Complete direct-source adapter checks

All six known-answer cases pass. Every physical model and processing case retains the supplied truth through training constraints, inherited conditioning, subdomain transfer and future prediction. Independent checks reconstruct the training rows and verify the future endpoints; truth is checked afterward.

The study uses its 36 fixture native oracle calls, 36 old/new bounds and 2,574 logical steps in 27.74 seconds. No aircraft observations or fits are used. The saved results and separate fixture/target reservations are verified.

Together with native qualification and the full-history screen, this clears the software prerequisites for a new complete generated width comparison. That comparison must cover all 48 regions and retain the original target. Passing here does not establish useful prediction widths, recording-scale adequacy or observational confidence.

Results · Frozen scope · Roadmap.