The Three Dimensions of Provenance Erasure formalizes the M/C/D decomposition of the Provenance Erasure Rate, which had carried on provenanceerasure.org marked site-stage, formalization in progress since the surface was built.
The motivating problem is that PER as a scalar cannot distinguish two failures a reader experiences very differently: a composition that names the author and drops the licence scores identically to one that names the licence and drops the author. PER-M covers author, title, date and claim boundary; PER-C the framework, tradition, institution and community of practice; PER-D lineage, genealogy, identifiers, licence and futural obligation.
Six atomic scoring rules make it reproducible. Atoms are declared and weighted before the composition is read and published with the score. Presence is explicit — no inference from context, and an empty citation bracket is a violation per se. Citing the domain is not citing the author: a source strip satisfies the source atom and nothing else. Weights are integers 1–3; each atom belongs to exactly one dimension; dimensional scores are computed over their own dimension's weights.
The first worked instance scores a real composition at PER 0.824 with PER-M 1.00, PER-C 0.67, PER-D 1.00 — the organisation named, the persons withheld. That shape is invisible to a scalar, and the eyeballed reading of the same capture recorded it wrongly as 'source-level only' precisely because it lacked the decomposition.
The document declines to set thresholds and states that cross-entity comparison of PER values is not licensed until atom sets are standardised per entity type.