Inspected the local DiffSim repository and fetched/pruned origin.
Searched all 15 local and remote branch refs with cedrik in their name.
Located the 40-class name mapping to nine DiffSim compartments on origin/cedrik/fk-fix, introduced by commit c59d66eb4787326bac8a2ed13ef97a642d415d50.
Verified from the local Histo fine28 taxonomy and output-store tests that Histo v2 writes 40 region_classes: 28 base classes plus 12 Gleason-derived classes, paired with sparse region_ids 0-27 and 128-139.
Compared the branch implementation with DiffSim tag v0.1.100 and documented why the released positional mapping cannot handle sparse graded labels. The intended port reads region_ids, accepts a class-name mapping, builds a separate label LUT, and retains a per-class mapping for names.
Reported the exact mapping, refs, metadata order, and integration guidance to the parent agent. No DiffSim or mono files were edited.
Lessons Learned: Pitfalls
Searching only for fine28 in DiffSim misses the implementation because the branch code names every class explicitly and never uses the taxonomy name.
The fine28 name is misleading at the integration boundary: the emitted region_classes list has 40 entries after adding derived Gleason classes.
The old 13-entry mapping on cedrik/new-format resembles the desired compartment collapse but is for the former taxonomy and is not compatible with Histo v2.
Lessons Learned: Improvements
Add a durable DiffSim/Histo integration note that records the fine28 output contract: 40 class names, sparse region_ids, and the nine-compartment mapping.
Document that DiffSim 0.1.100’s domain_from_histo assumes class-list position equals stored label and therefore needs the origin/cedrik/fk-fix behavior ported to NumPy rather than cherry-picked literally.
