Broader development intervals: fixed metadata rule
Selection
Use only the two eligible runs in the verified full-motion inventory for recording 187. Keep its source hashes, quality criteria and conservative development group. Do not rerun selection or inspect gyroscope values and model preferences to choose intervals.
For each run, select its first and last 2,401 receiver records. Each range is nominally twenty minutes. If the two ranges overlap, retain the first and record why the last was omitted. Never shift a range to obtain an easier fit.
Within a range, use offsets 1, 3, …, 2,399 as the 1,200 query epochs. The earlier 800 are training observations; the later 400 are prediction observations. Keep the original first twenty-minute pilot in the inventory as already used, so it cannot silently become another new test.
All selected ranges belong to the same development group. They can test consistency over different parts of that recording, but cannot establish independent replication. Seven ESA confirmation groups remain protected.
Metadata-only scope
Read the sealed inventory and preparation reports, checking their hashes and current development membership. Freeze the rule, source/environment binding, software checks and exact receiver-index ranges before future fitting. At most sixteen parent runs and thirty-two ranges are supported by this planning interface; excess metadata is rejected rather than silently truncated.
This step opens no original recording, decodes no measurements and performs no propagation or fit. It supplies no new timing, speed, heading, latitude or height statistics for individual windows. Parent-run averages must not be presented as window averages. The rule requires no numerical thread allocation.
Later gates
The ranges remain candidates until preparation independently checks their exact timing, source sequence, full IMU support and unchanged source hashes. Require actual support duration within half a second of twenty minutes; do not repair gaps or interpolate measurements. Preserve turns and climbs for explicit motion reconstruction. Preparation must save window-specific trajectory summaries.
Preparation and model fitting need separately qualified finite definitions after the current error-assumption assessment. This metadata freeze authorizes neither step and consumes no fitting allowance. Preserve numerical failures, physical accuracy failures and all preceding study limits.