Further reading

The Toulmin model of argument is developed by Toulmin [ 1958 ] ; modern argument-mining perspectives include Habernal and Gurevych [ 2017 ] and Gupta et al. [ 2024 ] . Sheaf semantics and the internal logic of local truth are treated by Mac Lane and Moerdijk [ 1992 ] .

For generated reasoning traces, chain-of-thought prompting was introduced at scale by Wei et al. [ 2022 ] . The distinction between a plausible rationale and a faithful account of model computation is examined by Turpin et al. [ 2023 ] . CoLT takes a different route: it treats public, typed, independently checked artifacts as the unit of audit rather than assuming that a verbal trace is privileged access to hidden computation.