Summary

Runtime-side companion to the lance-graph cross-workspace plan cesium-osm-substrate-v1.md (separate PR landed in parallel). This is D-OSM-1 — the stub that mirrors the shape of crates/cesium/src/arcgis_pbf.rs for the OpenStreetMap PBF ingest path.

What this PR ships

• crates/cesium/src/osm_pbf.rs (~470 LOC) — stub carrier types + live Q3 boundary helper.
• crates/cesium/src/lib.rs — pub mod osm_pbf; (registered between arcgis_pbf and spz).

The live xyz_to_tms_y boundary helper

OSM slippy-map tile coordinates use XYZ-order (Y top-down, web standard); Cesium's implicit_tiling uses TMS-style (Y bottom-up). These are not the same key. The boundary helper performs the one-line subtract y_tms = (2^z - 1) - y_xyz at ingest — the only Y-axis conversion point in the OSM ingest pipeline. Pinning it here ensures D-OSM-2 cannot regress it accidentally.

Per I-LEGACY-API-FEATURE-GATED: wire-format-vs-runtime boundary converted at ingest; runtime sees only one shape (TMS).

Stub carrier types (no osmpbf dep yet)

OsmNode, OsmWay, OsmRelation, OsmPbfBlock, OsmPbfHeader — unit-struct placeholders with rich /// doc comments showing the intended Q1 (b) v1-fallback shape (tags: Vec<(Box<str>, Box<str>)>) that migrates to Q1 (c) Tag-as-Class identity in a follow-up sprint. decode_pbf(_bytes) returns OsmPbfError::NotImplemented — signature pinned now so D-OSM-3 (SPO lift) and D-OSM-5 (splat-fit-geo) can reference it without waiting on D-OSM-2 wiring.

Why no real osmpbf dep yet

osmpbf v0.4 (b-r-u) is the only production-grade Rust OSM PBF reader (multi-year stable, lazy-decoded, parallelized). D-OSM-2 wires it as a default-on dep behind an osm feature flag on the cesium crate. This PR keeps the ndarray workspace compile cost unchanged until that gate fires.

Test plan

• cargo fmt -p cesium — clean.
• cargo test -p cesium --lib osm_pbf — 5/5 unit tests pass (Y-flip zoom 0/1/10 + idempotency over zoom 0-20 + stub-returns-error guard).
• cargo test -p cesium --doc osm_pbf — 7/7 doctests pass.
• My code introduces zero new clippy warnings.
• D-OSM-2 PR wires osmpbf v0.4 dependency and implements decode_pbf.

Pre-existing CI block on master (NOT introduced by this PR)

Running cargo clippy -p cesium --all-targets -- -D warnings on the merge target (origin/master at e2ec430) without my changes surfaces 9 pre-existing errors:

• crates/cesium/src/fixtures.rs:412-415 — 4 × assertions_on_constants (Rust 1.95 lint upgrade; needs const { assert!(..) } wrapping).
• crates/cesium/src/to_cam_soa.rs:518-519 — 5 × arithmetic-side-effects (identity_op / always returns zero).

These predate this PR and are unrelated to OSM work. They will fail CI for any cesium PR until cleared. Recommend a separate cleanup PR fixes these in master; happy to file one as a follow-up if desired — fixes are trivial (wrap with const { … } for fixtures, audit two lines in to_cam_soa.rs:518-519).

Q1/Q2/Q3 rulings preserved at the wire boundary

Per lance-graph cesium-osm-substrate-v1.md §2 coordination outcome (locked 2026-06-05):

• Q1 (tags → IR): doc-commented carrier shows v1 (b) Vec<(Box<str>, Box<str>)>; migration note flags Q1 (c) Tag-as-Class target.
• Q2 (NiblePath): qk_tms_path: [u8; 24] on each carrier; encoded as qk:<level>/<x>/<y_tms> ASCII bytes; matches implicit_tiling.rs subtree shape.
• Q3 (Y-axis): xyz_to_tms_y is the live boundary helper; pinned at ingest per I-LEGACY-API-FEATURE-GATED.

What's NOT in this PR

• Real osmpbf dependency (D-OSM-2).
• Real SPO triple lift (D-OSM-3, lance-graph-ontology side).
• Real SIMD DEM sampling (D-OSM-4, separate ndarray PR).
• The 3D Tiles writer crate (D-OSM-6, the genuine Rust gap).
• Fix for the 9 pre-existing master clippy errors.

Cross-references

• Lance-graph canonical plan (parallel PR): .claude/plans/cesium-osm-substrate-v1.md
• Sibling substrate plan: splat-native-ultrasound-v1.md (D-SPLAT-1 Gaussian3D reused in D-OSM-5)
• OGAR coordination: Q1/Q2/Q3 rulings locked 2026-06-05; OGAR docs PR queued behind this set so they cite D-OSM-* by ID
• arcgis_pbf.rs precedent: 428 LOC; this file mirrors its shape (top-level doc comment with proto/format spec, stub carrier types, error enum with NotImplemented variant)

Cross-PR coordination

This is one of two coordinated runtime-side PRs for the cesium-osm-substrate-v1 arc. OGAR session has the green light to file their docs PR after these merge so they can cite D-OSM-1..7 by ID. The OGAR docs PR will extend DOMAIN-INSTANCES.md §2.6 (geospatial as 6th domain instance) and RDF-OWL-ALIGNMENT.md §10 Phase 2c (ogar-from-osm-pbf adapter sequencing).

Triggered by user feasibility question on OpenStreetMap × Cesium × Gaussian-splat coupling.

Authored by session claude/lance-graph-ontology-review-Pyry3.