Joint Radar reference indexing comparison

All six alternating fresh-process runs are retained. The complete prepared-input calibration workflow changes from a median 49.293604 to 11.866233 s (4.154107x). The solver API changes from 45.601982 to 9.320259 s (4.892780x). The host was unreserved; variation is material and every timing is reported. No run is discarded.

The complete input control is byte-identical to the preceding profiled-Radar comparison. It contains all 2,762 public KITTI OXTS reference samples and 2,758 generated Radar-frame velocities, with deterministic noise and unequal weights. This is performance and parity evidence, not independently measured Radar or calibration accuracy evidence. No locked accuracy cohort is evaluated.

This workload fits all seven variables from an identity transform and zero clock offset, with the original 80-iteration bound, robust LM, initial-residual gate, train/holdout split, diagnostics and fourteen signed holdout probes. All six runs converge in five iterations, retain 2,099 eligible measurements and the same 659 initially rejected IDs, and make 171,502 interpolation calls. All numerical output, optimizer history and diagnostics match exactly.

Whole-process timing includes imports, input parsing, object construction, initial eligibility checks, complete optimization, interpolation counting, all diagnostics, serialization and saving. Downloading, control generation, source freezing and external auditing are untimed. Both variants use the same dependencies and one BLAS thread. protocol.json pins the initial transform, all solver options, actual source roots/commits, complete source capsules and manifests, launcher/helpers/schema, and input/control-generator hashes. Each worker records its actual imported files and normal Git identity. Scratch Git directories are not redistributed. The only changed runtime module is calibrex/solvers/radar_joint_spatiotemporal_solver.py.

control-generator.py.txt retains the original input-generation source; launcher.py.txt retains the measured workflow source. The download receipt and fetcher bind the original ZIP members by SHA-256 and CRC-32. index.json binds every other published byte, including empty stderr and validation logs.

The KITTI source data and derived numeric control are attributed to Andreas Geiger, Philip Lenz, Christoph Stiller and Raquel Urtasun, KITTI, under CC BY-NC-SA 3.0. The official KITTI raw page and original archive URL are retained in provenance. Source-code capsules retain the project's Apache-2.0 license.

Audit without raw data, a GPU or a calibration run:

python -m tools.benchmark_radar_reference_indexing --audit-only docs/assets/joint_radar_reference_indexing

The benchmark report contains every timing and the remeasurement command.