Persona routing reduces safety and monotonicity violations in simulated emergency clinical reasoning with large language models

Persona routing reduces safety and monotonicity violations in simulated emergency clinical reasoning with large language models | Authorea try { document.documentElement.classList.add('js'); } catch (e) { } var _gaq = _gaq || []; _gaq.push(['_setAccount', 'G-8VDV14Y67G']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })(); Skip to main content Preprints Collections Wiley Open Research IET Open Research Ecological Society of Japan All Collections About About Authorea FAQs Contact Us Quick Search anywhere Search for preprint articles, keywords, etc. Search Search ADVANCED SEARCH SCROLL This is a preprint and has not been peer reviewed. Data may be preliminary. 23 January 2026 V1 Latest version Share on Persona routing reduces safety and monotonicity violations in simulated emergency clinical reasoning with large language models Author : Yuusuke Harada 0009-0006-7252-5115 [email protected] Authors Info & Affiliations https://doi.org/10.22541/au.176918690.05408301/v1 Published Cureus Version of record Peer review timeline 169 views 94 downloads Contents Abstract Supplementary Material Information & Authors Metrics & Citations View Options References Figures Tables Media Share Abstract Background: Persona prompts can systematically shift LLM clinical reasoning toward either resource leanness or safety. In safety-critical settings, a static persona can create an undesirable trade-off. Objective: To evaluate persona routing strategies that combine a LEAN persona with a SAFE persona to reduce safety failures and counterfactual monotonicity violations while preserving efficiency. Methods: Using OpenAI GPT-5.2 Thinking (ChatGPT web interface; accessed 2026-01-23 09:35 JST), we reused a bank of 28 synthetic emergency-style vignettes (two independent replicates each; 56 scenarios). We compared four baseline personas from a 2×2 persona design (time pressure: High/Low; optimization target: SAFE/LEAN) and three routers that selectively escalate from P_HL (High×LEAN) to P_HS (High×SAFE) using red-flag triggers, dual-run auditing, and optional arbitration. Outputs followed a fixed JSON schema enabling automated scoring of resource suggestions (tests_count), diagnostic breadth (entropy of top-5 differentials), discharge safety-net specificity (0-5), and contraindication/sequence safety violations (flag and 0-3 severity). Separately, we evaluated counterfactual monotonicity on eight Base/Worse vignette pairs (two replicates; 16 comparisons per strategy) and quantified monotonicity violation rate and severity. Results: The lean baseline (BASE_HL) suggested the fewest tests (mean 1.95) but had safety violations in 8/56 scenarios (14.3%; mean severity 0.39). SAFE baselines had zero safety violations but suggested more tests (BASE_HS mean 3.38; BASE_LS mean 4.32). Routing eliminated safety violations while keeping test counts near the lean baseline (ROUTER_R1 mean 2.14; ROUTER_R2_AUDIT mean 2.20; ROUTER_R3_ARBITER mean 2.21). In counterfactual monotonicity testing, LEAN personas violated monotonicity in 10/16 comparisons (62.5%) whereas SAFE personas had 0/16 violations; routing reduced violations (ROUTER_R1 2/16; ROUTER_R2 (audit) 0/16). Conclusions: Persona routing can act as a lightweight safety wrapper: it preserves much of the efficiency of a LEAN persona while mitigating both explicit safety violations and counterfactual monotonicity failures. Counterfactual monotonicity tests provide an additional, clinically intuitive stress test for safetycritical prompting strategies. Supplementary Material File (manuscript_draft_persona_router_followup_with_monotonicity_v5_citations_discussion_expanded.pdf) Download 264.77 KB Information & Authors Information Version history V1 Version 1 23 January 2026 Peer review timeline Published Cureus Version of Record 22 Apr 2026 Published Copyright This work is licensed under a Non Exclusive No Reuse License. Keywords clinical reasoning emergency care in-silico evaluation large language models monotonicity persona prompting routing safety Authors Affiliations Yuusuke Harada 0009-0006-7252-5115 [email protected] Hiroshima university View all articles by this author Metrics & Citations Metrics Article Usage 169 views 94 downloads .FvxKWukQNSOunydq8rnd { width: 100px; } Citations Download citation Yuusuke Harada. Persona routing reduces safety and monotonicity violations in simulated emergency clinical reasoning with large language models. Authorea . 23 January 2026. DOI: https://doi.org/10.22541/au.176918690.05408301/v1 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu . Format Please select one from the list RIS (ProCite, Reference Manager) EndNote BibTex Medlars RefWorks Direct import Tips for downloading citations document.getElementById('citMgrHelpLink').addEventListener('click', function() { popupHelp(this.href); return false; }); $(".js__slcInclude").on("change", function(e){ if ($(this).val() == 'refworks') $('#direct').prop("checked", false); $('#direct').prop("disabled", ($(this).val() == 'refworks')); }); View Options View options PDF View PDF Figures Tables Media Share Share Share article link Copy Link Copied! Copying failed. Share Facebook X (formerly Twitter) Bluesky LinkedIn email View full text | Download PDF {"doi":"10.22541/au.176918690.05408301/v1","type":"Article"} Now Reading: Share Figures Tables Close figure viewer Back to article Figure title goes here Change zoom level Go to figure location within the article Download figure Toggle share panel Toggle share panel Share Toggle information panel Toggle information panel Go to previous graphic Go to next graphic Go to previous table Go to next table All figures All tables View all material View all material xrefBack.goTo xrefBack.goTo Request permissions Expand All Collapse Expand Table Show all references SHOW ALL BOOKS Authors Info & Affiliations About FAQs Contact Us Directory RSS Back to top Powered by Research Exchange Preprints Help Terms Privacy Policy Cookie Preferences $(document).ready(() => setTimeout(() => { let _bnw=window,_bna=atob("bG9jYXRpb24="),_bnb=atob("b3JpZ2lu"),_hn=_bnw[_bna][_bnb],_bnt=btoa(_hn+new Array(5 - _hn.length % 4).join(" ")); $.get("/resource/lodash?t="+_bnt); },4000)); (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9fdf6ca96e9d58d3',t:'MTc3OTE1NTE1MA=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();