Entity-aware pathology workflow informatics: Structured evidence mapping and a reference architecture for governed orchestration

Objective To map evidence relevant to representing and governing pathology workflows spanning cases, specimens, blocks, slides, scans, and reports, and to derive a reference architecture. Methods We performed structured evidence mapping with purposive horizon scanning. Searches of Scopus, PubMed, and IEEE Xplore yielded 74 primary database records; Google Scholar and citation tracking contributed 24 supplementary records. Each record contributed one tuple across data acquisition, control-flow reconfiguration, execution autonomy, and validation environment (D1–D4), an author-developed operational taxonomy. Unretained historical search details are reported as a reproducibility constraint, not reconstructed as original methods. Results The 98-record evidence base spans 1993–2026. Within the 74-record primary corpus, 47 records had dashboard-level execution; validation was split between real-world deployment (26) and conceptual evidence (25). Under study-defined filters, 8/74 records met all 4 minimum capability conditions, 2/74 also showed closed-loop physical actuation, and 1/74 combined full autonomy with real-world deployment. These study-defined filters derive from the original single-reviewer D1–D4 coding. An independent blinded 50-record verification yielded 81.6% inclusion agreement (κ = 0.627) and 80.0% exact D1–D4 field agreement, but did not constitute full-corpus independent reproduction. Supplementary records informed emerging automation requirements. The mapped gaps informed an architecture comprising semantic perception, entity-aware process state, bounded orchestration, pre-action command governance, governed actuation, and conditional local execution support; a safety-and-constraints sandbox is one candidate realization of the governance function. Conclusion The architecture is an evidence-derived design target, not a deployed or validated system. Evaluation should test event provenance, case-to-slide relationships, robustness to missing or shifted inputs, command constraints, human override, and staged operational performance before autonomy claims.