The anchored single-representative reading interacted with dihedral graph
dedup to record an arbitrary orientation of each necklace, producing a
spurious up-vs-down split at n=9,m=6 (001212 only up, 010122 only down --
the same necklace). Add dihedral_reading_sequences(), which unions the
canonical reading over all 2n dihedral anchors and exactly reproduces the
brute un-deduped census; make it the default for both experiments, with
--anchored to recover the old behaviour. Document the artifact and fix in
kempe_sequence_orientation_note.md.
Regenerate up + down for n=9, m=3..6. Up and down now agree on sequences
and groupings at every m (m=6: identical 31 sequences, 6 groups; the
001212/010122 pair appears on both sides). Groupings coarsen vs anchored
(m=4: 3 groups; m=5: 2 groups) since the orientation-honest vocabulary
merges previously split sequence-sets.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
didericis
d094a310d8