{"id":"3c61e5d0-009e-4c03-8d27-6b00f7ba189e","arxiv_id":"2604.10441","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"VeriSim adds six literature-derived noise types to patient simulations and shows medical LLMs lose 15-25% diagnostic accuracy under realistic conditions.","lead":"VeriSim is a simulation framework that adds realistic patient communication issues like memory gaps, low health literacy, and anxiety into medical AI tests while keeping the true medical facts intact. A smart generalist should read it to see why lab-perfect AI models may fail when real patients do not communicate clearly.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Whether the six noise dimensions and hybrid UMLS-LLM verification faithfully reproduce real patient behavior without simulation artifacts remains the primary uncertainty for the reported 15-25% accuracy drops.","rationale":"The reader's weakest_assumption directly identifies the same load-bearing point. With the full manuscript now available, the concern still centers on simulation fidelity rather than on the experimental execution or reporting. The clinician evaluation and open-source release provide partial independent support, but do not resolve the core assumption, so the UNVERDICTED stance with low confidence is appropriate.","tokens_in":1761,"tokens_out":415,"duration_ms":54048,"concrete_test":"Sample 100 simulated multi-turn conversations (stratified across the six noise dimensions), have two board-certified clinicians independently flag any instances where patient statements contradict the stated ground truth or introduce implausible clinical details; compute the rate of such violations and compare against the paper's verification pass rate. If clinician-flagged inconsistencies exceed 5%, the truth-preservation claim is weakened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central empirical claim (all seven models degrade under noise, with smaller models showing 40% greater degradation) depends on the simulation being both realistic and strictly truth-preserving. The framework derives six dimensions from medical communication literature and uses a hybrid UMLS-LLM check to enforce ground-truth adherence. However, the operationalization details—prompt construction for noise injection, exact verification rules for multi-turn consistency, and handling of cumulative effects across conversation turns—are the least secure part of the argument. If the LLM component of verification introduces or fails to detect subtle factual drift, or if the noise patterns create artificial failure modes not present in real clinics, the observed drops in diagnostic accuracy and increases in conversation length could be artifacts of the testbed rather than evidence of a Sim-to-Real gap. The reported clinician kappa > 0.80 provides some external support but is a single post-generation rating and does not directly test whether verification preserves ground truth under each noise dimension.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper introduces VeriSim, a configurable framework for truth-preserving patient simulation in medical AI evaluation. It derives six noise dimensions from peer-reviewed medical communication literature (e.g., recall limitations, health literacy barriers, stigma-driven non-disclosure), injects them controllably into patient responses, and enforces ground-truth adherence via a hybrid UMLS-LLM verification mechanism. Experiments on seven open-weight LLMs demonstrate significant degradation: diagnostic accuracy drops 15-25% and conversation length increases 34-55% under noise, with 7B models showing 40% greater degradation than 70B+ models; medical fine-tuning offers limited robustness. Board-certified clinician evaluation yields kappa > 0.80, and LLM-as-a-Judge is positioned as a scalable auxiliary metric. The work is released open-source to address the Sim-to-Real gap in medical AI benchmarks.","tokens_in":1951,"tokens_out":679,"duration_ms":36010,"significance":"If the simulation accurately captures real patient communication behaviors without introducing artifacts, the results would be significant for the field by establishing a rigorous, reproducible testbed that exposes robustness limitations in current medical LLMs beyond standardized benchmarks. The quantitative findings on model-size-dependent degradation, combined with clinician validation and open-source release, could drive more realistic evaluation practices and improvements in clinical AI systems.","major_comments":[{"comment":"The operationalization of the six noise dimensions and the hybrid UMLS-LLM verification (including prompt construction for noise injection, exact multi-turn consistency rules, and handling of cumulative effects) lacks sufficient detail to confirm that ground truth is strictly preserved and that no simulation artifacts drive the reported 15-25% accuracy drops (Methods section on framework design and verification).","section":"Methods (Noise Injection and Verification)"},{"comment":"The central empirical claims of 15-25% diagnostic accuracy drops and 34-55% conversation length increases across seven LLMs are presented without statistical tests, confidence intervals, error bars, or details on post-hoc analysis choices, undermining assessment of whether the degradation is robust or model-specific (Results section).","section":"Results"},{"comment":"Clinician validation reports kappa > 0.80 but is described as a single post-generation rating; it does not include per-noise-dimension tests for ground-truth preservation or checks against real clinical encounter data, leaving the truth-preserving claim insufficiently supported for the load-bearing simulation (Evaluation section).","section":"Evaluation"}],"minor_comments":[{"comment":"The abstract and text refer to 'LLM-as-a-Judge' as validated but do not report the specific agreement metrics, prompt templates, or comparison protocol with clinician ratings.","section":"Abstract and Evaluation"},{"comment":"Specific model identifiers, versions, and exact prompting templates for the seven open-weight LLMs should be listed in a table or appendix for full reproducibility.","section":"Experiments"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern on simulation fidelity is load-bearing and not fully resolved by the current description; addressing the major comments would strengthen the manuscript substantially. The work aligns well with the journal's scope on AI evaluation and healthcare applications."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their constructive comments, which have helped us identify areas for improvement in clarity and rigor. We address each major comment below and outline the revisions we will make to strengthen the manuscript.","responses":[{"response":"We agree that additional methodological transparency is warranted. In the revised manuscript, we will expand the Methods section with explicit prompt templates for each noise dimension, the precise multi-turn consistency rules enforced by the hybrid verifier, and a step-by-step description of how cumulative noise effects are applied while preserving ground-truth medical facts via UMLS entity linking and LLM consistency checks. We will also include pseudocode illustrating the verification pipeline to demonstrate that no simulation artifacts are introduced.","revision_made":"yes","referee_comment":"[Methods (Noise Injection and Verification)] The operationalization of the six noise dimensions and the hybrid UMLS-LLM verification (including prompt construction for noise injection, exact multi-turn consistency rules, and handling of cumulative effects) lacks sufficient detail to confirm that ground truth is strictly preserved and that no simulation artifacts drive the reported 15-25% accuracy drops (Methods section on framework design and verification)."},{"response":"We concur that statistical support is essential for interpreting the magnitude and reliability of the observed effects. We will revise the Results section to report paired statistical tests (e.g., Wilcoxon signed-rank tests with Bonferroni correction), 95% confidence intervals for all accuracy and length metrics, error bars on figures, and a clear description of the post-hoc analysis pipeline. These additions will allow readers to evaluate whether the 15-25% and 34-55% degradations are statistically robust and model-size dependent.","revision_made":"yes","referee_comment":"[Results] The central empirical claims of 15-25% diagnostic accuracy drops and 34-55% conversation length increases across seven LLMs are presented without statistical tests, confidence intervals, error bars, or details on post-hoc analysis choices, undermining assessment of whether the degradation is robust or model-specific (Results section)."},{"response":"We will strengthen the Evaluation section by reporting per-noise-dimension inter-annotator agreement (Fleiss' kappa) from the clinician reviews to confirm consistency across the six noise types. However, direct empirical checks against real clinical encounter data are not feasible within this study due to privacy regulations and the absence of matched real-world transcripts; our design instead relies on literature-derived noise dimensions validated through clinician review of simulated outputs.","revision_made":"partial","referee_comment":"[Evaluation] Clinician validation reports kappa > 0.80 but is described as a single post-generation rating; it does not include per-noise-dimension tests for ground-truth preservation or checks against real clinical encounter data, leaving the truth-preserving claim insufficiently supported for the load-bearing simulation (Evaluation section)."}],"tokens_in":1540,"tokens_out":647,"duration_ms":28871,"standing_objections":["Direct empirical validation of the simulated patient behaviors against real clinical encounter data, which would require access to protected health information and is outside the ethical and practical scope of the current work."]},"desk_editor":{"model":"grok-4.3","letter":"VeriSim introduces a configurable framework to test medical LLMs with patient noise that mimics real clinical issues like memory gaps and low health literacy. The key result is that seven open-weight models all show big drops in accuracy and longer conversations under this noise, with smaller models hit harder.","headline":"VeriSim gives a practical noise-injection framework for medical LLMs and reports clear accuracy drops, but the realism of the six noise dimensions and hybrid verification is the part that still needs checking.","tokens_in":2468,"tokens_out":140,"would_cite":false,"duration_ms":39734,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Medical LLMs lose 15-25% diagnostic accuracy when patient responses include realistic noise like memory gaps and anxiety.","keywords":["medical AI","patient simulation","LLM robustness","noise injection","diagnostic accuracy","clinical communication","sim-to-real gap","conversation evaluation"],"falsifier":"Running the same LLMs on transcripts from actual doctor-patient visits and measuring whether diagnostic accuracy drops by 15-25% and conversation length rises by 34-55%, matching the simulated results.","tokens_in":2666,"feed_emoji":"🩺","tokens_out":684,"duration_ms":49084,"temperature":0.7,"pith_summary":"The paper presents VeriSim as a way to test medical AI by adding controllable noise to simulated patient answers while keeping the underlying medical facts fixed. It draws six noise types from studies on how people actually talk to doctors, such as forgetting details or avoiding topics due to stigma. Tests on seven different LLMs show every model gets worse at diagnosis and needs longer conversations, with smaller models hurt more than big ones. Standard medical fine-tuning does not help much against this kind of noise. The work uses clinician reviews and a hybrid check to confirm the simulations stay realistic and truthful.","feed_headline":"Medical AI accuracy drops 15-25% with realistic patient noise","feed_subtitle":"New simulation shows LLMs need longer conversations and make more errors when patients forget details or avoid sensitive topics.","key_machinery":"The VeriSim framework, which uses six noise dimensions from medical communication literature combined with hybrid UMLS-LLM verification to add realistic patient noise while preserving ground truth.","core_discovery":"VeriSim is a truth-preserving patient simulation framework that injects controllable, clinically evidence-grounded noise into patient responses while maintaining strict adherence to medical ground truth through a hybrid UMLS-LLM verification mechanism. It operationalizes six noise dimensions derived from peer-reviewed medical communication literature, capturing authentic clinical phenomena such as patient recall limitations, health literacy barriers, and stigma-driven non-disclosure. Experiments across seven open-weight LLMs reveal that all models degrade significantly under realistic patient noise, with diagnostic accuracy dropping 15-25% and conversation length increasing 34-55%.","pith_inferences":["Current benchmark tests for medical AI may overestimate real-world performance because they lack this form of patient noise.","Training methods that explicitly include simulated communication barriers could close part of the observed performance gap.","The open-source release of the framework allows direct comparison of new models or fine-tuning approaches against the same noise conditions."],"forward_implications":["Diagnostic accuracy of all tested LLMs drops 15-25% when patient noise is added.","Conversation length grows 34-55% under the same conditions.","7B-parameter models degrade about 40% more than 70B+ models.","Medical fine-tuning on standard data gives only limited protection against communication noise.","Clinicians rate the simulations as high quality with inter-annotator kappa above 0.80."],"fun_headline_variants":["Medical AI loses 15-25% accuracy to realistic patient noise","Patient noise lengthens LLM conversations 34-55%","Smaller models degrade 40% more under patient noise","VeriSim tests medical AI with realistic patient noise"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The six noise dimensions drawn from medical literature plus the hybrid verification step capture real patient communication without creating new distortions.","fun_headline_variants_meta":{"raw":{"variants":["Medical AI loses 15-25% accuracy to realistic patient noise","Patient noise lengthens LLM conversations 34-55%","Smaller models degrade 40% more under patient noise","VeriSim tests medical AI with realistic patient noise"]},"model":"grok-4.3","cost_usd":0.011826,"raw_usage":{"total_tokens":5208,"prompt_tokens":740,"num_sources_used":0,"completion_tokens":65,"cost_in_usd_ticks":118262000,"prompt_tokens_details":{"text_tokens":740,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":4403,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":740,"tokens_out":65,"duration_ms":48082,"temperature":1.0,"reasoning_tokens":4403,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-10T16:22:23.703345+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Running the same LLMs on transcripts from actual doctor-patient visits and measuring whether diagnostic accuracy drops by 15-25% and conversation length rises by 34-55%, matching the simulated results.","supporting_citations":[],"review_version":1}