Long-history screen of source derivative bounds

Launch only after the separately frozen source-derivative native qualification passes all 24 controls. Preserve its candidate and unchanged physical bounds.

Use the same six generated 1,600-packet sources, saved training constraints, calibration boxes and future query times as the completed source screens. Compute one new bound per source. Compare identical reference maps first, then retain the minimum of the original, compiled, batched-source and new allowances on each axis. Do not round-trip those bounds through displayed metre values when forming the minimum.

Separate finite allowance

• Six identities, one bound each, at most 76,800 logical bound steps.
• Six setups and 36 exact training-support checks.
• Reserve 9,600 source-reference steps and 9,600 trained body bounds.
• Separately reserve six derivative setups and 9,600 compiled derivative appends covering reference/domain midpoint/end states.
• Four single-thread workers; cooperative 120 seconds per bound and 600 seconds overall. No retries, native trajectory evaluations or optimizer starts.
• No aircraft originals, LP proposals or protected ESA observations.

Freeze sources, inputs, comparison artifacts, compiler and environment before launch. Preserve failures and charges. Passing establishes source-map compatibility and completion on these histories. A separately frozen complete prediction comparison must still meet every original horizontal and vertical width target before withheld aircraft fitting proceeds. These are software development calculations, not new observational evidence or calibrated coverage.

Native qualification · Previous long-history screen · Roadmap.