Development evidence dataset: complete provenance indices

All 326 catalog entries and all 791 numerical-history rows are now indexed. The original 53 selected stretches and their 53 globe / zero flat / zero unresolved numerical preferences are unchanged. Version 1 remains sealed and independently checkable.

Acquisition and selection

The catalog census retains every URL, original-byte hash, reported size, observed byte length, storage evidence and eligibility state. The 232 retained sources contain 53 eligible stretches from 53 files; 179 retained files have no stretch under the current four-minute/high-speed policy. Retention does not itself mean scientific eligibility. Historical reference copies outside the pipeline are outside this storage census. No file was opened, removed or downloaded to produce the table.

Numerical operation provenance

The row index links every history entry to its exact publication record and execution manifest. Sixteen origin labels resolve to 23 execution manifests and 2 recorded numerical environments. Environment details preserve Python, NumPy, SciPy, thread settings and kernel metadata from each operation, rather than using today's environment or the environment of a later publication review.

There are 718 full-fit records, 10 conditional profiles and 63 endpoint-diagnostic records. These are not independent trials. Failed and interrupted results retain their original labels. The interrupted entry has no asserted evaluation count; its reserved allowance remains visible. Diagnostics may reuse an earlier fitted point: the direct source identity is preserved, but full point ancestry is not claimed where it has not been reconstructed.

Tables and interpretation

• All 53 stretches and 318 latest primaries.
• Complete numerical history, including failed records.
• Conservative recording groups and retained-file eligibility.
• Raw-gyro baselines, diagnostic studies, and validation gates.
• Input hashes, selection criteria, and assembly plan.

An IMU (inertial measurement unit) records turning and specific force. Its gyro increments supply turning observations; receiver positions supply route context. Numerical preference, descriptive gyro agreement, predictive adequacy and calibrated detection remain separate fields.

This is a development dataset, not the final validated paper dataset. The 27 conservative split groups are not verified independent flights. Qualified withheld predictions, observational covariance, independent confirmation and dependence-aware calibration remain open. Protected ESA measurements and model outcomes were not read. No new fit, model evaluation or original-file read occurred.

Run ~/venv/bin/python docs/paper-development-dataset-v2/build_dataset.py --check to verify both versions and their frozen inputs. --check-live also checks named current project inputs. --publish copies the completed snapshot without rebuilding tables or running analysis. --preflight validates current metadata joins without creating a snapshot. --check-public verifies the two public snapshots in a repository checkout without any private runtime directory or original corpus. Changed evidence requires a new version.

Previous version · Roadmap.