Papers for
data validation teams
Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.
Study treats AI agents like people to improve their safety and reliability
Do AI Agents need Preventive Care? An Epidemiologic Cohort Design for Agentic AI
Abstract: As AI agents increasingly perform tasks autonomously across sectors, including healthcare and biomedical research, understanding how their behaviors develop and influence outcomes over time becomes increasingly important. Existing evaluations largely focus on model capabilities and performance on predefined tasks. Here, we propose an epidemiologic cohort design for studying agentic AI, drawing on principles established in human population studies. The framework treats individual agents as members of a cohort, their underlying model as genetic makeup, accumulated work as their time scale, and naturally occurring behaviors as exposures that may influence subsequent outcomes. By following agents longitudinally, this approach enables investigation of behavioral determinants of adverse events, estimation of their potential preventable burden under causal assumptions, and characterization of behavioral changes over an agent's work life. We illustrate the design through two prospective observational cohorts examining curiosity and probing as a safety question, and data checking and analytical accuracy as a reliability question. Together, the framework and illustrative studies provide a generalizable blueprint for designing epidemiologic cohorts of AI agents.
Reasoning graph reliability varies with model and impacts answer confidence
Diagnostic Analysis for Large Language Model Reasoning
Abstract: Reasoning structures (e.g., chains and graphs) are increasingly used for probabilistic inference, factuality verification, and uncertainty estimation, but their structural reliability is rarely evaluated directly. We test whether this reliability holds through a proposed two-stage diagnostic evaluation in open-ended multi-hop QA. Stage 1 tests different capabilities required to construct reasoning graphs, showing that graph reliability varies with topology, prompt design, and model family. Stage 2 uses uncertainty estimation as a downstream test case. Generated-edge experiments show that graph-based computation is sensitive to upstream graph errors, with performance dropping relative to the gold-graph setting. Gold-graph experiments further show that graph structure provides little benefit when used only as prompt context, but improves the separation of correct from incorrect answers by confidence score and enables node-level diagnostics.
Routeaudit improves multi-verifier routing with adaptive interaction analysis
ROUTEAUDIT: Interaction-Aware Identification for Budgeted Multi-Verifier Routing
Abstract: Adaptive multi-verifier systems are commonly compared through endpoint quality-cost gaps, even when the verifier catalog, availability, accounting, information filtration, or scorer changes with the policy. We formulate verifier routing as a contract-conditioned identification problem. The contract records request support, verifier catalog, realized availability, resource accounting, online filtration, and post-trace scoring; a matched route contrast changes only the policy coordinate. ROUTEAUDIT adds three measurable objects to this contract. A contract lattice averages coordinate increments over every admissible bridge order and reports the resulting attribution together with its path sensitivity. A policy-independent response tape identifies paired sequential contrasts when adaptive policies reveal different observations. For incomplete matching, request-level bounds use whichever potential outcome remains observed and give a sharp finite-population interval. The protocol commits paid observations and ledger events before the oracle join and returns an attribution certificate for each comparison. On two held-out raw-tail caches, matched static SF+SA equals the cascade, assigning the apparent gains of 0.1797 and 0.1250 over full static to the verifier-set edge. On 1,319 held-out task requests, the learned and RLVR studies report quality 0.9522 and 0.9553 versus 0.9484 for matched static; the RLVR-static paired difference is +0.0068 with a request-paired interval $[0.0015,0.0122]$ and a training-seed-by-request hierarchical interval $[0.0006,0.0131]$. Controlled attribution recovery yields route mean absolute error 0.0011 and endpoint reconstruction error 0.0004. Factorial, bridge-order, and stochastic-provider studies evaluate the certificate interface; RLVR supplies a learned-policy stress test under the same identification contract.
Jev system reveals hidden uncertainty improves probability estimates
Jev thinks "I don't know'', but doesn't say it: Introducing Sys1Cal-v1 Dataset for Probability Calibration
Abstract: The appearance of Jev marked the era of System One Models, foundation models that return structured decisions with probability distributions rather than text. Aside from low cost and great speed, Jev's central promise is that these probabilities are calibrated: such claim is not backed by any public test and available external benchmarks evaluate confidence calibration, not whether every returned option probability has the right numerical meaning. To tackle this issue, we introduce Sys1Cal-v1, a dataset of True/False questions about a proposition $A$ for which the exact probability $P(A)$ is known by construction. Each item is queried through the three Jev primitives - Noul, Choice and Score - and evaluated by total variation distance from the ground-truth distribution, which can be used to estimate a soft accuracy of System One Models. We showcase the utility of Sys1Cal-v1 as a benchmark dataset by evaluating Jev and SemIf, an open-source Choice-style baseline. In this work, however, we focus even more deeply on Jev, by studying the calibration of its Score and Choice answers. In particular, we discover a peculiar behaviour that can be explained by assuming that Jev suppresses a third truth value, going beyond True and False. In other words, in \texttt{Choice} answers, $P(A)$ and $P(\neg A)$ are presented as if $P(A)+P(\neg A)=1$, while a term $P(U)\neq0$ is missing in the sum. Recovering $P(U)$ leads to an improvement of median soft accuracy in \texttt{Choice} answers from $0.771$ to $0.978$, suggesting that, even in binary decisions, Jev wants to answer with a third option:``I don't know''.
Typed decision models fail to align probabilities with probability rules
Beyond Calibration: Do a Typed-Decision Model's Probabilities Obey the Probability Axioms?
Abstract: Typed-decision models such as TypeSafe's Jev answer a declared yes/no or multiple-choice question about a state with a probability instead of text, and their evaluations report accuracy and calibration. Neither requires that the probabilities a model gives to logically related questions fit together item by item, which is what a system that acts on those probabilities needs. We test this property, coherence, with a battery of logically linked questions that needs no labels. For 160 items from ChaosNLI and PubMedQA, each with three mutually exclusive labels, we ask whether the label is X, whether it is not X, whether it is one of the other two labels, and which label applies. On 480 negation pairs, Jev's probabilities for "the label is X" and "the label is not X" miss summing to one by 0.064 on average (95% CI 0.055 to 0.072). Qwen3.8-27B, run from its official BF16 weights, misses by 0.293 with first-token probabilities and by 0.122 with verbalized probabilities. The gap to the first-token readout persists on pairs where both systems give similar probabilities, without double-negation labels, and after averaging Jev's repeated calls. Jev is not coherent either: its violations are about five times its repeat noise, and it over-endorses statements about single labels, so that its three single-label probabilities sum to 1.14 on average. The two systems also fail differently. Qwen3.8-27B's first-token readout under-endorses the complement of a label whether or not the question contains "not", rejecting both a statement and its negation in 196 of 480 pairs, and it does not become more coherent where it is more confident, whereas Jev's violations concentrate where its answer is uncertain. Because the checks need no labels, they expose biases that appear only when question forms are compared, and inconsistencies within items.