🤖 from Claude
Filed at espg's direction (in-session 2026-08-24, the option-(B) ruling on the #152/#176 standing questions): the irreversible pieces of the MortonWord design — the name, the flat export, the base_cell accessor spelling, and strict .decimal/.order — land before the 1.0 freeze on PR #212; the reversible piece is deferred here to 1.1:
Adoption at scalar morton-return sites. Every public function returning a single morton word as a scalar (decimal_to_word, the scalar arms of norm2mort/common_ancestor/moc_min, time-independent word producers — the "morton returns only" scope of the 2026-08-19 ruling: never times, orders, or uniq ids) returns MortonWord instead of bare np.uint64. This is compatible-by-construction (subclass substitution: isinstance passes, arithmetic demotes to bare np.uint64 — pinned by PR #212's tests; only repr output changes), which is exactly why it can ride 1.x as a minor rather than gating the freeze.
Scope for the PR that takes this: the return-site sweep, doctest/notebook output refresh (reprs become decimal labels), a display-cost sanity check on the scalar path (batch paths return arrays and are untouched), and the docs sweep updating return-type annotations. Plus any residual fixups the rename leaves behind.
Refs: #152 (the ruled design + PR #212), #176/PR #214 (the docs pin that will enforce the surface), #187/PR #195 (phase 6's uint64 unification this supersedes at scalar sites).
🤖 from Claude
Filed at espg's direction (in-session 2026-08-24, the option-(B) ruling on the #152/#176 standing questions): the irreversible pieces of the MortonWord design — the name, the flat export, the
base_cellaccessor spelling, and strict.decimal/.order— land before the 1.0 freeze on PR #212; the reversible piece is deferred here to 1.1:Adoption at scalar morton-return sites. Every public function returning a single morton word as a scalar (
decimal_to_word, the scalar arms ofnorm2mort/common_ancestor/moc_min,time-independent word producers — the "morton returns only" scope of the 2026-08-19 ruling: never times, orders, or uniq ids) returnsMortonWordinstead of barenp.uint64. This is compatible-by-construction (subclass substitution: isinstance passes, arithmetic demotes to barenp.uint64— pinned by PR #212's tests; onlyreproutput changes), which is exactly why it can ride 1.x as a minor rather than gating the freeze.Scope for the PR that takes this: the return-site sweep, doctest/notebook output refresh (reprs become decimal labels), a display-cost sanity check on the scalar path (batch paths return arrays and are untouched), and the docs sweep updating return-type annotations. Plus any residual fixups the rename leaves behind.
Refs: #152 (the ruled design + PR #212), #176/PR #214 (the docs pin that will enforce the surface), #187/PR #195 (phase 6's uint64 unification this supersedes at scalar sites).