{"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_reflexive_efference_fieldwork_002","proposal_index":2,"version":0,"title":"Research-Action Residuals for Reflexive Longitudinal Fieldwork","problem":"In longitudinal participant-observation of a tenant assembly, researchers may facilitate workshops, introduce maps or terminology, ask repeated questions, and circulate findings. These authorized research actions can change who speaks, which issues receive attention, and how participants describe their situation. When later observations depart from the prior pattern, the team lacks a prospective, synchronized account of which changes it expected its own actions to produce. Research-induced, independently emerging, and co-produced changes can therefore be conflated in retrospective interpretation.","actors":["Consented tenant-assembly participants","Participant-observers and workshop facilitators","Field researchers recording meetings","Qualitative analysts interpreting longitudinal change","Study principal investigator","Community advisory group","Research data steward","Research ethics authority"],"observable_state":"For each meeting, the team can align a timestamped log of researcher actions with complete authorized observations of subsequent interaction. The record shows whether expected consequences were stated before the action, which model version generated them, what actually occurred, and how later analytic memos attributed the change. Missing prospective predictions, recurrent prediction errors, analyst disagreement, and participant disagreement are observable.","consequence":"Without separating predicted self-generated consequences from unexplained change, the study may attribute its own elicitation or facilitation effects to community dynamics, overlook external developments, or erase the genuinely co-produced character of observed practices.","affected_objective":"Improve the traceability and contestability of causal interpretations in intervention-adjacent fieldwork while preserving the full interactional record and participants' authority to challenge researcher explanations.","intervention":"Before each already-authorized researcher action, duplicate its action record into a versioned forward model that predicts a bounded set of near-term observable consequences, such as agenda uptake, terminology reuse, participation shifts, or responses to a facilitation prompt. Record uncertainty and rival consequence patterns prospectively. After the meeting, code the corresponding actual observations with provenance and compute a structured residual: predicted consequence absent, predicted consequence stronger or weaker than expected, unpredicted change, timing mismatch, actor mismatch, or meaning that cannot be represented by the model. Weight residuals by observation reliability, model uncertainty, interpretive consequence, participant disagreement, and possible harm. Expected self-associated changes remain visible in the full record but do not automatically enter the scarce queue for claims about independently emerging change; calibrated residuals do. Participant objections, consent issues, conflict, coercion, safeguarding concerns, and uncodable meaning bypass cancellation and receive full-context review. Analysts periodically compare reconstructed accounts with raw field materials and participant interpretations, revise the forward model only through documented review, and suspend cancellation when effects are recursive, delayed, desynchronized, or outside scope.","structural_mapping":[{"archetype_element":"Prediction target and horizon","domain_realization":"For each declared research action, predict only specified observable interactional consequences during a bounded number of subsequent meetings; do not predict participant motives, sincerity, identity, or political legitimacy."},{"archetype_element":"Outgoing action copy","domain_realization":"A timestamped copy of the researcher's approved prompt, workshop, artifact, feedback statement, or facilitation move enters the analytic predictor before participants' subsequent responses are observed."},{"archetype_element":"Generative forward model","domain_realization":"A versioned model links classes of researcher action to uncertain consequence patterns, conditioned on meeting format and prior interaction while declaring excluded contexts and an accountable owner."},{"archetype_element":"Expected behavior","domain_realization":"The team records prospective expectations about sequence, timing, actors, terminology, participation, and agenda effects, including explicit null and rival expectations."},{"archetype_element":"Actual behavior","domain_realization":"Complete authorized audio, transcript, fieldnote, or structured observation records the realized interaction with timestamps, source quality, missingness, and consent status."},{"archetype_element":"Prediction comparator","domain_realization":"The comparison represents absent, amplified, attenuated, delayed, actor-shifted, semantically altered, and wholly unpredicted consequences rather than assigning one anomaly score."},{"archetype_element":"Prediction-error signal","domain_realization":"The residual is the portion of observed change not adequately accounted for by the prospective research-action model; it is a candidate for external, interactional, or model-revision explanation, not proof of any cause."},{"archetype_element":"Precision-weighted propagation","domain_realization":"Residual priority combines evidence quality, uncertainty, participant disagreement, interpretive consequence, rights relevance, and analytic cost; ambiguous or consequential cases receive fuller context rather than stronger automatic attribution."},{"archetype_element":"Reconstruction requirement","domain_realization":"An analyst must be able to reconstruct the provisional account from the predicted research consequence plus the structured residual and then compare it with the complete interaction record."},{"archetype_element":"Model synchronization and provenance","domain_realization":"Action recorders and analysts use the same model checksum, scope, coding definitions, and prediction timestamp; each residual remains attached to the action and observation that generated it."},{"archetype_element":"Update rule","domain_realization":"Validated misses enter a replay buffer and may revise action-consequence beliefs during scheduled reflexive review, with slow, reversible updates separated from immediate interpretation."},{"archetype_element":"Raw audit, drift, and fallback","domain_realization":"Independent full-record reviews, participant challenges, residual-structure tests, and validity clocks detect over-cancellation; recurrent mismatch or non-additive effects restore unfiltered analysis."},{"archetype_element":"Safety-critical bypass","domain_realization":"Consent withdrawal, coercion, retaliation risk, conflict, safeguarding concerns, participant objections, and uncodable cultural meaning bypass the predictor and travel with complete context."},{"archetype_element":"Attention budget","domain_realization":"The study explicitly limits the analyst time devoted to attribution review and measures whether residual processing saves any capacity after prospective prediction, coding, synchronization, and auditing costs."}],"mechanism_mapping":[{"mechanism_slug":"efference_copy_cancellation","role":"Copies each outgoing research action into a forward model, predicts its observable consequences, and separates those expected consequences from the unexplained remainder.","counterfactual_removal":"Without the action copy and forward consequence model, the proposal becomes a retrospective reflexivity memo rather than a predictive self-effect cancellation loop."},{"mechanism_slug":"innovation_residual_filter","role":"Carries uncertainty about both the action-consequence model and observation, using their relative reliability to determine how strongly a mismatch alters the provisional interpretation.","counterfactual_removal":"Without uncertainty-aware innovation processing, noisy field observations or overconfident expectations could generate categorical causal stories from weak mismatches."},{"mechanism_slug":"precision_weighted_error_gate","role":"Prioritizes residuals by reliability, uncertainty, interpretive consequence, participant disagreement, rights relevance, and review cost while retaining suppressed cases for audit.","counterfactual_removal":"Without the gate, trivial deviations could saturate review while subtle but consequential failures of the researcher's causal account remain buried."},{"mechanism_slug":"event_triggered_residual_reporting","role":"Routes validated, consequential failures of the action-consequence prediction to a dedicated attribution-review queue, with observation heartbeats distinguishing no residual from missing evidence.","counterfactual_removal":"Without event-triggered routing, residuals would not reallocate scarce analytic attention and would merely decorate the full field record."},{"mechanism_slug":"bayesian_model_update","role":"Represents action-consequence beliefs and their uncertainty, allowing reviewed evidence to revise them incrementally without treating one meeting as decisive.","counterfactual_removal":"Without bounded belief revision, the forward model would remain static or be rewritten informally after every outcome, enabling hindsight-driven expectations."},{"mechanism_slug":"model_version_checksum_handshake","role":"Confirms that the action recorder, comparator, and analyst interpret each residual against the same prospective model and coding definitions.","counterfactual_removal":"Without synchronization, an analyst could apply a residual to a revised expectation that did not exist when the action occurred."},{"mechanism_slug":"prediction_error_replay_buffer","role":"Stores material mismatches with the action, prospective prediction, raw-context pointer, participant response, model version, and later interpretation for review and regression testing.","counterfactual_removal":"Without contextual replay, model learning would rely on recollected surprises and could not distinguish prospective errors from retrospective narrative repair."},{"mechanism_slug":"residual_comparison_test","role":"Tests whether residuals retain systematic timing, actor, semantic, or meeting-type structure and compares them with a rival consequence model and raw-record sample.","counterfactual_removal":"Without comparison, the team could label structured model failure as independent social change or harmless observational noise."},{"mechanism_slug":"shadow_raw_channel_sampling","role":"Assigns an independent analyst to review selected complete interaction records without seeing the production prediction, then compares interpretations and identified changes.","counterfactual_removal":"Without an independent raw path, the research team's own expectations would define both what was cancelled and whether that cancellation was acceptable."},{"mechanism_slug":"raw_signal_fallback_switch","role":"Suspends cancellation and restores full-context analysis when effects are delayed, recursive, outside scope, version-skewed, ethically sensitive, or poorly reconstructed.","counterfactual_removal":"Without fallback, the model could continue subtracting meaningful social responses after its limited self-effect assumptions had failed."}],"causal_chain":["An already-authorized researcher action is recorded prospectively rather than reconstructed after its outcome is known.","A scoped, versioned forward model predicts an uncertain set of near-term observable consequences of that action.","The subsequent interaction is captured independently and completely with provenance and explicit missingness.","A structured comparator separates matched consequences from absent, altered, delayed, actor-shifted, or unpredicted observations.","Precision weighting determines which unexplained residuals warrant scarce causal-attribution attention, while protected conditions bypass cancellation.","Analysts reconstruct the provisional account as predicted self-associated consequence plus residual and verify it against full contextual evidence.","Residuals that survive review motivate investigation of external events, interaction effects, boundary errors, or model misspecification rather than immediate causal conclusions.","Independent raw review and participant challenge reveal observations that the research-action model wrongly explained away.","Reviewed residuals update the forward model slowly, with versioning that preserves what was predicted before each outcome.","Structured drift, failed reconstruction, ethical concern, or non-additive feedback disables cancellation and restores full-record interpretation."],"baseline":"Researchers maintain reflexive field diaries, intervention logs, analytic memos, and complete meeting records, then discuss possible reactivity retrospectively. This preserves context and supports reflexivity but does not require a prospective action-consequence prediction, a model-relative residual, synchronized reconstruction, precision-weighted propagation, or an independently audited cancellation loop.","nearest_rivals":["Reflexive field diaries: document researcher positionality and perceived influence, but usually do not generate testable consequence predictions before each action or compute structured residuals afterward.","Process tracing: evaluates causal mechanisms using sequences and evidence, but does not make a forward model of researcher-generated observations the baseline that is subtracted from the incoming record.","Member checking or participant validation: lets participants challenge interpretations, but does not itself create continuous prediction, residual routing, model synchronization, or an attention budget.","Researcher-intervention logs: preserve provenance for facilitation and elicitation, but an action inventory alone neither predicts consequences nor distinguishes matched effects from unexplained change.","Interrupted time-series or before-and-after comparison: can estimate aggregate changes around an intervention, but may not represent interactional meaning, actor-specific residuals, evolving researcher actions, or raw-context fallback.","Conventional anomaly detection: can flag departures from prior meetings, but does not specifically predict and cancel consequences generated by the research team's own outgoing actions."],"remaining_contrastive_claim":"The design's remaining contrastive claim is structural: it prospectively copies an outgoing research action into a synchronized forward model, compares predicted with observed interactional consequences, treats the calibrated unexplained remainder as the primary signal for attribution review and model learning, and governs cancellation through independent raw review, participant challenge, protected bypasses, and full-context fallback.","authority_safety":{"decision_authority":"The principal investigator may authorize shadow analytic use only for research actions and observations already permitted by the protocol. The ethics authority controls changes to consent, recording, intervention, or data use; the data steward controls raw-record access; and the community advisory group can challenge protected classes, model boundaries, and interpretations of researcher influence.","authorized_first_step":"For a bounded set of already-planned and already-approved research activities, record prospective action-consequence predictions and evaluate them in shadow mode without changing what researchers do, what participants receive, or which account remains the analytic record of record.","excluded_actions":["Adding or modifying participant-facing interventions to improve prediction quality","Using a residual as proof that change was external, endogenous, or caused by a participant","Suppressing expected observations from the complete research record","Predicting participant motives, sincerity, political commitments, credibility, or cultural worth","Reducing consent, confidentiality, safeguarding, withdrawal, or mandatory-reporting protections","Automatically revising the action-consequence model from live meetings","Using the system to steer tenant-assembly decisions or evaluate individual participants","Treating participant adoption of researcher language as mere contamination without participant interpretation","Applying the model outside its declared action class, site, meeting format, or time horizon","Allowing low residual scores to override participant objections or contradictory raw evidence"],"halt_rollback":"Halt cancellation for the affected scope if prospective predictions are missing, model checksums diverge, raw records are unavailable, participant objections are suppressed, reconstruction disagreement exceeds the preregistered budget, errors cluster by participant role or protected group, residuals remain temporally or semantically structured, effects extend beyond the validity horizon, or model-and-audit work consumes the intended analytic capacity. Freeze updates, retain every prediction and comparison, revert to full-record reflexive analysis, and require principal-investigator, ethics, and community review before resumption."},"negative_tests":{"strongest_counterevidence":"The strongest counterevidence is that researcher and participant action are recursively co-produced rather than additively separable: the same prompt acquires different meanings through participants' responses, so subtracting a predicted researcher effect systematically removes the interactional phenomenon the study should explain.","problem_falsifier":"The problem is falsified if the fieldwork contains no participant-facing researcher actions during the analytic period, if complete intervention logs already support timely and consistently prospective attribution, or if analysts and participants find no consequential ambiguity between research-associated and independently emerging change.","intervention_falsifier":"The intervention is falsified if prospective residual analysis does not improve traceability or reviewer agreement over ordinary reflexive logs after including its maintenance cost, or if it increases hindsight repair, over-cancellation, participant disagreement, subgroup-specific error, privacy risk, or unsupported causal attribution.","risks":["The forward model may imply a false mechanical separation between researcher and community action.","Expected researcher effects may be subtracted even when their social meaning is the central finding.","Participants may experience the model as researchers claiming privileged knowledge of their responses.","Prospective predictions may discipline facilitators toward behaviors that make the model look accurate.","Slow or indirect effects may fall outside the horizon and be mislabeled as external change.","The model may encode dominant researchers' assumptions about normal participation.","High-residual cases may expose identifiable dissent, conflict, or vulnerability.","A shared wrong model can align analysts around the same causal error.","Repeated updates may normalize increasing researcher influence instead of revealing it.","The added prediction and audit work may exceed the analytic capacity it is intended to free." ]},"next_evidence_step":"Pre-register a shadow pilot covering no more than six already-scheduled, protocol-approved tenant-assembly meetings. Before each approved researcher action, lock a versioned prediction of bounded observable consequences and at least one rival pattern. Preserve the complete meeting record. Have one analyst apply the residual procedure and a second analyst, blinded to the predictions, conduct ordinary full-record reflexive analysis; invite the existing community advisory group to review discrepant interpretations. Compare prospective completeness, reconstruction disagreement, unsupported attribution, participant or advisory disagreement, role-specific misses, structured residuals, fallback frequency, privacy incidents, and total analytic time. Do not use pilot residuals to alter meetings or substantiate causal claims.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 addresses analyst-attention overload from repetitive benefit-intake fieldnotes by encoding deviations from an expected encounter template so changing institutional practices surface sooner. This proposal addresses a different problem: causal and reflexive ambiguity created by the research team's own actions in longitudinal participant-observation. Its intervention begins with an efference copy of an outgoing research action, predicts the action's consequences, and routes the unexplained remainder for attribution review. It is independently adoptable in studies with researcher interventions even when full-record review is not backlogged, and proposal 1 can operate where researchers do not intervene at all. The causal paths, decision objectives, primary users, evidence tests, and principal failure modes are therefore distinct.","revision_record":{"parent_version":null,"progress_targets_addressed":["Specified a second independently adoptable sociology-and-anthropology problem","Preserved predictive residual structure through a prospective research-action forward model","Established operational authority, participant safeguards, falsifiers, rivals, rollback, and bounded evidence","Explained material diversity from sealed proposal 1"],"conceptual_changes":["Instantiated predictive residual processing as reflexive cancellation of predicted research-generated consequences rather than compression of repetitive ethnographic observations.","Made the outgoing researcher action—not an institutional encounter template—the feedforward input.","Defined residuals as prompts for competing causal explanations rather than evidence of external causation."],"operational_changes":["Required prospective action logs, locked predictions, rival patterns, version checks, independent raw review, participant challenge, and full-context fallback.","Restricted the first step to shadow analysis of already-approved activities across no more than six meetings.","Prohibited participant-facing changes and operational use of residual classifications."],"evidence_changes":["Defined a prospective comparison with blinded full-record reflexive analysis and community review.","Included unsupported attribution, structured residuals, participant disagreement, role-specific misses, privacy incidents, fallback use, and total analytic cost."],"claim_changes":["Made no claim of novelty, prevalence, demand, or effect size.","Limited the proposal to a testable architecture for traceable reflexive attribution under explicit non-additivity and over-cancellation risks."]}}