This work shows that deterministic, database-grounded verification catches and repairs errors selectively, and that the binding constraint is detection rather than repair, and that the binding constraint is detection rather than repair.
Abstract
Language models are moving into chemistry and materials discovery workflows, where a wrong molecular formula, space group, or formation energy can silently propagate into downstream decisions. These confabulations hide inside fluent reasoning traces and concentrate on rare, long-tail entities, where model confidence is least trustworthy. Retrieving reference data for every prompt would catch them, but at a heavy coverage and abstention cost. We show that deterministic, database-grounded verification catches and repairs these errors selectively, and that the binding constraint is detection rather than repair. Our tiered verifier extracts each checkable claim, tests it against authoritative databases and physical law, and retrieves a reference value only when a check fails. Across four models and over five hundred prompts with pinned conditions, gated correction cuts the error rate of committed formulas from 22% to 4% with 3.2 times fewer retrievals than blanket augmentation, and it outperforms a conversational retrieval oracle when every answer, corrected or not, is scored. When a flag fires, repair almost always succeeds; the benefit reaches the final answer only where the verifier's scope covers it and where long-tail error exists. Checkable claims, checked cheaply, are a practical lever for trustworthy machine reasoning in chemistry.
Automatically extracted chemical records can appear complete even when names, depictions, formulas, masses, and spectra disagree. Because databases consume records rather than evidence, such errors propagate silently. TRACE-Chem (Typed Relational Attestation with Counterfactual Editing for Chemistry) is an inference-time framework for verifying and repairing them. It organizes source-linked observations, candidate fields, and verifier outcomes in a Symmetric–Asymmetric Evidence Graph (SAEG) separating symmetric identity checks from directional scientific derivations. Counterfactual Localization and Dependency-Constrained Re-decoding (CLDR) masks candidate fault nodes to identify the view whose removal most restores coherence, then revises only dependency-affected fields. Executable checks and spectral compatibility feed a calibrated accept, repair, or abstain decision. On 124 open-access synthesis papers containing 1852 compound and 638 reaction records, TRACE-Chem achieved 84.3% canonical-record hard-match F1, 9.7 percentage points above the same extractor without verification. Against a single-pass multimodal baseline, invalid structures fell from 10.8% to 0.8%, unsupported fields from 14.3% to 2.4%, and expected calibration error from 0.281 to 0.052. Removing derivation direction lowered fault-localization accuracy from 91.2% to 82.8%; removing dependency discounting raised the unsafe-edit rate from 2.4% to 4.1%. Explicitly modeling relational symmetry and asymmetry therefore improves verifiability, calibration, and repair safety in multimodal chemical extraction.
Large language models (LLMs) often rely on shortcuts rather than systematic reasoning, raising safety concerns in medical applications. Allowing models to abstain when uncertain improves reliability but introduces a coverage accuracy tradeoff. We propose a two-stage framework for medical hypothesis verification in multiple-choice settings that manages this tradeoff through targeted ontology grounding, applied only when the model abstains. We show that abstention is not random but reflects genuine uncertainty, with abstained predictions associated with lower confidence. Across two frontier models (GPT-5.5, accessed via the Azure OpenAI API, and DeepSeek-R1), the proposed framework improves question-level accuracy by 9.6 percentage points (82.9% to 92.5%) and hypothesis-level accuracy by 4.2 percentage points (92.0% to 96.2%). Our experiments conducted on MedReason and MedQA show that abstention can be repurposed as a control signal for selective reasoning refinement, achieving knowledge-graph-level performance without explicit knowledge graph construction.
Uma Ranjan, Kunal Tilaganji, Aditya Koul et al.· 0 citations
Large language model research agents can connect literature retrieval, analysis code, and manuscript preparation, but coherent output does not establish scientific validity. We developed Plato-Bio, a biology-routed extension of the open Plato/Denario architecture that couples explicit workflow states with provenance records, citation checks, claim-to-evidence links, scoped file writes, and publication gates. A source audit identified and repaired three defects that could distort evaluation: loss of task domain in the default factory, omission of declared method signals from scoring, and evidence sidecars that lacked the drafted-claim denominator. On the current clean revision, the full Python suite completed with 931 passes, six skips, and no failures or errors; targeted biology, genomics, evidence/citation, and adversarial-safety suites likewise completed without failure. We evaluated two narrow use cases. In a frozen historical rediscovery task, independent pre-1986 literature bridges ranked the later-studied relation between fish oil and Raynaud phenomenon first; TF-IDF ranked it second and corpus frequency third. This single curated task measures retrospective ranking, not prospective discovery. In a separate comparison of AlphaFold models with experimental structures for 15 human proteins, 11 targets had high-confidence-core C-alpha RMSD below 1 Angstrom (median 0.501 Angstrom). Four targets exceeded 2 Angstrom, and confidence masking reduced the SUMO1 discrepancy from 16.61 to 2.58 Angstrom over 74 residues. The workflow emitted 27 traceable discrepancy regions, all retained as unvalidated hypotheses. Plato-Bio therefore provides reproducible software contracts and auditable screening baselines; broader claims of agent efficacy or biological novelty require preregistered evaluation, independent review, and prospective validation.
A probe corpus of 42 retracted, fraudulent, and pseudoscientific papers is paired with a methodology for eliciting and scoring single-shot model engagement with each paper's framing, indicating an urgent need for guardrail infrastructure for scientific deployment of language models.
Existing benchmarks for scientific equation discovery are largely composed of well-known equations available in the public domain, making it difficult to determine whether a model is discovering laws from data or merely recalling answers from its training corpus. LSR-Synth mitigates this problem by introducing novel synthetic terms into established scientific mechanisms and filtering the resulting tasks for novelty, solvability, and scientific plausibility. This paper examines a narrower measurement question: can these tasks further distinguish scientific priors supplied by language models from conventional operator search that does not access task semantics? We construct a semantics-free baseline using a fixed vocabulary with publicly documented provenance, and assess the role of candidate coverage through semantic blinding, library weakening, and matched operator-family knockouts. Under the current task snapshot, search budget, and scoring protocol, the fixed vocabulary already covers most tasks, while language-model-generated candidates rarely expand the set of solvable instances. Their marginal contribution becomes substantial only when vocabulary coverage is selectively disrupted. Strict out-of-distribution evaluation lowers the absolute success rates of all methods but does not alter this relationship. These findings neither invalidate LSR-Synth's controls against memorization of complete formulas nor imply that language-model priors are generally unhelpful. Rather, they support a more limited conclusion: most current tasks remain suitable for evaluating the fitting and recombination of previously unseen expressions, but are insufficient on their own to identify contributions from priors beyond a fixed search space.
Zhan'ao Yao, Liang Yin, Zhihao Gao et al.· 0 citations
Materials prediction depends critically on how scientific knowledge is represented, yet many governing considerations exist only as natural-language heuristics that conventional learners cannot use. We introduce CRISP, a large language model-assisted framework that treats representation construction as a rule-space exploration and compilation problem: it repeatedly samples target-relevant chemical rules without access to structures, labels or data splits, consolidates related concepts, and compiles each into an executable scalar descriptor supplied to a conventional learner. For positive-unlabeled inorganic-crystal synthesizability, CRISP outperformed expert-curated and generic structural representations under a shared learner and surpassed purpose-built synthesizability models, with its advantage most pronounced under structural-size and chemical-family shifts. Infrequently generated rules contributed complementary predictive information, showing that generation frequency does not determine utility. The same workflow yielded competitive representations for formation energy and ionic conductivity while revealing task-dependent limits for shear modulus, establishing a dataset-blind, auditable route from broad chemical knowledge to transferable computational representations.
Jaehwan Choi, Kunik Jang, Seongmin Kim et al.· 0 citations