RSH-FRENET-PARALLEL-V1 · SE(3) PREFIX SCAN

One recurrence.
Many ordered transforms.

The f64 Rust/WASM reference converts each interval into a local rigid transform. WebGPU builds those intervals in parallel, composes them through deterministic doubling passes, emits the complete path, reads it back, and compares every component before any benchmark statement is allowed.

PARALLEL RESEARCH PROFILE

Run conformance and timing together.

Loading the parallel Rust/WASM module…

CURRENT EVIDENCE

LOADING

WebGPU adapter not requested yet.

Scan passes
—
Max position residual
—
Max frame residual
—
Max schedule residual
—
Frame norm error
—
Frame orthogonality
—
WASM median
—
GPU median + readback
—
Observed speedup
—
Claim state
NONE

An observed speedup is scoped to this adapter, browser, sample count, and timing method. It is not a universal hardware claim and never replaces the canonical geometry receipt.

WHY A NEW CONTRACT?

Associativity has rules.

The v2.6 sequential path applies modified Gram–Schmidt after every step. That nonlinear projection cannot be inserted into an associative prefix operation without changing the mathematics. The parallel contract therefore composes midpoint Rodrigues SE(3) transforms in a named doubling order and validates that order independently.

local interval transforms
        ↓
build_local[0 … n−1]
        ↓
scan offsets 1, 2, 4, 8 …
        ↓
capture midpoint → emit centred path
        ↓
readback + f64 residual gates + benchmark sidecar