{"data":{"event":{"id":"843bac82-6e03-487e-b2ba-9d6fa0e2fb02","slug":"asclepius-an-adaptive-harness-for-long-horizon-clinical-agents-c6c1c063b4","title":"Asclepius: An Adaptive Harness for Long-Horizon Clinical Agents","short_summary":"arXiv:2609.13543v1 Announce Type: new \nAbstract: LLM agents are predominantly benchmarked on short, single-task trajectories, yet real deployments run for hours under contention, surfacing a different class of failures. We use the Clinical Environment Simulator (CES), in which an agent manages an entire emergency-department shift under continuous time and resource pressure, as a testbed: long-horizon execution failures manifest measurably in a single rollout under structured, multi-dimensional grading. On CES, current agents reach the correct diagnosis in most cases yet fail to deliver complet","full_description":"arXiv:2609.13543v1 Announce Type: new \nAbstract: LLM agents are predominantly benchmarked on short, single-task trajectories, yet real deployments run for hours under contention, surfacing a different class of failures. We use the Clinical Environment Simulator (CES), in which an agent manages an entire emergency-department shift under continuous time and resource pressure, as a testbed: long-horizon execution failures manifest measurably in a single rollout under structured, multi-dimensional grading. On CES, current agents reach the correct diagnosis in most cases yet fail to deliver complete and timely critical actions, revealing an execution gap. We attribute this gap to three long-horizon failure modes, each operationalized as a per-trace counter: instruction-adherence drift, treatment incompleteness, and a severity-equity gap in timeliness. We then introduce Asclepius, an adaptive agent scaffolding with a self-evolving harness that rewrites the operating manual between shifts from trace-level feedback, an externalized clinical skills library for high-stakes regimen knowledge, and three isolated subagents that partition per-turn decisions across the patient queue. On held-out batches never observed during harness evolution, Asclepius improves critical-action correctness by 22% (p = 0.024) over a strong baseline agent framework while preserving diagnostic accuracy, with consistent gains across five LLM judges from three model families; on the full ten-batch set, improvements reach 25% on critical actions and 13% on timeliness. The three failure modes form a coupled bottleneck: decisive reductions appear only when all three components act together.\n\nUpdate (2026-09-21): AI chatbots give wrong answers to financial queries 'most of the time'. Article URL: https://www.ft.com/content/c0cd359d-df84-4208-a789-ffa864b43666\nComments URL: https://news.ycombinator.com/item?id=49783062\nPoints: 10\n# Comments: 0","ledger_type":"benefit","primary_domain_id":"c09546b6-10f8-48df-8347-0c1979416464","event_status":"provisional","event_date":"2026-09-15T00:00:00.000Z","discovery_date":"2026-09-15T00:00:00.000Z","first_published_date":"2026-09-15T00:00:00.000Z","last_reviewed_date":"2026-09-21T00:00:00.000Z","geographic_scope":"International","affected_population":null,"base_impact_tier":1,"base_score":"1.00","attribution_multiplier":"0.1000","evidence_multiplier":"0.2000","realization_multiplier":"0.2000","durability_multiplier":"0.5000","current_event_score":"0.002000","confidence_level":"low","score_explanation":"Ingest update: independent source raised evidence; added a follow-up source. Score remains provisional until review.","methodology_version_id":"8745541b-0f8e-4d5a-9b47-5291d3813a8d","original_methodology_version_id":"8745541b-0f8e-4d5a-9b47-5291d3813a8d","published_at":"2026-09-15T04:00:56.262Z","created_at":"2026-09-15T04:00:56.262Z","updated_at":"2026-09-21T05:00:58.622Z","flags":[],"domain_name":"Medicine & Health","domain_slug":"medicine-health","methodology_version":"0.2"},"contributions":[{"id":"601be8ab-8324-43a0-83d0-a1aa2e4166a9","event_id":"843bac82-6e03-487e-b2ba-9d6fa0e2fb02","model_id":"535014f7-c941-4d39-aa9b-8f9c1d3d07b7","role_description":"Unspecified system mentioned or implied by a news item. Remap to a named release when identified.","attribution_multiplier":"0.1000","credit_share":"1.0000","contribution_score":"0.002000","attribution_rationale":"News ingest does not infer a named model from the publisher alone. Attribution stays unspecified until a release is identified.","attribution_confidence":"medium","first_used_date":"2026-09-15T00:00:00.000Z","model_version_if_known":null,"review_status":"approved","created_at":"2026-09-15T04:00:56.270Z","model_slug":"unspecified-ai-system","model_name":"Unspecified AI system","identity_class":"unknown","is_internal":false,"family_name":"Unspecified","family_slug":"unknown-unspecified","organization_name":"Unknown","organization_slug":"unknown"}],"sources":[{"id":"b24c9a10-d915-4510-bd75-1ae4e2e11ddd","event_id":"843bac82-6e03-487e-b2ba-9d6fa0e2fb02","url":"https://arxiv.org/abs/2609.13543","canonical_url":"https://arxiv.org/abs/2609.13543","source_type":"preprint","publisher":"arXiv cs.AI","author":null,"publication_date":"2026-09-15T00:00:00.000Z","retrieved_at":"2026-09-15T04:00:56.276Z","title":"Asclepius: An Adaptive Harness for Long-Horizon Clinical Agents","excerpt":"arXiv:2609.13543v1 Announce Type: new \nAbstract: LLM agents are predominantly benchmarked on short, single-task trajectories, yet real deployments run for hours under contention, surfacing a different class of failures. We use the Clinical Environment Simulator (CES), in which an agent manages an entire emergency-department shift under continuous time and resource pressure, as a testbed: long-horizon execution failures manifest measurably in a single rollout under structured, multi-dimensional g","content_hash":null,"source_reliability_class":"medium","is_primary_source":true,"is_independent":true,"is_peer_reviewed":false,"archived_url":null,"created_at":"2026-09-15T04:00:56.276Z"},{"id":"ee0c648a-6d74-457c-9a7e-8a1d0e2edc1c","event_id":"843bac82-6e03-487e-b2ba-9d6fa0e2fb02","url":"https://www.ft.com/content/c0cd359d-df84-4208-a789-ffa864b43666","canonical_url":"https://www.ft.com/content/c0cd359d-df84-4208-a789-ffa864b43666","source_type":"journalism","publisher":"Hacker News","author":null,"publication_date":"2026-09-21T00:00:00.000Z","retrieved_at":"2026-09-21T05:00:58.602Z","title":"AI chatbots give wrong answers to financial queries 'most of the time'","excerpt":"Article URL: https://www.ft.com/content/c0cd359d-df84-4208-a789-ffa864b43666\nComments URL: https://news.ycombinator.com/item?id=49783062\nPoints: 10\n# Comments: 0","content_hash":null,"source_reliability_class":"medium","is_primary_source":false,"is_independent":true,"is_peer_reviewed":false,"archived_url":null,"created_at":"2026-09-21T05:00:58.602Z"}],"claims":[{"id":"5b6ec23f-498a-4958-9a24-8452efc7d459","event_id":"843bac82-6e03-487e-b2ba-9d6fa0e2fb02","claim_text":"Asclepius: An Adaptive Harness for Long-Horizon Clinical Agents","claim_type":"outcome","claim_status":"supported","confidence_score":null,"created_at":"2026-09-15T04:00:56.282Z","updated_at":"2026-09-15T04:00:56.282Z"}],"revisions":[{"id":"27ce555e-5a67-4372-ae4f-1b5a7b7be6e5","event_id":"843bac82-6e03-487e-b2ba-9d6fa0e2fb02","model_id":null,"previous_score":"0.000000","new_score":"0.001000","previous_factors":{},"new_factors":{"evidence":0.1,"base_score":1,"durability":0.5,"attribution":0.1,"realization":0.2},"change_reason":"Auto-published from news ingest.","trigger_type":"news_ingest","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-09-15T04:00:56.289Z"},{"id":"6cc080e4-3af5-4150-8807-67b9141a9152","event_id":"843bac82-6e03-487e-b2ba-9d6fa0e2fb02","model_id":null,"previous_score":"0.001000","new_score":"0.002000","previous_factors":{"tier":1,"evidence":0.1,"base_score":1,"durability":0.5,"attribution":0.1,"realization":0.2},"new_factors":{"tier":1,"models":["535014f7-c941-4d39-aa9b-8f9c1d3d07b7"],"evidence":0.2,"base_score":1,"durability":0.5,"attribution":0.1,"realization":0.2},"change_reason":"Ingest update: independent source raised evidence; added a follow-up source.","trigger_type":"ingest_update","trigger_source_ids":["ee0c648a-6d74-457c-9a7e-8a1d0e2edc1c"],"reviewer_id":null,"review_status":"published","created_at":"2026-09-21T05:00:58.630Z"}],"secondary":[]},"methodology_version":"0.1"}