data quality teams

Abstract: Restricting chain-of-thought (CoT) faithfulness evaluation to correctly answered instances changes the population being measured. We study this selection step in a seven-variable structural causal model and derive the exact observed gap $Δ= \operatorname{Cov}(A, C)/P(C=1)$. The model distinguishes difficulty confounding from shortcut-induced collider bias and states the conditional-independence assumption needed to express the gap through latent response propensities. A protocol review of 18 papers distinguishes answer-correctness restrictions from agreement-based selection, benchmark construction, and training-data filtering. We evaluate four sub-ceiling deployments on arithmetic tasks and observe positive selection gaps in all four. The primary GLM-5.3 run yields a +7.4 percentage-point gap ($n = 1000$, $p < 0.0001$), and pooled Cochran-Mantel-Haenszel analysis gives OR = 1.70 ($p < 10^{-6}$). An assumption-free covariance decomposition assigns 42.8% of the GLM-5.3 gap to between-stratum variation in interventionable-operation count, leaving the majority unattributed. When counterfactual compliance is observable for every instance, evaluating all traces directly estimates population faithfulness without correctness filtering.

Abstract: Approximate functional dependency (AFD) strength measures are commonly used to prioritize candidate dependencies, including when selected dependencies support downstream tasks such as error detection. This practice implicitly assumes that dependencies ranked higher by strength are also better candidates for the task they are intended to support. We test this assumption directly for error detection. We evaluate detection value through precision and recall without imposing a single weighting on different error types, and introduce an explicit false-positive/false-negative cost only when a single candidate-selection objective is required. Across three real-world relations from an established AFD benchmark, $1{,}155$ corruption-and-detection runs, and $513{,}480$ candidate-level evaluations, we find a clear asymmetry. Strength measures tend to track precision positively, but their associations with recall are substantially less stable, with both magnitude and direction varying across measures and experimental settings. Strength-based prioritization is also unreliable at the ranking level: top-$k$ sets chosen by strength often have little overlap with those favored by actual detection performance, and strength-based selections can incur substantial regret. An important part of the recall mismatch is associated with structural detection opportunity under the candidate equivalence-class partition, which tracks recall more consistently overall than the strength measures we evaluate. Corruption mechanism and several static structural properties account for selected effects but do not provide a common explanation across settings, leaving part of the mismatch unresolved. Overall, AFD strength and detection value are distinct and should be evaluated separately.

Abstract: Interpreting properties of LLM-generated data is important for understanding its utility and limitations across learning tasks. In this work, we characterize synthetic data through sample-level learnability, studying variation among LLM families and scales, alongside human-written data as a reference. We first generate synthetic datasets spanning single- and multi-label classification, labeling, and tree prediction tasks. We then derive empirical data distributions from encoder training dynamics for both machine and organic data, and estimate the robustness of these distributions across encoders. Finally, we evaluate how data selection strategies based on these learnability signals affect both data sources differently.

Abstract: Merchant matching resolves a noisy payment descriptor to a retrieved merchant entity or returns no match. A key challenge in curating training labels is distinguishing teacher abstention from evidence that no acceptable entity exists: false no-match labels contaminate pseudo-labeled data, while conservative labeling reduces coverage. We investigate whether agreement between two local large language model judges improves pseudo-label reliability. A label is retained only when the judges agree, with separate ordered thresholds for selections and abstentions that guarantee disjoint positive and negative label sets. Retrospective replay on 2,000 expert-annotated queries shows that higher selection thresholds can improve positive-label purity, whereas higher abstention thresholds increase false no-match labels. At thresholds (0.86, 0.80), Muse Glimmer 30B and Gemma 4 31B jointly label 1,633 queries (81.7% coverage) at 96.88% purity; positive and negative purities are 99.47% and 93.38%. This exceeds either constituent model at the same thresholds by more than two percentage points, with lower coverage. A split-half check finds only 0.14 percentage points of threshold-selection optimism. A symmetric threshold of 0.86 adds 40 erroneous no-match labels, while 46 false abstentions persist even with no confidence threshold. Across five matched within-model comparisons, higher reasoning effort yields no clear F0.5 gain and increases median latency by 1.8-5.0 times. These results motivate separate thresholding and auditing for positive and negative pseudo-labels. The study establishes label purity, not student utility; fresh-data curation and student fine-tuning remain necessary to demonstrate downstream value.

Abstract: Whitening a foundation-model embedding and using its squared norm as a training-free likelihood surrogate is motivated by the observation that whitened coordinates often appear approximately standard normal. We show that this observation follows from the projection central limit theorem and therefore does not imply a Gaussian joint distribution. Across multiple encoders and three training objectives, we find systematic over-dispersion of the whitened radius relative to the Gaussian reference, including against distributional clones with identical mean and covariance. We further show that the commonly reported agreement between empirical and theoretical norm statistics is an algebraic consequence of in-sample whitening and does not constitute evidence for Gaussianity. We identify the mechanism behind this behavior: whitening reverses the encoder's spectral hierarchy, shifting the contribution to the squared norm toward near-degenerate directions that encode predominantly noise. In these directions, the dominant variability is governed by a single input-dependent scale. We estimate this scale from two moments and use it to predict, without additional free parameters, the cross-dependence between disjoint spectral halves. These results indicate that the squared whitened norm is better interpreted as a Mahalanobis measure of semantic atypicality than as a log-likelihood. This interpretation explains both its practical effectiveness and its calibration failures: the statistic can rank and detect atypical samples consistently with nonparametric density estimates and across encoders trained with different objectives, while Gaussian tail thresholds can be inaccurate by orders of magnitude. etc.

Abstract: Counting equivalent pairs is a common way to report transformation-audit coverage, but it can substantially overstate the constraints imposed by an audit: pairs generated from the same semantic object are correlated, and complete orbit graphs contain algebraically redundant edges. We therefore distinguish four complementary quantities---edge count $m$, effective contrast rank $s$, population support rank $r$, and graph spectral gap $η$---and characterize their roles in audit coverage and deployment reliability. Under a rank-$r$ Gaussian contrast model, a population-invariant calibrated reader exists exactly when the anchor has a component in $\ker T$. When an audit has rank $s < r$, its unobserved risk is $R^\star/U$, with $U \sim \operatorname{Beta}((r-s)/2,s/2)$; when $s \ge r$, exact calibrated interpolation is infeasible. The same distinction appears in orbit topology: a spanning tree imposes the same exact-null constraints as a complete graph, while a sharp graph Poincare inequality propagates edge-level drift to an entire orbit at a cost proportional to $1/η$. Cyclic audits can additionally yield zero pair-level leave-one-out error without holding out any semantic object. To address these failures, we derive exact block-Woodbury leave-one-orbit-out updates and introduce a source-disjoint deployment gate over finitely many candidate readers. The gate retains the original reader unless uncertainty bounds certify lower drift within a prescribed clean-utility budget. etc..