deployed-multilingual-medical-triage-system-most-accuracy-gain-comes CLM-0376In a deployed multilingual medical triage system, most of the accuracy gain comes from writing the clinicians' implicit decision tree into an explicit scoped rule set rather than from moving the final decision out of the LLM: replacing a flat danger-sign list with 231 structured rules in the same end-to-end prompt raised recall from 0.565 to 0.837 and F1 from 0.606 to 0.689, while further handing the decision to a deterministic rule engine over LLM-extracted symptoms left accuracy statistically unchanged (recall 0.810, F1 0.702).
Capability: Following an unfamiliar procedure
Observed on
Short, code-switched WhatsApp maternal/newborn queries in seven Indian languages, 427 held-out test queries, Gemini-2.5-Flash, labels from the organisation's own clinical protocol..
Sources
- Validated by: Stepwise ablation of four systems on 427 held-out queries with stratified bootstrap CIs; full decomposition matched the rules-in-prompt variant on recall and F1 (overlapping intervals) at 1.5x cost, plus deployment over 152,421 queries with 17.8% over-escalation and no reported rise in missed emergencies.. Measured on one organisation's 427-query test set (147 emergencies), single LLM, labels from in-house clinicians with 83% raw agreement. Run-to-run variance checked on validation and small relative to architecture differences. [truncated]
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- When medical LLMs are given a clinical practice guideline in context and asked to produce the decision pathway a case requires, final-outcome accuracy does not track adherence to the guideline-defined step sequence: models can reach the right outcome while omitting, reordering, skipping or padding steps, and medically fine-tuned models score no better on pathway consistency than the general models they were tuned from.Competence in a specialist domain · reviewed by AI
- In LLM fact-checking over supplied evidence, high veracity accuracy does not imply the verdict depends on that evidence: when the model's own cited sentences are iteratively removed, off-the-shelf models (GPT-4o-mini, Gemini-2.5-Flash-Lite, Qwen-2.5-32B, Llama-3.1-8B) keep their SUPPORT/REFUTE labels at a non-zero plateau instead of abstaining, and training with paired supervision that requires NOT_ENOUGH_INFO on the evidence-ablated input removes most of that persistence without losing accuracy.Checking claims against evidence · reviewed by AI
- When an LLM judge is given a prior evaluation score as context metadata on a revision, its rubric scores shift toward that uninformative prior — seven of eight tested models (API and open-weight, 0-5 scale, 20 fixed texts) show a negative total shift, and on a categorical compliance task the metadata blocks 48% of error corrections and flips about 10% of previously correct judgments to the anchored wrong label.Judging another's work as well as a careful reviewer would · reviewed by AI
- On complex clinical diagnosis from published case reports (rare disease and challenging cases), running several presenter-rebutter debater pairs on one backbone and synthesizing their revised hypotheses with an aggregator raises exact-match diagnostic accuracy and rated safety over the same model prompted once, while giving no measurable gain on multiple-choice clinical knowledge questions.Getting more from several models than from one · reviewed by AI
- For detecting findings-conclusion inconsistencies in oncology PET/CT radiology reports, a 15M-parameter BERT pre-trained on in-domain reports reaches higher balanced accuracy and lower false-positive rate than zero- or few-shot prompted 27B-70B open-weight LLMs, and LoRA fine-tuning the 70B model only matches the small model rather than beating it.Spotting an error in a given artifact · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: An ablation on comparable triage data where supplying the explicit rule set inside the prompt gives little or no gain over a flat danger-sign list, while moving the decision into a deterministic engine produces the bulk of the accuracy improvement. Drafted stance toward many-failures-come-from-not-following-domain-policy: supports -- The baseline's errors came from the model reasoning from its own training-shaped priors because the clinical decision policy was never given to it.