Follow up items for the week of June 27, 2026

[espg]

🤖 from Claude

Sweep of closed PRs from the last 10 days (June 17–27, 2026): #53, #55, #57, #58, #59, #70, #72, #76, #84. Items already individually tracked as open issues (#78, #79, #80, #81, #83) or fully addressed in the prior sweep issue (#73) are excluded. Items (7)–(10) are carry-forwards from #73 that are not yet individually tracked.

---

Items

(1) // floor-division (Morton coarsening) operator — deferred — PR #72

PR #72 added moc_min / common_ancestor and documented // as the reduce-by-common-coarsening sibling of clip2order. The operator was explicitly called "deferred" in the PR body. No tracking issue exists. Scope: MortonIndex // n coarsening to a given order, matching Python // semantics.

(2) Morton __repr__ deferred to a separate PR — PR #76

PR #76's body explicitly notes: "no morton __repr__ is added here — that was deferred to a separate PR." No issue has been opened. A useful repr (e.g. MortonCell(order=12, base=3, ...) rather than a raw int) would significantly improve interactive usability and debugging.

(3) res2display() order cap silently wrong above order 20 — PR #76

res2display() in tools.py loops only to order 20 and prints km-only units. With the packed-u64 encoding now supporting order 29, the function returns None for any order above 20 — silently giving wrong output for the ~11 new usable orders. Should extend to order 29 with readable units (m at fine scales).

(4) Earthdata-gated notebooks not publicly Binder-runnable — PR #76

Three of the five updated example notebooks (velocity/grounding line workflows) require authenticated earthaccess downloads and cannot run on public Binder without credentials. PR #76 asked: "Leave as-is, or open a follow-up to cache a small anonymous subset in-tree?" — no decision was made. CLAUDE.md §4 states notebooks "should be runnable on binder" with "publicly (anonymously) reachable" data, so this is a compliance gap.

(5) split_base_cells dict entry ordering — decision pending — PR #84

PR #84 added split_base_cells with an open question on dict ordering: first-seen (insertion order from traversal) vs sorted by base cell id. This is an API/behavior decision that affects reproducibility and testability. No determinism test can be written until this is decided.

(6) split_base_cells Python wrapper vs Rust group_by kernel — decision pending — PR #84

A sorted-output variant of split_base_cells could be a Python wrapper (sort the existing dict result) or a new Rust group-by kernel. The current implementation is pure Python. PR #84 raised this as an open question for @espg. No decision recorded.

(7) CI versioning automation — PR #55 / carry-forward from #73

Still unaddressed: the most recent Cargo.toml bump commit message reads "REALLY needs to be automated!!!" (commit 1cf29c2, June 25). Manual Cargo.toml bumps remain the only release path and will keep causing friction on every release. Risk: version drift between the git tag that triggers build-wheels.yml and the Cargo.toml version that sets the wheel version.

(8) mort2norm FORCE_PYTHON remnant — PR #49 / carry-forward from #73

mort2norm's _python_mort2norm fallback, the os.environ env-var switch, and the os import remain in tools.py. CLAUDE.md §7 states the Rust extension is the sole runtime path — this remnant is inconsistent with that guarantee. Scope: one function + ~5 lines of cleanup.

(9) Clippy useless_conversion 27 warnings — decision pending — PR #57 / carry-forward from #73

27 pre-existing clippy warnings remain on main from the pyo3/numpy ? idiom (false-positive clippy::useless_conversion — the conversion is load-bearing). PR #57 left the decision open: keep as-is (noisy clippy output) or add #[allow(clippy::useless_conversion)] at the crate root. Tracked broadly under #48 but never individually filed.

(10) Benchmark scripts reference removed FORCE_PYTHON functions — PR #49 / carry-forward from #73

benchmarks/benchmark_rust_vs_python.py and benchmark_morton_bbox.py still set MORTIE_FORCE_PYTHON. For the four removed functions this env var is now a no-op; the benchmarks silently measure only the Rust path while claiming to compare Rust vs Python. Not run by CI so no failure, but misleading. Scope: small update or deletion.

---

Priority / severity table

Priority	Item	Severity	Source PR
1	res2display() order cap at 20 — silent wrong output for orders 21–29	High	PR #76
2	CI versioning automation — repeated manual pain, explicitly flagged by committer	Medium-High	PR #55
3	Earthdata-gated notebooks — 3 notebooks can't run on public Binder (CLAUDE.md §4 gap)	Medium	PR #76
4	mort2norm FORCE_PYTHON remnant — violates §7 Rust-only guarantee	Medium	PR #49
5	Morton __repr__ missing — deferred usability gap, no issue filed	Medium	PR #76
6	// Morton coarsening operator — deferred API gap, no issue filed	Medium	PR #72
7	split_base_cells ordering decision — blocks writing determinism tests	Low-Medium	PR #84
8	split_base_cells Python vs Rust implementation site — performance decision	Low	PR #84
9	Benchmark scripts reference removed FORCE_PYTHON — misleading dead code	Low	PR #49
10	Clippy useless_conversion 27 warnings — decision pending (broadly tracked in #48)	Low	PR #57

[espg]

🤖 from Claude

This is a backlog digest with ten items but no per-item tracking, so the useful discuss move is to propose how to triage and label it so a later routine run can actually pick the items up (an unlabeled item is out of scope — CLAUDE.md §5). Below is a recommended bucketing into the four routine labels, then the open decisions and risks.

Recommended triage

A — mechanical small-fix (no design call needed; bundle into one claude/small-fixes-<date> PR):

• (3) res2display() order cap at 20. Highest-severity item here — silent wrong output for orders 21–29 now that the packed-u64 encoding reaches 29. res2display() in tools.py loops to 20 and prints km-only; extend to 29 with m-scale units. Pure, testable against pinned values.
• (8) mort2norm FORCE_PYTHON remnant. One function + ~5 lines (_python_mort2norm, the os.environ switch, the os import) — directly violates the §7 Rust-only guarantee. The other four twins were already removed in PR remove MORTIE_FORCE_PYTHON parity fallbacks #49; this is the leftover.
• (10) Benchmark scripts set MORTIE_FORCE_PYTHON. Now a no-op for the removed functions, so the benches silently measure Rust-only while claiming a comparison. Small update or deletion.

B — enhancement + implement (real feature surface, each its own issue/PR):

• (1) // Morton coarsening operator. MortonIndex // n → coarsen to order n, the deferred sibling of clip2order/moc_min from PR moc_min / common_ancestor: deepest-common-ancestor reduction #72. New public API + tests.
• (2) Morton __repr__. MortonCell(order=…, base=…) instead of a raw int — deferred from PR Update example notebooks for order-29 packed encoding + binder wheel #76, a genuine interactive-usability win.

C — needs an @espg decision before code (label discuss until resolved):

• (5) split_base_cells dict ordering — first-seen (traversal/insertion order) vs sorted-by-base-id. Blocks writing any determinism test (mirrors the prefix_trie.morton_polygon tie-break non-determinism #83 tie-break call).
• (6) split_base_cells Python wrapper vs Rust group_by kernel — a sorted variant could sort the existing dict in Python or be a new Rust group-by. Performance/placement call; (5) and (6) are coupled — one decision settles both.
• (9) Clippy useless_conversion (27 warnings) — load-bearing pyo3/numpy ? conversions flagged as false positives. Keep as-is (noisy) vs #[allow(clippy::useless_conversion)] at crate root. Already broadly under Version 1.0 Sweep (Follow up to new morton datatype) #48; just needs the yes/no.

D — policy/compliance question (not a code task yet):

• (4) Earthdata-gated notebooks not Binder-runnable. Three of the five updated notebooks need authenticated earthaccess downloads, which §4 says notebooks must not require ("publicly/anonymously reachable" data). This is a real compliance gap but the fix shape is a decision: (a) cache a small anonymous in-tree subset and rewire those notebooks to it; (b) mark them clearly as credentialed-only and exclude from the Binder set; (c) leave as-is and amend §4. Needs your call.

E — infra, out of routine scope without explicit sign-off:

• (7) CI versioning automation. Repeatedly flagged ("REALLY needs to be automated!!!", commit 1cf29c2). But §1/§7 bar the routine from touching .github/workflows/, the tag-driven publish, or release infra without you naming it explicitly. Recommend you either file it as a human-owned issue or explicitly authorize the routine to draft the bump workflow. I won't touch it otherwise.

Risks / unknowns

• Don't double-file. None of these ten have their own issue yet, but (5)/(6) are owned by PR add split_base_cells #84's open thread and (9) sits under Version 1.0 Sweep (Follow up to new morton datatype) #48 — relabel rather than open duplicates for those.
• (3) is the one with active correctness impact (silent wrong output today); the rest are cleanup, deferred features, or decisions. If you only greenlight one bucket, make it A.
• Sequencing: A and B are independent and safe to start as soon as labeled; C must resolve first or its tests can't be written; D and E are yours to direct.

Question (one pass)

Want me to translate this into actual labeled issues on the next run? Concretely: (i) open one small-fix PR for bucket A (items 3, 8, 10); (ii) file (1) and (2) as enhancement+implement issues; (iii) leave (4),(5),(6),(7),(9) for your decision (I can draft them as discuss issues if you'd like the thread). Reply with which of (i)/(ii)/(iii) to proceed on and I'll act. Marking waiting.

---

Generated by Claude Code