Evidence dataset
The development snapshot, version 2 brings the selected observations, numerical results and validation status together without changing the completed analysis.
| Included | Coverage |
|---|---|
| Selected stretches and source/track metadata | 53 |
| Latest primary model/processing-case comparisons | 318 |
| Preserved numerical-history records | 791 |
| Conservative development split groups | 27 |
| Retained-file eligibility outcomes | 232 |
| Raw-gyro geographic baseline comparisons | 159 |
| Catalog entries with reconciled storage/eligibility | 326 |
| Numerical records linked to publication and operation environments | 791 |
The history includes 244 records not marked converged. They remain visible alongside later successful work; some entries are diagnostics rather than optimizer starts. The latest complete numerical disposition remains 53 globe, zero flat and zero unresolved.
An IMU (inertial measurement unit) records turning and specific force. Its gyro increments supply the observed turning; receiver positions supply the route context. The snapshot keeps those roles, source fingerprints and unknown instrument identities explicit. Conservative groups do not establish a count of independent flights.
This version is for development and audit. Qualified withheld-prediction adequacy, observational error assessment, independent confirmation and calibrated decisions remain open. The acquisition-flow and operation/environment indices are now complete: 58 initial no-level rejections, 35 later rejections and one verified duplicate removal reconcile the 326 entries with 232 retained files. Sixteen history origins resolve to 23 execution manifests and two recorded numerical environments. Later publication reviews retain their own environment separately from the operations they describe. The snapshot does not label numerical preferences as calibrated detections or claim to be the final paper dataset.
The tables and checker use frozen input
copies and recorded SHA-256 hashes. Later project changes cannot silently alter
this version. Assembly and checking read saved metadata, not original recordings,
and run no model predictions or fits. --check-public works in a repository
checkout without the runtime data directories. Protected ESA outcomes are not
part of the dataset. Version 1 remains sealed.