{"actors":["Discovery-services librarian","Metadata librarian","Systems librarian","Public-services staff member who receives search-failure reports","Library patron represented through de-identified query traces"],"affected_objective":"Restore findability for materials implicated in unexplained zero-result searches while avoiding unsupported conclusions about patron intent or metadata fault.","arm":"COMMON_P1","authority_safety":{"authorized_first_step":"The discovery-services librarian may create a de-identified case record, inspect existing search and indexing traces, run non-public test queries, and recommend a reversible probe.","decision_authority":"The metadata librarian authorizes catalog-record changes; the systems librarian authorizes index or interface changes; collection or access-policy decisions remain with their existing owners.","excluded_actions":["Do not identify or contact a patron from query traces.","Do not infer protected traits, intent, or competence from search language.","Do not bulk-edit metadata, suppress records, change ranking globally, or alter access policy on the strength of a working explanation.","Do not represent the selected explanation as a confirmed cause."],"halt_rollback":"Stop the probe if it exposes identifiable query data, changes public results beyond the bounded test, or produces evidence inconsistent with the working explanation; revert any test configuration and reopen the ranked explanation set."},"baseline":"Staff handle zero-result reports case by case, often moving directly from the visible failed query to a familiar remedy such as adding a synonym, correcting a record, or blaming indexing, without a shared record of observations, rivals, selection rationale, or revision triggers.","candidate_id":"abductive_explanation_selection__library_information_science__COMMON_P1","causal_chain":["A patron query returns zero results even though staff can identify one or more plausibly relevant holdings.","The system records the query, timestamp, interface, filters, result count, candidate holdings, and index state separately from staff interpretations.","A librarian constructs a bounded explanation set including vocabulary mismatch, metadata omission, indexing lag, interface or filter state, and a lower-probability access-control or record-suppression fault.","Candidates are compared for trace coverage, timing, mechanism plausibility, background frequency within local cases, missing evidence, and predictions about controlled test queries.","A best-so-far explanation is selected only if it materially distinguishes itself; otherwise the case remains an unresolved ranked set.","The selected explanation receives a provisional status, an action limit, and explicit observations that would weaken or overturn it.","A low-regret discriminating probe is run against a test index or non-public query path.","Probe results update the explanation record and route any justified correction to the role authorized to make it."],"cell_id":"abductive_explanation_selection__library_information_science","consequence":"Without an auditable explanation-selection step, a plausible first story can direct metadata or system work toward the wrong layer, leave the actual findability failure intact, and make later contradictory evidence difficult to interpret.","diversity_from_prior_proposals":"Not assessed against other proposals under runtime isolation; this candidate is internally distinguished by treating a failed retrieval event as a defeasible explanation case rather than immediately as a cataloging defect.","experiment_id":"eoa_inverse_innovation_exp13_second_slot_policy60_20260806","intervention":"Install a Zero-Result Explanation Case protocol in the discovery-service workflow. For each bounded reported failure, staff record the explanandum and raw traces, enumerate required rival explanations, score their explanatory fit in a small matrix, label the best account as working or preserve a tie, specify action limits and revision triggers, and run one reversible query or index probe chosen to discriminate among the leading accounts.","mechanism_mapping":[{"counterfactual_removal":"Without the matrix, staff can favor a familiar explanation without showing how it accounts for timing, filters, metadata, and rival traces.","mechanism_slug":"inference_to_best_explanation_matrix","role":"Compares the bounded candidate explanations against explicit retrieval-specific fit criteria and records why a leader does or does not outrank its nearest rival."},{"counterfactual_removal":"Without the memo, observations, interpretations, confidence, and authority limits can be lost during handoff between service, metadata, and systems staff.","mechanism_slug":"explanatory_case_memo","role":"Preserves the failed-query observation, candidate set, provisional selection, unresolved evidence, ownership, and permissible next action."},{"counterfactual_removal":"Without a disconfirming probe, a vocabulary-mismatch story can become self-sealing because every failed query is treated as further support for it.","mechanism_slug":"disconfirming_probe_plan","role":"Derives a controlled test whose possible outcomes distinguish the working explanation from metadata, indexing, interface-state, and access-control rivals."},{"counterfactual_removal":"Without an evolving log, later index refreshes or record changes cannot be related reliably to the explanation that motivated them.","mechanism_slug":"abduction_log","role":"Tracks changes to the ranked explanations as new traces and probe outcomes arrive."}],"nearest_rivals":["Automatic zero-result alerts, which identify retrieval failures but do not compare explanations for them.","A standing synonym or query-expansion list, which applies a predefined remedy without establishing whether vocabulary mismatch explains the case.","Routine metadata-quality review, which can detect record defects but may overlook filter state, indexing timing, interface behavior, or access-control causes.","Usability testing, which studies interaction patterns but does not provide a case-level best-so-far causal account for an operational failure."],"negative_tests":{"intervention_falsifier":"Across a bounded retrospective sample, blinded reviewers using the protocol do not choose more discriminating next probes or revise explanations more consistently than reviewers given the same traces in the existing workflow.","problem_falsifier":"Existing case records already separate observations from interpretations, preserve credible rivals, document why one account wins, constrain action by confidence, and specify revision triggers; the inferred structural problem would then be absent.","risks":["The fit matrix may create false precision from qualitative judgments.","Staff may treat the highest-ranked explanation as confirmed despite its provisional label.","Required documentation may delay handling of simple cases.","Query traces may reveal sensitive interests if de-identification and access controls fail.","Base-rate-common explanations may crowd out a rare access or suppression fault with greater downside.","Organizational authority may influence scoring while appearing to be explanatory fit."],"strongest_counterevidence":"If independently reproduced test queries show that a deterministic, documented rule maps the observed failure to one known catalog or index condition with no credible rival, abductive selection adds little and a rule-based diagnostic is more appropriate."},"next_evidence_step":"Select a bounded set of 12 previously resolved, de-identified zero-result cases spanning metadata, indexing, interface-state, and unresolved outcomes. Give two reviewers only the traces available at the original decision time; have each complete the protocol, then compare their selected probes and revision triggers with the subsequently observed resolution. This is a read-only retrospective exercise and authorizes no catalog or production-system changes.","observable_state":"A de-identified case record shows the exact query string or normalized token sequence, timestamp, discovery interface, active filters, result count, candidate known-item identifier when available, relevant metadata fields, index-refresh status, and staff report. Interpretations such as “terminology mismatch” or “bad metadata” are stored separately.","prior_art_status":"UNSEARCHED","problem":"When a discovery search unexpectedly returns no results for material that appears to be held, library staff must infer which hidden condition produced the failure from incomplete traces. Vocabulary mismatch, missing or malformed metadata, stale indexing, interface state, access controls, and genuine collection mismatch can look alike at the point of service. The first plausible explanation can therefore harden into an operational conclusion before rivals are compared.","proposal_index":1,"remaining_contrastive_claim":"The candidate specifically adds case-level selection among competing hidden causes, with defeasibility and discriminating probes; it is not merely failure detection, generic metadata auditing, query expansion, or alternative listing.","revision_record":{"claim_changes":["Initial version; no claims revised."],"conceptual_changes":["Initial version instantiates abductive explanation selection around unexplained zero-result discovery searches."],"evidence_changes":["No external evidence consulted; prior art remains unsearched."],"operational_changes":["Initial version bounds the first step to a read-only retrospective review of 12 de-identified cases."],"parent_version":null,"progress_targets_addressed":["Concrete explanandum and observable traces","Explicit rival set and fit comparison","Provisional selection or preserved tie","Authority-calibrated reversible action","Falsifiers, risks, and bounded evidence step"]},"schema_version":1,"structural_mapping":[{"archetype_element":"Surprising Observation Record and Explanandum Boundary","domain_realization":"A discovery query produced zero results despite an identified plausibly relevant holding; the bounded target is that query-result discrepancy, not a general judgment about the catalog or patron."},{"archetype_element":"Observation–Interpretation Separation","domain_realization":"Query, filters, result count, record fields, and index timestamps are recorded as traces; claims such as vocabulary mismatch or indexing failure are recorded as explanations."},{"archetype_element":"Candidate Explanation Set","domain_realization":"The set includes a leading account, common vocabulary and metadata causes, an interface-state rival, and a lower-probability access-control or suppression fault that would change action if true."},{"archetype_element":"Explanatory Fit Criteria","domain_realization":"Candidates are compared on coverage of observed traces, temporal alignment, mechanism plausibility, local background frequency, required ad hoc assumptions, missing evidence, and discriminating predictions."},{"archetype_element":"Best-So-Far Explanation or Tie","domain_realization":"The case records either one materially better working explanation or a ranked unresolved set when the traces do not discriminate."},{"archetype_element":"Defeasibility Status and Action–Confidence Boundary","domain_realization":"The output is labeled working, contested, or unresolved and may authorize inspection or a test query, but not an irreversible metadata, ranking, access, or collection decision."},{"archetype_element":"Discriminating Evidence Plan","domain_realization":"A controlled query variant, test-index refresh, filter replay, or field-level record inspection is chosen for its ability to separate the leading explanation from its nearest rival."},{"archetype_element":"Revision Trigger and Explanation Owner","domain_realization":"A named discovery-services librarian must reopen the case when a prediction fails, a rival gains supporting traces, or the selected account requires an unrecorded exception."}],"title":"Zero-Result Explanation Cases for Library Discovery","version":0}