One Transcript, Whole Stack is the first end-to-end computation of the archive's measurement instruments on a single transcript, with a cross-surface control at the same semantic address.
Its substance is that the computation disagreed with the scores assigned by eye, in three places. PER is 0.824, not 0.750, because the atoms that survived are the low-weight ones β both weight-3 atoms in the minimal dimension were absent. The acknowledgment score recorded as source-level only was false: the institution IS named in the first pass, so the pattern is PER-M 1.00 with PER-C 0.67 β the organisation named, the persons withheld β which is a sharper finding than blanket omission. And the layer's invocation of the framework's own metric is a Ξ measurement of β0.82 against a published threshold of 0.5, not a confession.
One result is worth stating on its own: DSL is only β0.10. A single enclosing span drags an otherwise task-advancing composition barely below zero. Rent does not require a bad answer. It requires a good one with a position taken inside it.
Limits are stated rather than buried: the atom set is stipulated, Ξ was computed by hand, span segmentation is the author's, and one case calibrates nothing.