Reproducibility archive for "When Transparency Warrants Trust: A Systematic Review of Consumer Reliance on AI Recommendations"

This archive contains the complete evidence base, decision record and analysis code for asystematic review of how information about AI recommendations affects consumer trust,reliance and calibration. CONTENTS Search and screening open_search_log_COMPLETE.csv All 103 API requests across topical and expansion pathways, with timestamps, reported totals and saved counts search_queries.json The 12 topical query strings as executed sentinel_recall.csv Recall check against 12 known-relevant reports records_deduplicated.csv 1,889 deduplicated topical records citation_expansion_raw.csv 3,231 citation-expansion records open_evidence_master.csv 4,093 unique master records open_evidence_priority.csv 1,162 records prioritised by rule-based ranking direct_evidence_tier.csv 259 records assessed at title/abstract AI_Trust_Narrowed_Corpus_screening.xlsx 100-record reconciliation, codebook, protocol lock stage3_adjudication_registry.csv 38 reports sought in full text, with adjudication status Evidence and appraisal combined_study_registry.csv 33 studies with design, sample, exposure family, outcomes, dependence structure, source anchors and appraisal combined_effect_registry.csv 32 exact effects on their valid published or calculated scales risk_of_bias_registry.csv Six-domain design-appropriate appraisal for all 33 studies report_family_map.csv 27 reports mapped to 33 studies, with version linkage included_reference_manifest.csv Bibliographic record for all 27 included reports EVIDENCE_BASE_LOCK.md Corpus lock FINAL_POOLING_DECISION.md Quantitative-synthesis gate and reopening rule RISK_OF_BIAS_REPORT.md Appraisal narrative Code scripts/ Regenerates the manuscript's Tables 1-3 and Figure 2 from the registries, and re-verifies every reported count and effect KEY FIGURES 27 reports, 33 studies, 8,822 nominal participants, 32 exact effects.Appraisal: 6 some concerns, 25 high risk, 2 critical; no study rated low risk.No meta-analysis was conducted. On a common standardised scale the corpus supplies nineHedges' g values drawn from three independent samples. REPRODUCIBILITY Every count reported in the article can be regenerated from these files. The selection flowreconciles exactly: 1,889 topical records plus 3,231 citation-expansion records give the 5,120captured; 4,093 remain after deduplication; 1,162 prioritised; 259 assessed at title/abstract;100 reconciled; 38 sought in full text; 23 assessed and 15 unresolved; 12 eligible through thecore route plus 15 by citation chaining, giving 27 reports and 33 studies. LIMITATIONS OF THE UNDERLYING SEARCH Users should read these before reusing the corpus. The search used open scholarlyinfrastructure and does not claim coverage of Scopus, Web of Science, PsycINFO, BusinessSource Complete, the ACM Digital Library or IEEE Xplore. Ninety of the 103 runs retrieved asingle relevance-ranked page of 100 records and stopped. A sentinel check of 12known-relevant reports recovered seven. Twenty of the 23 records classified as direct evidencecould not be obtained in full text. Screening and extraction were reconciled against fixedrules but were not duplicated by two independent reviewers, and no inter-rater agreementstatistic is available. The review was not prospectively registered. WHAT IS NOT INCLUDED No copyrighted article full texts are redistributed. Verbatim publisher abstracts have beenremoved from the record-level files; each row retains its DOI, from which the abstract can beretrieved directly. Local file paths have been removed. LICENCE Data and documentation: CC BY 4.0. Code in scripts/: MIT. See LICENSE.txt.Bibliographic metadata originates from Crossref and OpenAlex. RELATED ARTICLE 10.5281/zenodo.21954317