Complete prediction-width comparison with the direct source bound

The direct source method passes its mathematical/reference checks, all 24 native controls, the six full-history checks and all six complete-adapter known-answer cases. This study tests usefulness on the complete original generated prediction problem. Preserve earlier results and failures unchanged.

Retain all six model/processing groups and eight mounting regions: 48 fixed identities. Use the same six 1,600-packet sources, training at 0.8, 1.4, 2.4, 3.6, 4.8 and 6.0 seconds and prediction at 7.2 seconds. These are generated software cases, not aircraft replications.

For each region verify the original parent support certificate, reference and child-domain transfer. Recompute six training bounds and one future bound, retaining all 51 expression columns and 78 enclosing coordinates. Transfer reference-center differences outward under the unchanged 1-micrometre agreement limit. Preserve inherited constraints, the 1 mm generated observation allowance, arithmetic margin, eight contraction passes and 1,024-row ceiling.

Independently reconstruct the new training rows and audit all six future endpoints. Only after the prediction or empty proof is saved may known truth be checked. Union every surviving region within each group; retain both the fresh range and its intersection with the qualified 192-box range. Every axis of every complete group must meet the original twofold-width target. Missing or failed regions prevent qualification. Do not select favorable partial unions.

Separate finite allowance and cost

• 48 identities, 336 propagation bounds and 4,838,400 logical steps.
• 48 reference setups and child proofs; 288 exact parent support verifications.
• 336 setups each for source rows, domain derivatives and point derivatives; 537,600 steps each, plus 537,600 direct-formula steps and source-body bounds.
• 1,728 training-row checks, 48 contractions and 48 truth audits.
• At most 288 one-second LP proposals and 1,152 helper/independent future support checks, in addition to the 288 parent checks.
• Four workers with one numerical thread each. Preserve the 120-second bound limit; 7,200 seconds overall, checked cooperatively.
• No retries, allowance expansion, signals, native trajectory oracles, fits, aircraft reads or protected ESA observations.

The full-history future bounds take approximately 64–114 seconds. Budget roughly 75–100 minutes for the complete comparison, with a two-hour ceiling. Freeze sources, inputs, targets, compiler and environment before launch. The coordinator publishes saved results after completion without refitting.

Passing establishes usefulness only on this short generated problem. Recording-scale error bounds, whole-program floating execution, observational adequacy and calibrated confidence remain distinct requirements before new aircraft conclusions.

Adapter qualification · Full-history screen · Previous complete comparison · Roadmap.