{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp06_four_proposal_generalization60_20260803","cell_id":"predictive_residual_processing__sociology_anthropology","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"prp_residual_ethnographic_encounter_review_001","proposal_index":1,"version":0,"title":"Residual Ethnographic Review for Changing Benefit-Access Encounters","problem":"In a multi-site ethnographic study of consented public-benefit intake encounters, fieldworkers repeatedly document stable procedural sequences, roles, and interaction patterns. Full chronological review consumes limited analyst attention, so deviations such as new gatekeeping practices, altered kin mediation, language switching, or discretionary reinterpretation of rules may not be examined until later coding. Simple exception reporting would be unsafe because an official or researcher-defined baseline could normalize coercion, exclusion, or minority experiences.","actors":["Consented participants in observed benefit-access encounters","Field researchers producing encounter notes","Qualitative analysts reviewing observations across sites","Study principal investigator","Research data steward","Community advisory panel","Research ethics authority"],"observable_state":"The study's review queue contains full encounter narratives in which recurring steps and interaction patterns are restated. Analysts can measure review latency, repeated-content share, and whether material deviations identified during later full-note coding were visible during initial review. Each observation has a timestamp, site, provenance, consent status, and complete raw note.","consequence":"Scarce interpretive attention is spent reprocessing expected material while changes in access practices or social relations can reach investigators late. If the baseline itself reflects institutional assumptions, consequential minority or rights-bearing observations may also be misclassified as routine.","affected_objective":"Allocate ethnographic review attention sooner to consequential changes while retaining the contextual fidelity, contestability, participant protections, and raw evidence required for valid qualitative interpretation.","intervention":"Create a shadow-mode, versioned expected-encounter model for narrowly defined site and service contexts. Before each review window, the model predicts an encounter skeleton: likely sequence of steps, participating roles, access decisions, mediation patterns, language channels, and uncertainty. A field researcher maps the actual consented observation to the same structure and records signed edits—additions, omissions, substitutions, reorderings, timing changes, role changes, and uncodable material. A consequence-and-precision rule routes selected residuals, their model version, uncertainty, and enough surrounding context to an analyst queue; the analyst can reconstruct the encounter skeleton as prediction plus edits and request the full narrative at any time. Rights, consent, coercion, discrimination, safeguarding, identity-sensitive ambiguity, and researcher-uncertainty classes bypass suppression and carry the complete note. All raw notes remain independently stored. Random and risk-stratified full-note audits, model-version checks, periodic full-baseline review, residual-distribution monitoring, and explicit fallback conditions determine whether residual-mode triage may continue. Validated residuals inform slow, reviewed revisions of the expected-encounter model rather than automatic redefinition of normality.","structural_mapping":[{"archetype_element":"Prediction target and observation boundary","domain_realization":"Predict the structured sequence, roles, channels, and access outcomes of one consented intake encounter within a declared site, service type, and review window; do not predict participant identity, motive, or cultural worth."},{"archetype_element":"Generative model with scope, horizon, uncertainty, and owner","domain_realization":"A versioned expected-encounter template is conditioned on site and service context, applies only to the next review window, carries uncertainty by field, and is owned by the study team under community and ethics oversight."},{"archetype_element":"Expected and actual behavior","domain_realization":"The template records the expected encounter skeleton before review; the field note and structured observation record the realized sequence with timestamp, provenance, missingness, and uncodable content."},{"archetype_element":"Signed or structured prediction comparator","domain_realization":"The comparator represents additions, omissions, substitutions, reordering, timing shifts, role changes, and unresolved narrative material rather than collapsing difference into a single anomaly score."},{"archetype_element":"Precision and consequence weighting","domain_realization":"Routing weights combine observation quality, researcher confidence, interpretive ambiguity, access consequence, subgroup coverage, and rights relevance; small reliable rights-related deviations outrank large but noisy stylistic differences."},{"archetype_element":"Residual propagation and reconstruction","domain_realization":"The analyst receives the structured edits, baseline version, uncertainty, and local context, reconstructs the encounter skeleton, and can open the complete narrative without permission from the predictor."},{"archetype_element":"Bounded update rule","domain_realization":"Material, validated residuals enter a review buffer; model revisions occur only at scheduled reviews, are attributable and reversible, and cannot automatically absorb coercive or exclusionary practices into the expected baseline."},{"archetype_element":"Model synchronization and provenance","domain_realization":"Fieldworker and analyst interfaces must use the same template checksum; every residual identifies the model version, site scope, observation source, transformations, and time."},{"archetype_element":"Freshness, drift, and error budget","domain_realization":"Templates expire after a declared interval. Sustained structured residuals, subgroup reconstruction gaps, rising uncodable content, or cumulative consequential omissions exhaust the error budget and suspend residual triage."},{"archetype_element":"Independent raw-signal audit","domain_realization":"A data steward preserves all authorized raw notes, while analysts review a random baseline sample plus risk-stratified full notes without relying on the production template."},{"archetype_element":"Decompression and safety bypass","domain_realization":"Consent concerns, coercion, discrimination, safeguarding, identity-sensitive ambiguity, researcher uncertainty, checksum failure, or drift immediately route the full note and disable residual-only handling for the affected scope."},{"archetype_element":"Attention budget and verified silence","domain_realization":"A fixed analyst-review budget is declared before the pilot; observation heartbeats and missingness states distinguish an expected encounter from a missing note, failed channel, or uncompleted comparison."}],"mechanism_mapping":[{"mechanism_slug":"predictive_codec","role":"Uses matched expected-encounter templates so structured edits can serve as the review payload while the analyst reconstructs the encounter skeleton as baseline plus correction.","counterfactual_removal":"Without matched prediction and reconstruction, the system becomes ordinary selective coding or exception reporting rather than predictive residual processing."},{"mechanism_slug":"precision_weighted_error_gate","role":"Ranks residuals using observation reliability, uncertainty, access consequence, rights relevance, subgroup coverage, and review cost while retaining suppressed items for audit.","counterfactual_removal":"Without consequence-aware weighting, visually large but interpretively weak differences could displace small, reliable indicators of exclusion or coercion."},{"mechanism_slug":"event_triggered_residual_reporting","role":"Routes threshold-crossing structured deviations to the near-term analyst queue while heartbeats and explicit missingness make silence interpretable.","counterfactual_removal":"Without event-triggered routing, full chronological review remains the scarce channel and informative deviations receive no preferential attention."},{"mechanism_slug":"model_version_checksum_handshake","role":"Prevents analysts from interpreting an edit against a different site template or revision than the field researcher used.","counterfactual_removal":"Without version compatibility, a valid residual could reconstruct a false encounter while appearing internally coherent."},{"mechanism_slug":"shadow_raw_channel_sampling","role":"Sends random and risk-stratified complete notes through an independent audit review to measure what the expected template and gate suppressed.","counterfactual_removal":"Without independent raw sampling, the team could not detect interactional meaning or minority experiences that its own baseline fails to represent."},{"mechanism_slug":"model_drift_monitoring","role":"Tracks structured changes in residual distributions, uncodable material, reconstruction disagreement, subgroup errors, and template age.","counterfactual_removal":"Without drift and staleness checks, a changing institutional practice could be repeatedly treated as isolated noise or silently absorbed."},{"mechanism_slug":"periodic_full_state_resynchronization","role":"Requires scheduled full-baseline review and reconciliation of templates, raw notes, code definitions, and active versions.","counterfactual_removal":"Without full resynchronization, small interpretive and version differences could accumulate across sites and analysts."},{"mechanism_slug":"raw_signal_fallback_switch","role":"Restores full-note review when validity, safety, consent, synchronization, or reconstruction conditions fail, with hysteresis before returning to residual mode.","counterfactual_removal":"Without fallback, the model would remain an obligatory filter precisely when its assumptions are least trustworthy."},{"mechanism_slug":"prediction_error_review","role":"Examines material misses with provenance and decides whether the model, observation method, analytic boundary, or downstream response requires revision.","counterfactual_removal":"Without governed review, residuals would generate alerts but would not safely update the shared interpretive model."}],"causal_chain":["Repeated encounter structures permit a narrow, versioned model to generate an expected encounter skeleton before analyst review.","The actual consented observation is represented in the same units while the full narrative remains independently stored.","A structured comparator produces model-relative edits that preserve direction, type, uncertainty, and provenance.","A precision-and-consequence gate routes informative residuals within the fixed attention budget, while protected classes bypass the gate with full context.","Analysts reconstruct expected context from the synchronized template plus residual and can demand the raw note whenever interpretation requires it.","Validated deviations receive earlier focused review and enter a replay buffer for later model, method, or boundary review.","Random raw-note audits and scheduled full-baseline reviews expose meaning that the predictor suppressed or could not encode.","Drift, version mismatch, consequential reconstruction error, or safety conditions suspend residual processing and restore full-note review.","Only reviewed, attributable residuals may revise the next model version, closing the prediction-observation-learning loop without allowing automatic normalization."],"baseline":"Field researchers retain complete narrative notes for every authorized encounter; analysts review them chronologically or during periodic coding with a conventional qualitative codebook. This preserves context but offers no maintained prediction, reconstructive residual message, synchronized baseline, or explicit attention allocation to unexpected changes.","nearest_rivals":["Rapid ethnographic assessment: accelerates synthesis through focused collection and team debriefs, but does not require prediction-plus-residual reconstruction, synchronized model versions, or raw-channel fidelity tests.","Conventional qualitative coding and framework matrices: organize full notes into categories and comparisons, but coded excerpts are not necessarily model-relative residuals and do not reconstruct observations from a shared prediction.","Critical-incident or exception reporting: foregrounds unusual encounters, but often leaves the baseline, uncertainty, suppressed stream, and silence semantics implicit.","Anomaly detection over coded events: can flag improbable observations, but a flag alone does not provide a reconstructive residual channel, bounded learning loop, independent raw audit, or full-signal fallback.","Routine full-note review with priority tags: preserves all context and allows triage, but does not suppress expected representation through a synchronized generative model or use residuals to update that model."],"remaining_contrastive_claim":"The design's remaining contrastive claim is structural: a narrow expected-encounter model is shared and versioned; the primary triage representation is a reconstructive, precision-weighted edit against that model; validated edits update it; and independent full-note audits, protected bypasses, resynchronization, and decompression govern what prediction is allowed to suppress.","authority_safety":{"decision_authority":"The principal investigator may authorize only a shadow analytic pilot within the existing research protocol. The data steward controls raw-note access; the research ethics authority governs any change in collection, consent, or data use; and the community advisory panel reviews protected classes and interpretations that could redefine local conduct as normal or deviant.","authorized_first_step":"Run a retrospective shadow comparison on deidentified notes already authorized for analysis, without changing fieldwork, participant treatment, official benefit decisions, or the analysts' record of record.","excluded_actions":["Using residual scores to judge participant credibility, deservingness, risk, or cultural conformity","Changing eligibility, service delivery, enforcement, or other benefit decisions","Discarding, overwriting, or withholding authorized raw notes","Reducing existing consent, confidentiality, safeguarding, or mandatory-reporting protections","Automatically updating the expected-encounter model from live observations","Treating institutional procedure as the normative standard for participant behavior","Deploying residual-only review outside the tested site, service type, population, or validity window","Using protected attributes to lower propagation priority","Representing channel silence as confirmation without a heartbeat and explicit missingness state"],"halt_rollback":"Immediately halt residual triage for the affected scope if a protected bypass fails, model versions diverge, raw-note access is unavailable, reidentification risk appears, consequential reconstruction errors exceed the preregistered budget, errors cluster by subgroup or site, residual structure indicates drift, or total model-and-audit effort erases the attention saving. Freeze model updates, preserve all logs, revert to full-note review, notify the data steward and principal investigator, and require ethics and community review before restarting."},"negative_tests":{"strongest_counterevidence":"The strongest counterevidence would be that interactional meaning is distributed through apparently routine narrative detail, so analysts using prediction plus residual repeatedly reach different interpretations from independent full-note reviewers—especially for minority, indirect, relational, or context-dependent observations—even when structured reconstruction looks accurate.","problem_falsifier":"The inferred problem is falsified if measured full-note review has no material queue, latency, or attention constraint; if later coding rarely finds consequential changes that initial review missed; or if encounter structure is too variable to predict within a defensible interpretive error budget.","intervention_falsifier":"The intervention is falsified if shadow use does not reduce review burden after including model maintenance and audits, or if it increases missed consequential deviations, subgroup disparities, interpretive disagreement, false escalation, privacy exposure, or fallback burden relative to full-note review.","risks":["The expected template may reify official procedure or dominant-group assumptions as normal.","Researchers may alter notes to fit available residual categories.","Slowly worsening exclusion may be absorbed through model updates rather than escalated.","Outlier-focused storage may expose identifiable or sensitive participant experiences.","Analysts may lose interactional context even when the encounter skeleton reconstructs correctly.","Published thresholds may invite gaming or comfort-driven tuning.","Residual alerts may create a deficit-focused portrayal of a community or institution.","Random audits may still miss rare harms, while risk-stratified audits may reproduce known blind spots.","Model maintenance, synchronization, and review may consume more attention than they release.","A rights-related event may be miscoded below the bypass threshold."]},"next_evidence_step":"Pre-register and run a bounded retrospective shadow pilot on up to 60 deidentified, already-authorized encounter notes spanning no more than three study sites. Build the expected templates from an earlier subset, lock one version, and have one analyst triage prediction-plus-residual packets while an independent analyst reviews complete notes. Compare review time including model maintenance, reconstruction agreement, material-deviation recall, false escalations, rights-class and subgroup-specific misses, audit disagreement, version failures, fallback frequency, and privacy incidents. Do not replace the full-note record or operational workflow during this test.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Not applicable within this sealed one-proposal response.","revision_record":{"parent_version":null,"progress_targets_addressed":["Initial complete specification of a concrete sociology-and-anthropology problem","Preservation of the archetype's prediction, residual, reconstruction, update, synchronization, audit, and fallback structure","Explicit authority limits, protected bypasses, rollback conditions, rivals, and falsifiers","Bounded closed-book first-evidence design"],"conceptual_changes":["Instantiated the archetype as versioned residual review of ethnographic encounter sequences rather than generic anomaly detection.","Defined the residual as structured edits to an expected interaction sequence while treating uncodable narrative as a full-signal trigger.","Separated descriptive predictability from normative acceptability so recurring harmful conduct cannot become safe to suppress merely by becoming expected."],"operational_changes":["Specified synchronized site-scoped templates, heartbeats, provenance, consequence weighting, random and risk-stratified raw audits, scheduled resynchronization, and raw-note fallback.","Restricted initial use to retrospective shadow analysis of already-authorized data.","Assigned authority across the principal investigator, data steward, ethics authority, and community advisory panel."],"evidence_changes":["Defined a preregistered, bounded comparison against independent full-note review.","Included total maintenance cost, reconstruction agreement, consequential and subgroup-specific misses, fallback use, and privacy incidents as decision evidence."],"claim_changes":["Made no claim of novelty, prevalence, demand, or effect size.","Limited the claim to a testable structural intervention and its contrast with named rival approaches."]}}