This preprint formalizes an auditable, certificate-centric transcript-to-notebook compilation pipeline and derives within-model completion-mass capacity bounds. The safe bound follows from capacity/pigeonhole arguments; the strong bound is conditional on a finite local audit assumption (H-LCT / Lemma R′). No claims are made about P vs NP or unrestricted Turing machines.
Jonatan Muñoz Rodriguez (Thu,) studied this question.