_record":{"claim_changes":["Initial version; all statements are framed as a testable proposal rather than established effects."],"conceptual_changes":["Initial formulation maps discrete events to narrative-source resets and continuous change to within-source epistemic and evaluative modulation."],"evidence_changes":["Initial bounded pilot specifies one text, 24 passages, two annotators, formal rivals, fixed claims, and passage-level failure measures."],"operational_changes":["Initial safeguards assign interpretation authority to the project lead, preserve ambiguity through competing annotations, and prohibit automated or public use."],"parent_version":null,"progress_targets_addressed":["Concrete domain problem and decision task","Observable state and decision consequence","Discrete, continuous, and hybrid boundary rules","Causal and mechanism mappings","Nearest-rival contrast","Authority, rollback, falsifiers, and risks","Bounded first evidence step"]},"schema_version":1,"structural_mapping":[{"archetype_element":"Decision Need","domain_realization":"Decide which passages may be attributed to the same narrative source and used together when adjudicating a claim about narrator reliability."},{"archetype_element":"Process Change Signature","domain_realization":"Narrative position exhibits gradual changes in access, certainty, and alignment within a voice, interrupted by source-changing events such as quotation, inserted documents, frame crossings, and corrections."},{"archetype_element":"Cost of False Smoothness","domain_realization":"Interpolating across a source-changing event can merge assertions made by different voices and misassign responsibility for knowledge or distortion."},{"archetype_element":"Cost of False Discreteness","domain_realization":"Forcing gradual irony, distancing, or uncertainty into fixed states creates artificial interpretive cliffs and hides variation inside each label."},{"archetype_element":"Granularity Choice","domain_realization":"Use a hybrid model with discrete source regimes and continuous, dimension-specific traces within regimes."},{"archetype_element":"Step Boundary","domain_realization":"A boundary requires a cited textual marker that changes the attributable source or explicitly retracts or corrects the standing account; disagreement is retained rather than resolved by default."},{"archetype_element":"Continuity Assumption","domain_realization":"Access, certainty, and alignment may be interpolated only between adjacent observations assigned to the same source regime and only when no contrary textual marker appears."},{"archetype_element":"Measurement Resolution","domain_realization":"Annotate at paragraph resolution normally and at clause resolution around proposed source-changing events; record the resolution with every judgment."},{"archetype_element":"Transition Validation","domain_realization":"Review whether marked events correspond to cited textual changes, whether continuous traces cross an unmarked source switch, and whether boundaries alter admissible evidence for the fixed interpretive claims."},{"archetype_element":"Hybrid Boundary Rule","domain_realization":"A validated source-changing event terminates the prior continuous trace and starts a new attribution regime; ambiguous events preserve parallel traces until adjudication."},{"archetype_element":"Approximation Error Check","domain_realization":"Compare hybrid, pure-state, and smoothed-trajectory outputs for missed source changes, invented stages, cross-boundary evidence aggregation, and altered claim adjudication."}],"title":"Hybrid Narrative-Source Trace for Reliability Adjudication","version":0}