{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp06_four_proposal_generalization60_20260803","cell_id":"predictive_residual_processing__futurism_foresight","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"prp-foresight-residual-delphi-002","proposal_index":2,"version":0,"title":"Residual-Round Delphi for Tracking Expert Belief Change","problem":"A multi-round Delphi exercise asks experts to resubmit complete probability, timing, confidence, dependency, and rationale fields for every proposition in every round. Successive responses may contain many stable fields, but the facilitator must process the whole questionnaire to find the belief revisions that matter. Repetition consumes the panel's response budget, material changes can lose their explanatory context in aggregate summaries, and missing responses can be mistaken for unchanged judgments.","actors":["Experts who submit anonymous judgments","Delphi facilitators who define propositions and administer controlled feedback","Elicitation-method owners who maintain response schemas and prediction rules","Independent audit reviewers who compare residual reconstructions with full responses","Decision sponsors who receive the resulting uncertainty assessment","People and groups who could be affected by decisions informed by the elicitation"],"observable_state":"Across successive rounds, field-level response records can show how many probability, timing, confidence, dependency, and rationale entries remain unchanged; completion timestamps and omissions show the response burden; reconstruction checks show whether a prior-round expectation plus submitted corrections reproduces each expert's current judgment; and randomly assigned from-scratch questions reveal changes that an editable baseline may have suppressed.","consequence":"If stable fields consume most of a fixed expert-attention budget, less time remains for reconsidering contested propositions and explaining material belief changes. Facilitators may also aggregate apparent stability without distinguishing confirmed continuity from nonresponse, baseline anchoring, or a failed questionnaire path.","affected_objective":"Preserve independently reasoned, uncertainty-aware expert judgments across Delphi rounds while concentrating limited panel and facilitator attention on consequential belief revisions without suppressing dissent, ambiguity, or protected concerns.","intervention":"Replace selected full-repeat Delphi rounds with a governed residual-round protocol. For every expert-proposition pair, the system freezes a versioned expected next-response vector before the round: probability or range, timing quantiles, confidence, dependency judgments, and rationale tags, usually initialized from the last verified response and adjusted only by a declared prediction rule. The expert receives the matching baseline and must actively confirm it or submit signed and categorical corrections, including changed rationale. The portal reconstructs the complete current response as baseline plus residual and shows it to the expert for confirmation before acceptance. A precision-and-consequence rule requires fuller explanation for large, confident, high-consequence, minority, or dependency-changing residuals; it does not rank experts by conformity. Nonresponse remains an explicit missing state, never an implied confirmation. Random propositions are answered from scratch before the baseline is revealed, every designated high-stakes proposition travels in full, and scheduled full-response rounds resynchronize the panel record. Version mismatch, unexpected residual structure, audit disagreement, increased uncertainty, high disagreement, or reconstruction failure switches the affected proposition or expert to a full questionnaire. Validated residuals update the prediction model for later rounds only through an attributable review; they do not automatically change the panel conclusion or sponsor decision.","structural_mapping":[{"archetype_element":"Prediction target and observation boundary","domain_realization":"The target is an individual expert's next-round structured judgment for a defined proposition and horizon, not the eventual future event or a manufactured consensus."},{"archetype_element":"Generative model state","domain_realization":"A versioned participant-proposition baseline predicts probability, timing, confidence, dependencies, and rationale tags, with uncertainty about each predicted field."},{"archetype_element":"Model scope and horizon","domain_realization":"Each baseline is authorized only for the named expert, proposition, elicitation round, response schema, and forecast horizon and expires when any of those change."},{"archetype_element":"Expected and actual behavior","domain_realization":"The frozen expected response exists before the round; the expert's actively confirmed or corrected current judgment is the observation."},{"archetype_element":"Prediction comparator","domain_realization":"A field-aware comparator preserves signed probability and timing changes, confidence direction, added or removed dependencies, categorical reversals, and rationale changes."},{"archetype_element":"Prediction-error signal","domain_realization":"The residual contains only confirmed changes plus enough baseline identity and context to reconstruct the expert's complete response."},{"archetype_element":"Precision weighting and residual budget","domain_realization":"Routing and explanation requirements consider change magnitude, stated confidence, proposition consequence, disagreement, minority status, and facilitator capacity while preserving all submitted residuals in the record."},{"archetype_element":"Residual propagation channel","domain_realization":"Facilitators receive reconstructed response vectors and a prioritized residual-round digest rather than manually comparing repeated full questionnaires."},{"archetype_element":"Confidence and uncertainty state","domain_realization":"Expert confidence, baseline-prediction uncertainty, response completeness, and facilitator confidence in reconstruction remain distinct."},{"archetype_element":"Update rule","domain_realization":"Reviewed prediction errors may revise the next-round response predictor, uncertainty, thresholds, or proposition boundaries, but do not overwrite the expert's submitted judgment."},{"archetype_element":"Model synchronization and provenance","domain_realization":"The expert interface and facilitator record use the same schema and baseline checksum; each residual carries participant pseudonym, proposition, round, version, timestamp, and transformation history."},{"archetype_element":"Freshness and drift","domain_realization":"Baselines expire by round, while systematic audit changes, correlated residuals, or rising disagreement indicate that last-round responses no longer predict the panel's judgment process."},{"archetype_element":"Reconstruction requirement","domain_realization":"Before submission, the expert verifies the complete response reconstructed from the frozen baseline and residual; audit software independently repeats the reconstruction."},{"archetype_element":"Raw audit and resynchronization","domain_realization":"Random propositions are answered from scratch without baseline exposure, and scheduled rounds collect complete responses from the relevant panel."},{"archetype_element":"Fallback and decompression","domain_realization":"Full questionnaires replace residual entry for stale or incompatible baselines, reconstruction errors, unusual residual patterns, audit disagreement, or participant request."},{"archetype_element":"Safety-critical bypass","domain_realization":"Propositions involving physical safety, rights, irreversible harm, severe distributional consequences, protected disclosures, or unusually high disagreement require complete judgments and rationales."},{"archetype_element":"Attention budget","domain_realization":"The protocol explicitly budgets expert response minutes, facilitator comparison time, audit work, model maintenance, and fallback administration rather than optimizing only questionnaire length."}],"mechanism_mapping":[{"mechanism_slug":"predictive_codec","role":"Maintains matched response baselines in the expert interface and facilitator record so a complete judgment can be reconstructed from prediction plus correction.","counterfactual_removal":"Without matched prediction and reconstruction, the protocol becomes survey skip logic rather than predictive residual processing."},{"mechanism_slug":"delta_or_differential_encoding","role":"Represents signed probability, timing, confidence, dependency, and rationale changes instead of retransmitting stable fields.","counterfactual_removal":"Every round would still carry complete judgments, leaving the repetitive attention problem unchanged."},{"mechanism_slug":"model_version_checksum_handshake","role":"Verifies the proposition wording, response schema, prior response, and prediction rule before accepting a residual.","counterfactual_removal":"A correction could be applied to a revised question or obsolete baseline and silently reconstruct the wrong judgment."},{"mechanism_slug":"event_triggered_residual_reporting","role":"Surfaces material belief changes to facilitators while requiring an explicit no-change confirmation and maintaining missingness separately.","counterfactual_removal":"Facilitators would either inspect all fields equally or misread silence as confirmed stability."},{"mechanism_slug":"precision_weighted_error_gate","role":"Allocates explanation and review effort according to confidence, consequence, disagreement, source expertise, and minority-position protection rather than change magnitude alone.","counterfactual_removal":"Large uncertain changes could crowd out small but well-supported or high-consequence revisions."},{"mechanism_slug":"confidence_threshold_table","role":"Versions the rules for confirmation, expanded rationale, audit, facilitator review, and full-response fallback by proposition risk and uncertainty.","counterfactual_removal":"Escalation would depend on ad hoc facilitator judgment and could drift toward a convenient response-volume target."},{"mechanism_slug":"shadow_raw_channel_sampling","role":"Assigns random and risk-stratified propositions to from-scratch response before baseline exposure and compares them with residual-mode reconstructions.","counterfactual_removal":"The protocol could not detect changes suppressed by anchoring to the predicted response."},{"mechanism_slug":"periodic_full_state_resynchronization","role":"Collects scheduled complete responses that reset each expert-proposition baseline and bound accumulated semantic error.","counterfactual_removal":"Small reconstruction, rationale, or schema errors could persist across all later rounds."},{"mechanism_slug":"model_drift_monitoring","role":"Tracks residual distributions, missingness, audit disagreement, confidence shifts, and baseline expiry by proposition and participant segment.","counterfactual_removal":"A panel-wide reconsideration could be treated as isolated edits under a stale prediction model."},{"mechanism_slug":"raw_signal_fallback_switch","role":"Restores complete questionnaire entry when validity, compatibility, audit, safety, or reconstruction conditions fail or when an expert requests it.","counterfactual_removal":"The residual interface could continue anchoring or truncating judgments after its assumptions cease to hold."},{"mechanism_slug":"prediction_error_review","role":"Determines whether each material miss arose from a changed expert belief, ambiguous proposition, controlled feedback, prediction error, interface effect, or boundary failure.","counterfactual_removal":"The prediction model could absorb interface-induced behavior and mistake it for genuine expert learning."},{"mechanism_slug":"bayesian_model_update","role":"Updates uncertainty about the response predictor after reviewed residuals while retaining uncertainty and sensitivity to alternative priors.","counterfactual_removal":"Prediction revisions would be informal and could react equally to reliable belief change, noise, or questionnaire artifacts."}],"causal_chain":["The facilitator defines proposition-specific response vectors, horizons, protected classes, and attention budgets.","Before each residual round, the system freezes a versioned prediction of every in-scope expert-proposition response.","Experts actively confirm or correct predicted fields, while missing submissions remain explicit and selected questions are answered from scratch.","The comparator encodes signed and categorical changes with provenance rather than retransmitting stable fields.","The matched facilitator record reconstructs each complete response and the expert confirms the reconstruction.","Precision and consequence rules concentrate rationale and facilitator review on material changes, disagreements, and protected minority positions.","Controlled feedback uses verified complete responses, while reviewed residuals separately teach the next-round prediction model.","From-scratch audit questions and scheduled full rounds test for anchoring, semantic loss, and accumulating baseline divergence.","Drift, incompatibility, audit disagreement, safety classification, or participant request restores full-response collection for the affected scope."],"baseline":"A conventional Delphi protocol collects a complete structured response and rationale for every proposition in each round, provides controlled aggregate feedback, and asks the panel to answer the full instrument again. Facilitators compare rounds after collection, but no forward response model determines representation, reconstruction, audit sampling, or fallback.","nearest_rivals":["A standard full-round Delphi process; it preserves complete responses but repeatedly spends attention on stable fields.","An adaptive survey with branching and prefilled answers; it can shorten a questionnaire but does not require a versioned predictive baseline, reconstructive residual, drift monitoring, independent raw audit, or full-state resynchronization.","A tracked-changes questionnaire showing edits from the previous round; it records differences after the fact but lacks a scoped next-response prediction, uncertainty weighting, model update, and governed fallback.","A consensus summary that asks only dissenters to comment; it centers deviation from the group rather than deviation from each expert's own predicted judgment and can suppress stable minority positions.","Bayesian expert aggregation; it can combine judgments into a posterior but does not reduce repeated elicitation through model-relative residual communication and synchronized reconstruction.","A facilitator's manual list of changed answers; it identifies revisions after full collection and therefore does not reallocate expert response effort."],"remaining_contrastive_claim":"The proposal should be preferred only if a versioned participant-specific response prediction plus residual can reconstruct complete expert judgments while reducing total panel-and-facilitator effort, and if from-scratch audits and full-round resets show that anchoring, minority suppression, rationale loss, and maintenance costs remain within predeclared tolerances.","authority_safety":{"decision_authority":"The elicitation-method owner may define pilot propositions, schemas, baseline rules, and audit cadence. Each expert retains authority over their judgment and may demand full-response mode. An independent governance reviewer approves protected classes and anchoring tests. Decision sponsors retain authority over any strategy or policy informed by the panel.","authorized_first_step":"Run one bounded, non-decision-bearing crossover exercise in which participants assess a fixed set of mock foresight propositions over two rounds. Randomly assign proposition blocks to conventional full re-entry or residual entry, retain a from-scratch audit subset, and prevent outputs from informing live strategy.","excluded_actions":["Using the residual protocol for a live high-consequence decision before the bounded test and governance review","Treating nonresponse as confirmation of a predicted judgment","Automatically moving an expert's estimate toward the panel median","Ranking or excluding experts because their residuals are frequent, unusual, or dissenting","Revealing expert identity through distinctive residuals or rationale text","Changing proposition wording, response schemas, or baseline parameters without a new version","Automatically updating official forecasts, scenarios, policies, or resource allocations","Suppressing complete responses for protected or high-disagreement propositions","Using reduced questionnaire length to add unbudgeted propositions during the test"],"halt_rollback":"Halt residual mode for an affected proposition or participant if a reconstruction mismatch occurs, the audit subset reveals an unreported material change, anonymity is weakened, a protected response is compressed, version compatibility fails, or crossover workload exceeds its declared budget. Roll back by collecting the complete response, restoring the last verified baseline and threshold versions, preserving all audit traces, and withholding the affected aggregate until independent review."},"negative_tests":{"strongest_counterevidence":"In the crossover exercise, residual entry produces fewer independently reconsidered changes, weaker or less complete rationales, greater convergence toward displayed baselines, worse preservation of minority positions, or no reduction in total response and administration effort after audits, model maintenance, and fallback are counted.","problem_falsifier":"The problem is falsified for the selected elicitation if most response fields materially change between rounds, full questionnaires fit comfortably within the declared attention budget, repeated entry improves reasoning or error discovery, or facilitator comparison is not a meaningful constraint.","intervention_falsifier":"The intervention is falsified if baseline-plus-residual reconstruction fails semantic equivalence checks; from-scratch questions uncover material changes absent from residual-mode answers; residual entry increases anchoring relative to full entry; protected, dissenting, or high-uncertainty judgments lose rationale or visibility; missingness becomes ambiguous; or total operating cost is not lower at the required fidelity.","risks":["Displayed predictions may anchor experts to their prior judgments or to facilitator assumptions.","A shared response predictor may encourage artificial convergence and model lock-in.","Precision weights may privilege credentialed or majority viewpoints and suppress unfamiliar expertise.","Distinctive residuals may weaken Delphi anonymity.","Experts may game change thresholds to obtain attention or avoid follow-up questions.","Compact residual rationales may omit causal context necessary to interpret a changed estimate.","Frequent fallback may erase any attention benefit and encourage weakening the safeguards.","Prediction updates may learn effects caused by the interface or controlled feedback rather than independent belief change.","Version mismatch may apply a valid correction to the wrong proposition wording or response scale.","Saved response capacity may be filled with additional propositions, eliminating the intended attention budget." ]},"next_evidence_step":"Pre-register a single two-round, non-decision-bearing crossover using a fixed set of mock long-horizon propositions. Freeze each round-two baseline before presenting it; assign matched proposition blocks to full re-entry or residual entry; require active confirmation; and collect a concealed from-scratch audit subset before revealing predicted values. Compare total response and facilitator time, exact and semantic reconstruction agreement, frequency and direction of revisions, blinded rationale-quality ratings, retained disagreement and minority positions, missingness, audit-discovered changes, fallback use, and maintenance effort. Define material-change, anchoring, anonymity, protected-signal, and halt thresholds before examining results.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 addresses external evidence throughput in a distributed horizon-scanning network: it predicts scenario-driver states, encodes new observations as residuals, and reallocates a central review queue. This proposal addresses repeated human judgment elicitation inside a Delphi process: it predicts each expert's next-round response, reconstructs complete judgments from corrections, and reallocates participant reconsideration and facilitator-comparison effort. Its unit of prediction is an expert-proposition response rather than an environmental driver assessment; its primary failure test is anchoring and loss of dissent rather than missed source evidence; and its full-state control is a complete response round rather than a full scanning packet. It can be adopted as a standalone Delphi protocol without operating a horizon scan or using Proposal 1's evidence-routing system.","revision_record":{"parent_version":null,"progress_targets_addressed":["Created an independently adoptable second candidate with a different problem, intervention, actors, decision process, and causal path from proposal 1.","Preserved prediction, comparison, residual communication, reconstruction, model update, synchronization, raw audit, and fallback as governing structure."],"conceptual_changes":["Moved the prediction target from external scenario-driver evidence to longitudinal expert judgments.","Made attention savings arise from residual elicitation across Delphi rounds rather than residual triage across scanning sources.","Added interface-induced anchoring as a central model-failure pathway."],"operational_changes":["Specified participant-proposition baselines, active confirmations, full-response reconstruction, from-scratch audit questions, protected full-answer classes, scheduled full rounds, and participant-requested fallback.","Bounded the first test to a non-decision-bearing crossover exercise."],"evidence_changes":["Added direct tests of reconstruction, anchoring, rationale preservation, disagreement, anonymity, workload, and model-maintenance cost.","Defined problem and intervention falsifiers specific to repeated expert elicitation."],"claim_changes":["Made no claim of novelty, prevalence, demand, or effect size.","Conditioned adoption on attention cost, semantic fidelity, independent reconsideration, dissent protection, and audit performance."]}}