feat(contract): rail_geometry — the address places the node

.claude/board/LATEST_STATE.md

@@ -1,3 +1,33 @@
+## 2026-08-13 — rail-trie geometry registered: the address places the node
+
+### Current Contract Inventory — 1 new module, 1 new ClassView resolver
+
+- **`contract::rail_geometry`** — deterministic node placement from the rail
+  registers: `RailAxis` (Taxonomy/Mereology), `RailCarving`
+  (`InterleavedPairs` 6×stride-2 in the key facet | `AxisSlab` 12(+12
+  discontiguous cont)×stride-1), `RailPath` (hole rule: a value after a zero
+  is not ancestry; empty path = the lane's dominant root), `TriePlacement`
+  (`ring` = depth, `arc` ∈ [0,1) = radix fraction of the slots), and
+  `dual_rail_placement` (primary axis PLACES, secondary OVERLAYS — two
+  hierarchies on one canvas, only one of them places). The neo4j-shaped
+  invariant is proven, not styled: a child's arc lands inside its parent's
+  half-open interval `[arc, arc + 256^-depth)`, siblings order by slot,
+  and the placement is a pure function of the row — two loads render
+  identically, no solver, no scene model. The f64 boundary is pinned as a
+  passing test (exact through level 6; order-preserving beyond; a glove
+  needing deeper discrimination reads `slots()` directly).
+- **`ClassView::rail_carving(class, axis)`** — the reading, registry-resolved
+  per class like `edge_codec_flavor`: default = canon zero-fallback (key
+  facet pairs at `4..16`); a bake that measured its way to a different
+  carving overrides. The slab variant exists because a consumer bake
+  MEASURED the pair reading and rejected it for its hierarchy (44.25 % of
+  paths fit vs 99.62 % in twelve per-axis levels). Selection only — no
+  carving changes `NODE_ROW_STRIDE`.
+- **Boundary kept:** not a renderer (every glove reads ONE resolved
+  placement; none re-derives), not a distance (the CLAM-side geodesic lives
+  with the compute crate). Compute-side counterpart: ndarray `clam_v3`
+  (`RailSpec` mirrors the two carvings; merged there first).
+
 ## 2026-08-12 — open-review sweep of #920–#929 — three ledger figures were wrong, and one audit method cannot do what it claims
 
 ### Current Contract Inventory — no new types (corrections to merged #927/#928 entries; append-only, corrected here not in place)

crates/lance-graph-contract/src/class_view.rs

@@ -1110,6 +1110,28 @@ pub trait ClassView {
         crate::canonical_node::EdgeCodecFlavor::CoarseOnly
     }
 
+    /// How this class's rail register is carved, per axis — the reading a
+    /// renderer (or a tree-distance) uses to lift the node's rail path out of
+    /// the row. Same registry-resolution pattern as
+    /// [`edge_codec_flavor`](ClassView::edge_codec_flavor): the default is the
+    /// canon zero-fallback (the key facet's interleaved `X:Y` pairs at
+    /// `4..16`), and a bake that measured its way to a different carving —
+    /// e.g. per-axis value slabs — overrides per class. Selection only: every
+    /// carving reads within the existing row, never a stride change.
+    ///
+    /// Downstream, [`rail_geometry`](crate::rail_geometry) turns the resolved
+    /// carving into a deterministic placement (`ring` from depth, `arc` from
+    /// the slots) — the key prerendering the node, zero value decode, one
+    /// resolved projection under every glove.
+    #[inline]
+    fn rail_carving(
+        &self,
+        _class: ClassId,
+        axis: crate::rail_geometry::RailAxis,
+    ) -> crate::rail_geometry::RailCarving {
+        crate::rail_geometry::RailCarving::zero_fallback(axis)
+    }
+
     /// Which value-slab schema preset this class materialises in
     /// [`NodeRow::value`](crate::canonical_node::NodeRow::value).
     ///

crates/lance-graph-contract/src/lib.rs

@@ -130,6 +130,7 @@ pub mod plan;
 pub mod property;
 pub mod proprioception;
 pub mod qualia;
+pub mod rail_geometry;
 pub mod rbac;
 pub mod tekamolo_facet;
 pub use qualia::{