{"resourceId":"synagent-verify-falsify-licoo2-20260916","versions":[{"version":"external-8c37e036a82d81edc2ca5349e3a54a18102c8eeba4a1f21040c32f4fb17a6285","resource":{"id":"synagent-verify-falsify-licoo2-20260916","title":"SynAgent: 18 autonomous LiCoO2 runs with verify–falsify reasoning locate 650–690 °C window (best 61.6 at 670 °C)","organization":"University of Tokyo / Rigaku / Mitsui Knowledge Industry (Takahara et al.)","sector":"University self-driving laboratory / materials synthesis AI","geography":"Japan (University of Tokyo dLab); transfer limits for other materials/systems","publishedAt":"September 16, 2026 (arXiv v1)","publicationDate":"2026-09-16","eventDate":null,"sourceName":"arXiv","sourceLabel":"Academic preprint — SynAgent multimodal SDL agents","sourceUrl":"https://arxiv.org/html/2609.18598v1","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["developers-agents","knowledge-work","infrastructure","operating-model"],"finding":"SynAgent couples multimodal LLM agents to a physical automated lab with adaptive analysis-skill generation and a verify–falsify proposal scheme that maintains an explicit understanding document. In 18 LiCoO2 (001) sputtering experiments, skills for XRD intensity ratio and SEM morphology were created at iteration 1 and reused; ordering onset was bracketed between 630 and 650 °C with a high-quality window at 650–690 °C and best I(003)/I(006)=61.6 at 670 °C; 15 predictions were judged supported and 3 refuted, with falsify proposals driving major revisions.","sledRelevance":"Why now: Unarchived 16 Sep measured autonomous-lab campaign—extends prior agent-bench editions by showing falsification-driven process understanding on real instruments, not only code reproduction scores.","evidence":"18-iteration table; 650–690 °C window; peak 61.6 at 670 °C; 15/3 supported/refuted; skill generation/reuse claims.","architectureImplications":"SDL stacks need skill registries, MaiML-like measurement contracts, and logged proposal/reflection turns—not optimizer APIs alone.","governanceImplications":"Require precommitted predictions and human review of generated skills before treating campaign narratives as institutional knowledge.","securityPrivacyImplications":"Sandbox generated analysis code; control instrument command paths; retain immutable campaign logs.","caveats":"Single variable; one system; one visit per condition; LLM self-judgment; preprint; BO matched re-run absent.","streamIds":["research"],"roles":{"sales":"Sales framing: Materials and SDL programs want autonomous campaigns that leave human-readable process understanding, not only optimized samples from black-box Bayesian loops. SynAgent evidence supports a bounded pitch around multimodal agents that generate analysis skills and deliberately falsify their own claims—reporting an 18-experiment LiCoO2 thin-film campaign that located a narrow 650–690 °C growth window with peak I(003)/I(006)=61.6 at 670 °C and 15/18 predictions judged supported—not a claim that any materials system will self-discover under campus procurement. Discovery should map instruments, MaiML data paths, and human review of skills. Unsupported: multi-variable optimality guarantees, skill correctness without human review, or BO replacement ROI.","engineering":"Engineering framing: Fit is a multi-agent SDL decision layer (proposal/reflect/understand) with empty skill library, adaptive skill generation, and verify–falsify proposal modes over a physical interconnect lab. Architecture integration points are MaiML measurement files, XRD/SEM instruments, and persistent session logs—not a generic chat UI. Prerequisites: standardized experiment data format, controllable condition grid, and vision-capable model access. Constraints: single temperature variable; one materials system; single visit per condition; LLM self-judgments of support/refute; comparison to prior BO on same platform not re-run under matched budgets here. Security: protect instrument control and measurement stores; review generated analysis code before reuse. Proof of value is frozen campaign history tying claims to experiments plus skill self-tests.","delivery":"Delivery framing: Pilot SynAgent-style verify–falsify loops on one instrumented materials objective with human review of generated skills and understanding documents. Dependencies: SDL/core engineering, materials PI, research computing, and quality review. Ownership: PI owns scientific claims; automation team owns loop reliability. Skills adoption teaches staff to demand falsify proposals and skill registration evidence. Governance checkpoints: skill registration review, prediction commitment before run, and end-of-campaign understanding audit. Acceptance:  documented temperature map with cited experiments and written limits on single-variable transfer; risks include confirmation bias if falsify mode is dropped and unverified analysis skills."},"roleDetails":{"sales":{"customerProblem":"Self-driving labs optimize scalars but often leave opaque reasons and untested failure regimes.","stakeholders":["Materials PIs","SDL/core directors","Research computing","Research integrity"],"discoveryQuestions":["Which objective and grid are fixed?","Who reviews generated skills?","How are falsify proposals scheduled?","What human-readable understanding is required at close?"],"valueHypothesis":"Verify–falsify multimodal agents can recover abrupt process thresholds and human-readable maps (e.g., 650–690 °C window) when skills are generated from real measurements and predictions are precommitted.","potentialEngagement":"Single-campaign pilot on one materials objective with skill-review gates and exported understanding document.","unsupportedClaims":["Guaranteed discovery on any chemistry","Skill correctness without review","Matched BO superiority claim without re-run"]},"engineering":{"fit":"LLM multi-agent SDL loop with adaptive analysis skills and verify–falsify proposal modes.","architectureIntegration":"Connect dLab/MaiML instrument bus to agent session store; register analyze.py skills after self-test; log proposals/reflections.","prerequisites":["MaiML or equivalent standardized measurements","Instrument automation ready","Vision-capable model","Condition grid config"],"constraints":["Single controllable variable in demo","One visit per condition","LLM judges support/refute","Japan university platform"],"security":"Sandbox skill execution; control instrument write paths; retain immutable campaign logs.","proofOfValue":["Precommitted predictions vs outcomes table","Registered skills reused across iterations","Understanding document citing experiments"]},"delivery":{"work":"Configure campaign, run N autonomous iterations with skill generation, export understanding and judgment table for PI review.","dependencies":["SDL engineering","Materials characterization","Research computing","PI scientific ownership"],"ownership":"PI owns claims; automation lead owns loop operations.","skillsAdoption":"Train staff to require falsify proposals and to treat skill outputs as provisional until reviewed.","governanceCheckpoints":["Skill registration approval","Prediction commitment check","End-of-campaign audit"],"acceptanceCriteria":["Exported iteration table","Human-readable four-region (or equivalent) map","Written single-variable limits"],"risks":["Dropped falsify mode","Unverified skills","Overgeneralizing LiCoO2 results"]}},"retrievedAt":"2026-09-22T03:18:53Z","enrichedAt":"2026-09-22T03:18:53Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Shifts labor from manual DOE bookkeeping to reviewing skills/understanding documents; still needs materials experts for claim acceptance.","procurementImplications":"Buy SDL decision layers only with exportable understanding artifacts and falsify-mode requirements—not black-box optimizers alone.","operatingModelImplications":"Fund human skill-review capacity and campaign audits alongside robot/instrument capital.","sourceVerification":{"openedUrl":"https://arxiv.org/html/2609.18598v1","referenceExcerpt":"narrow optimal growth window at 650–690 ∘C","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}