Database-specific Boolean search strings (term groups defined below; search reproducibility companion)
Paper: Evidence for Whom? Contestability and Socio-Technical Assurance in High-Risk AI Governance
Last search: June 2026. Window: 2015-2026 (LLM/agentic emphasis 2022-2026). Language: English.

Term groups
  SYSTEM   = ("artificial intelligence" OR "machine learning" OR "deep learning" OR
              "large language model" OR LLM OR "foundation model" OR "AI agent" OR agentic)
  EVAL     = (evaluation OR assessment OR audit OR benchmark OR testing OR assurance OR
              "conformity assessment" OR verification OR validation)
  METHOD   = (fairness OR bias OR robustness OR adversarial OR safety OR "red team" OR
              "red-teaming" OR interpretability OR explainability OR calibration OR
              uncertainty OR "data governance" OR "model card" OR datasheet OR
              "human oversight" OR monitoring OR "drift detection")
  GOV      = ("NIST AI RMF" OR "AI Risk Management Framework" OR "EU AI Act" OR
              "Regulation (EU) 2024/1689" OR "ISO/IEC 42001" OR "conformity assessment" OR
              "high-risk AI")

--------------------------------------------------------------------------------
IEEE Xplore (Command Search; metadata)
  ("All Metadata":SYSTEM) AND ("All Metadata":EVAL) AND ("All Metadata":METHOD)
  Filters: 2015-2026; English. Governance pass: ("All Metadata":GOV) AND ("All Metadata":EVAL).

ACM Digital Library (Advanced Search; Abstract)
  Abstract:(SYSTEM) AND Abstract:(EVAL) AND Abstract:(METHOD)
  Filters: 2015-2026.

Scopus (TITLE-ABS-KEY)
  TITLE-ABS-KEY(SYSTEM) AND TITLE-ABS-KEY(EVAL) AND TITLE-ABS-KEY(METHOD)
  AND PUBYEAR > 2014 AND ( LIMIT-TO(LANGUAGE,"English") )
  Governance pass: TITLE-ABS-KEY(GOV) AND TITLE-ABS-KEY(EVAL).

Web of Science (TS = Topic)
  TS=(SYSTEM) AND TS=(EVAL) AND TS=(METHOD)
  Timespan 2015-2026; document types: Article, Review, Proceedings Paper.

arXiv (cs.LG, cs.AI, cs.CY, cs.CR, stat.ML; via API/listing)
  (abs:SYSTEM AND abs:EVAL AND abs:METHOD) in the listed categories; 2015-2026.
  Governance pass: abs:GOV AND abs:EVAL.

Snowballing: forward/backward citation tracing of included records via Google Scholar.
Official corpus (hand-collected, not via the above queries): NIST AI 100-1, NIST AI 600-1,
EUR-Lex Reg. (EU) 2024/1689, ISO/IEC 42001:2023, ISO/IEC 23894:2023.
Reused corpus: a previously published, arXiv-verified agent-fairness corpus, reused and re-verified for this review.

--------------------------------------------------------------------------------
Search reproduction (re-run June 2026, for transparency)
  arXiv (cs.LG, cs.AI, cs.CY, cs.CR, stat.ML), governance x evaluation query
    (GOV) AND (EVAL):  168 records (via the arXiv API).
  The broad method query (SYSTEM AND EVAL AND METHOD) returns a much larger
    pre-screening universe; the verified corpus reflects the structured per-family
    search followed by title/abstract and full-text screening (see PRISMA figure).
  Scopus (institutional access, June 2026):
    main (SYSTEM AND EVAL AND METHOD): 228,040 records (exceeds the 20,000 export cap)
    governance (GOV AND EVAL):         1,423 records  -> search_exports/scopus_governance.csv
  Web of Science (institutional access, June 2026):
    main (TS: SYSTEM AND EVAL AND METHOD): 165,342 records (exceeds export cap)
    governance (TS: GOV AND EVAL):         832 records   -> search_exports/wos_governance.csv
  The broad main query is a pre-screening universe; the verified corpus reflects the
    structured per-family search + screening (see PRISMA figure), not the broad query.
  IEEE Xplore and ACM Digital Library result counts vary with the web UI; the
    per-database candidate yields from the original search were not separately retained.
  These re-run counts are a reproducibility snapshot, not the original screening counts.
