Research on clinician scheduling: data files https://capillaryhealth.com/data Data release 0.6.0, 2026-11-14. PubMed searched on 2026-11-12. List of topics, version 1.3.0. Courtesy of the U.S. National Library of Medicine. The counts reflect PubMed on the day it was searched, not papers added since. The National Library of Medicine does not endorse these files or Capillary Health. WHAT IS COUNTED A paper names a topic when its title uses one of the topic's phrases, or its abstract uses them at least twice. For most topics it must also name health care staff or a hospital, or use the word shift, roster or rota (the scope column). The files call each topic a practice, in their names and their columns. A paper's year is the year of its journal issue, and its types are the ones PubMed gives it. HOW THE COUNTS WERE CHECKED For each topic, 40 papers were drawn at random with the seed 20261112 from those that name it, and read. A count shows only when at least 32 of 40 are on topic. A topic with fewer papers has all of them read. The reading rule: A paper is on topic when it studies, reviews or reports on the practice as used for health staff. A paper that uses the phrase with another meaning, or mentions it only in passing, is off topic. A second, separate reading of 225 of the same papers agreed with the first on 219. TRIALS trials.csv lists each trial that compared one way of scheduling or staffing clinicians with another: the length of a shift, the kind of call, the rest between shifts, a planned nap, the way handoffs are done or the number of staff on duty. A trial of a drug, a light, a training course or a sleep aid among shift workers is not in it. The studies came from three places: the 443 papers that name a topic and that PubMed lists as trials; the 602 studies on ClinicalTrials.gov whose title or intervention names a topic; and the 42 trials that cite one of nine well-known schedule trials. A paper and a ClinicalTrials.gov record count as one study when either names the other: 1,024 studies. An AI model (Claude, made by Anthropic) read each study three times against a fixed list of questions, from its ClinicalTrials.gov record and the titles and abstracts of its papers. Every answer carries a quote, and code checks that the quote is in the study's text; an answer stands when two of the three readings give it. Each study's three readings make three pairs, 3,072 in all. In 2,961 of them (96%), the two readings agreed on whether the study belongs in the list. No person checked each reading. Of 9 studies with known answers, set aside before the full reading, 9 were put in or left out correctly. When a trial's papers do not give its ClinicalTrials.gov number, each paper can count as a trial of its own. The list does not say which group did better: open the trial's record or paper for its results. ClinicalTrials.gov searched on 2026-11-14, from data it processed on 2026-11-13. We summarized each record and kept only the fields in trials.csv. WRITTEN RULES rules.csv quotes each limit from its document in at most 40 words, with its section and dates. The document governs. Each quote was found by code in a copy of its document saved the day it was read. The copies are kept, not shared: ACGME and the publishers of the guidance hold copyright in theirs. FILES practice-list.csv: Each scheduling topic with its group, the phrases papers use for it, the other words a paper needs, its subject headings, guides and tools. id: the topic's identifier name: its name group: its group phrases: the phrases a paper may use, separated by |; a phrase in double quotes is matched exactly scope: the other words a paper needs: staff (health care staff or a hospital), shift (shift, roster or rota), or none when empty mesh: PubMed subject headings of its own, with their MeSH identifiers caveat: a note on the topic guide_slugs: Capillary guides on it, under /resources/ tool_paths: Capillary tools for it practice-list.json: The list's date, the matching rules, the words for health care staff and the words for shifts. practices.csv: Each topic's papers from 2021 to 2025 and from 2016 to 2020, the change, its papers in all years, the trials and reviews that name it, its sample result and its research grade. practice: the topic's identifier name: its name group: its group papers_2021_2025: papers that name it, 2021 to 2025 papers_2016_2020: papers that name it, 2016 to 2020 change: papers 2021 to 2025 over papers 2016 to 2020, given when the earlier window has 30 or more papers_all_years: papers that name it, all years heading_papers: papers under its own subject headings, all years randomized_trials: randomized controlled trials among the papers that name it systematic_reviews: systematic reviews among them meta_analyses: meta-analyses among them sample_read: sampled papers read sample_on_topic: sampled papers read as on topic research_grade: A for 300 or more papers from 2021 to 2025, B 100 to 299, C 30 to 99, D 10 to 29, E fewer than 10 practice-papers.csv: One row for each paper and topic: the PubMed identifier, the phrase or heading it matched and where, its year and its publication types. practice: the topic's identifier pmid: the paper's PubMed identifier basis: phrase or heading matched: the phrase or heading that matched place: title, abstract or heading major: for a heading, whether it is a major topic of the paper year: the year of the journal issue types: PubMed's publication types, separated by | checks.csv: Each topic's sample under each version of the list: the seed, the papers read, the papers on topic, and whether its count shows. practice: the topic's identifier list_version: the version of the list the sample was drawn under seed: the seed of the draw read: papers read on_topic: papers read as on topic stayed: yes when the count shows, no when it fell short, pending while being read details.json: The data release, the versions of the count script and of the list, and the day each source was read. README.txt: What each file holds and its columns, the sources and their terms, and how to cite the data. rules.csv: Each written limit on clinicians' hours and staffing: who it covers, its words as its document prints them, its section, document and dates, and the SHA-256 of the copy its quote was found in. rule_id: the limit's identifier: the rule-maker and the section, also its anchor on /data/rules body: the rule-maker jurisdiction: where it applies binds: rule (an accreditor's requirement), law, or guidance covers: who it covers practice: the topic it limits, by its identifier in practice-list.csv limit_text: the limit in the document's own words, at most 40 words limit_value: the number in the limit, when it has one unit: the number's unit averaged_over: the period the limit is averaged over, when it is section: the section, article or paragraph, as the document numbers it document_title: the document's title document_url: the address the quote was read at effective: the day the limit took effect, when the document states it read_on: the day the document was read capillary_check: the rule in Capillary's duty-hour check that applies it, when one does document_sha256: the SHA-256 of the copy the quote was found in quote_found_on: the day code found the quote in that copy trials.csv: One row for each trial and the topic it tests: its registry number, papers, design, what it compared and measured, the staff it studied, its size, countries and posted results. practice: the topic the trial tests, by its identifier in practice-list.csv; a trial that tests several topics has a row for each nct_id: its ClinicalTrials.gov number, empty when it has none; numbers are joined by | when it has more than one pmids: the PubMed identifiers of its papers, joined by | doi: the DOI of its main paper, as the paper's PubMed record states it first_author: the first author of its main paper: the earliest that PubMed types as a trial, else its earliest year: the year of its main paper, or of its start in the registry when it has no paper title: the title of its main paper, or its registry title journal: the journal of its main paper design: randomized, cluster-randomized (units or programs randomized), crossover (the same people or units under each schedule in turn), non-randomized, before-after, single-group or other compared: what it compared, in the readers' words, at most 25 words outcomes: what it measured, joined by |: patient-safety, staff-sleep, staff-health, staff-injuries, training, staffing-cost staff_group: the staff it studied, joined by |, by the identifiers in practice-list.json, or other participants: the people it enrolled, the staff when it counts staff and patients apart; empty when not stated or the readings differ units_or_programs: the units, wards, teams, hospitals or programs it compared; empty when not stated or the readings differ countries: the countries in its registry record, joined by | results_posted: the day it first posted results on ClinicalTrials.gov, empty when it has not fields_version: the version of the questions the readers answered trials.json: The same rows, with the day ClinicalTrials.gov was searched, how many studies were read and the readers' scores. references.csv: The key papers on each topic, the ones our guides cite and well-known studies, with their PubMed identifier or DOI and where each came from. practice: the topic's identifier pmid: the paper's PubMed identifier, empty when it has none doi: its DOI, empty when it has none label: the paper as the guide or the list of well-known studies names it from: guide: and the address under /resources/ of a Capillary guide that cites it, or spec for a well-known study of the topic LICENSE These files are shared under the Creative Commons Attribution 4.0 International license (CC BY 4.0): https://creativecommons.org/licenses/by/4.0/ You may copy, change and share them for any purpose, including commercial use. Cite them as below, link to the license and say if you changed them. The quotes in rules.csv, each limit in its document's own words, fall under their publishers' terms, not this license. CITE ConductScience. (2026). Research on clinician scheduling (Version 0.6.0) [Data set]. Capillary Health. https://capillaryhealth.com/data @misc{conductscience_clinician_scheduling_2026, author = {{ConductScience}}, title = {Research on clinician scheduling}, year = {2026}, version = {0.6.0}, publisher = {Capillary Health}, howpublished = {\url{https://capillaryhealth.com/data}}, note = {Data set. PubMed searched on 2026-11-12} } RELEASES 0.6.0 (2026-11-14) Trials: trials.csv and trials.json list each trial that compared one way of scheduling or staffing clinicians with another, by the topic it tests, with its ClinicalTrials.gov number and papers, its design, what it compared and measured, the staff it studied, its size, its countries and whether it posted results. An AI model read each study three times against a fixed list of questions; README.txt says how the list was made and how well the readings held up. references.csv: the key papers on each topic, the ones our guides cite and well-known studies, by PubMed identifier or DOI. README.txt says searched, not read, for the day a source was searched. 0.5.2 (2026-11-14) The files are shared under the CC BY 4.0 license: anyone may copy, change and share them for any purpose, with credit. README.txt and details.json name the license. The license does not cover the quotes in rules.csv, which fall under their publishers' terms. The pages and README.txt call the list's 66 items scheduling topics, not practices. The files keep the word practice in their names and columns. No counts changed. 0.5.1 (2026-11-13) Washington's overtime law, RCW 49.28.140, now links to an official copy on the state legislature's file server, because the law's web page asks programs not to read it. The quote and the limit are unchanged. 0.5.0 (2026-11-13) Written rules: rules.csv quotes each written limit on clinicians' hours and staffing from its document, with its section, its dates and the day code found the quote in a saved copy: ACGME's duty-hour requirements, the EU Working Time Directive, California's nurse-to-patient ratios, the state laws that restrict mandatory overtime for nurses, and guidance from professional bodies. 0.4.0 (2026-11-12) Counts by phrase: for each practice, the papers that name it in their title or abstract, by year, by publication type and by the staff they name, from PubMed read on 2026-11-12. A count shows only after at least 32 of 40 sampled papers, or four in five of a smaller sample, were read as on topic: 57 of the 66 practices. The list of practices, version 1.3.0: a paper counts when its title uses one of a practice's phrases or its abstract uses them twice. Versions 1.2.0 and 1.3.0 narrowed the phrases of 28 practices whose samples fell short; checks.csv keeps each version's sample. New files: practices.csv, practice-papers.csv, checks.csv, details.json and README.txt. 0.3.0 (2026-11-12) For each year, the papers under the three scheduling headings whose indexing finished on or after April 1, 2022, when NLM moved its indexing to software, read from the same summaries. 0.2.0 (2026-11-12) PubMed read on 2026-11-12: the papers NLM indexed under Personnel Staffing and Scheduling, Work Schedule Tolerance and Shift Work Schedule, each paper once, by year and by publication type, and the papers under the headings of the three practices that have one. 0.1.0 (2026-11-12) The list of 66 scheduling practices in 11 groups, list version 1.0.0, as practice-list.csv and practice-list.json.