Every finding we have published, with its caveat

Newest first. Each entry says how many people or items it rests on and what it cannot show. Superseded versions stay listed beside the version that corrects them.

Releases, newest first

  • 2026-10-27. CRLN Findings, October 2026: most misses were partial, not unsafe. Size: 91 learners' scored practice decisions, July 13 to October 10, 2026. Caveat: Practice decisions in the Practice Lab, not workplace performance. Headline revised October 27, 2026: the earlier title generalized from two decisions. Dated web edition, no DOI.
  • 2026-10-21. CRLN-PRE-002 results, version 1.0.0: do AI models say when they are unsure?. Size: 200 keyed items, each shown in 3 option arrangements, pre-registered. Caveat: A different measure on a different item set from CRLN-AIE-001. The practitioner comparison is exploratory, not matched. DOI: 10.5281/zenodo.23544672.
  • 2026-10-12. CRLN Benchmark, edition 2026-10: how trial-operations AI decides, compared with Practice Lab learners. Size: AI models from three vendors and a pooled Practice Lab learner baseline, on a small public set of site decisions. Caveat: The decisions are from the public practice catalog, so a model may have seen them. A high score here is not evidence about decisions outside this set. Dated web edition, no DOI.
  • 2026-09-29. CRLN-AIE-001 version 0.2.1: correction. Size: A correction only, no new data. Caveat: Retires the held-out set described in version 0.2, whose answers could be read without an account, and withdraws the one result that rested on it. DOI: 10.5281/zenodo.23043069.
  • 2026-09-29. CRLN-TRG-001 release 1.7.4, our own open standard, built on ICH E6(R3). Size: A competency framework for fourteen clinical research roles, CC BY 4.0. Caveat: Our own standard, not issued or approved by any outside body. Its jurisdiction crosswalk is provisional. DOI: 10.5281/zenodo.23040215.
  • 2026-09-12. CRLN-AIE-001 version 0.2: AI models and practitioners on the same decisions. Size: Seven AI models and practitioners on the same 114 clinical trial conduct decisions. Caveat: Reversed the conclusion of version 0.1. The models met the standard on 95.1 to 98.9 percent of responses, near the ceiling where this instrument can tell them apart. DOI: 10.5281/zenodo.22728905.
  • 2026-09-11. CRLN-AIE-001 version 0.1 (superseded). Size: Kept as a citable record. Caveat: Superseded. It compared an unweighted model mean with a weighted human mean, and version 0.2 reversed its conclusion. None of its figures are current. DOI: 10.5281/zenodo.22708752.

Standing records, kept current

  • CRLN-HB-001: the human baseline. Size: The practitioner baseline CRLN-AIE-001 compares models with, published separately. Caveat: The concept DOI always resolves to the newest version; cite a version DOI to pin one. DOI: 10.5281/zenodo.22690692.
  • CRLN-TRG-001, every release. Size: The concept DOI, which always resolves to the newest release. Caveat: Releases change; the change log names what changed in each. DOI: 10.5281/zenodo.22049549.
  • Who uses the standard: the adopters register. Size: Organizations that have told us they use CRLN-TRG-001, each listed with consent. Caveat: Self-declared and not audited by CRLN. A floor, not a count of everyone who uses it. Dated web edition, no DOI.
  • Validation ledger. Size: A live count of assessment items reviewed and passed by independent practicing professionals, by role. Caveat: Individual items pass review one at a time. It is not a claim about the assessment as a whole. Dated web edition, no DOI.
  • Methodology. Size: How readiness is scored and aggregated. Caveat: Aggregates are reported only for groups of five or more people. Dated web edition, no DOI.
  • Corrections log. Size: Every defect found in a number we reported. Caveat: With the figure before and after, and how each was caught. Dated web edition, no DOI.