Derivation boundaries: A structural source of faithfulness failure in large language models
Journal of Computer and Forensic Sciences · 2026
DOI: 10.5937/jcfs5-68363 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Computer and Forensic Sciences, 2026. Verfasst von Vaishnavi Sreekumar. Die Arbeit nennt 16 Literaturangaben.
Abstract (Originalsprache)
Faithfulness to supplied information is not uniform in large language models: some rules, thresholds, or features shape outputs correctly while others do not, with nothing in a model's own explanations distinguishing the two. An industrial classification task motivates this observation, where supplying a classifier's exact decision thresholds improved accuracy while degrading minority-class recall, raising the question of where such failures occur. Because that setting involved a closed model and proprietary data, the same question is examined in two domains: airline baggage fee and tax calculation. Across domains, models reliably handle directly available values but fail when a known fact must be transformed into a new value through a rule-conditioned lookup, points named here as derivation boundaries. Two error patterns recur there: misassignment, in which a valid value is assigned to the wrong side of a rule boundary, and fabrication, in which a value unsupported by the rules is introduced and justified as though retrieved correctly. Layer-wise analysis using the logit lens suggests a mechanistic explanation. Directly available values stabilize early, whereas boundary-dependent values commit later and less stably, with correct alternatives often remaining competitive until the final layers. Faithfulness failures concentrate at derivation boundaries rather than spreading uniformly, a pattern layer-wise analysis links to delayed internal resolution.
Zitieren
APA
Sreekumar, V. (2026). Derivation boundaries: A structural source of faithfulness failure in large language models. Journal of Computer and Forensic Sciences. https://doi.org/10.5937/jcfs5-68363
Vancouver
Sreekumar V. Derivation boundaries: A structural source of faithfulness failure in large language models. Journal of Computer and Forensic Sciences. 2026. doi:10.5937/jcfs5-68363
BibTeX
@article{Sreekumar2026,
author = {Sreekumar, Vaishnavi},
title = {Derivation boundaries: A structural source of faithfulness failure in large language models},
journal = {Journal of Computer and Forensic Sciences},
year = {2026},
doi = {10.5937/jcfs5-68363}
}
EUVIMED