Finite source-prefix equivalence controls
Use eighteen identities: three source streams (reference, domain derivatives, point derivatives), three generated candidate configurations and two processing configurations. Source arithmetic itself is independent of the Earth candidate; these labels keep complete generated-case coverage without implying a shape test.
Each source has sixteen nonuniform packets. Copy prefixes after two and eight packets, then evaluate a knot-related query and its adjacent binary64 values. Compare every remaining source value and all 27 derivative columns bit for bit with independent original cold streams. State-header sizes intentionally differ. Check that each copy leaves its parent unchanged and retains its original packet position and allowance. Reject a clipped history, an earlier/boundary query and a changed reference binding.
Fixed work allowance
- Eighteen identities; one worker and one numerical thread; at most 600 seconds.
- Per identity: 174 compiled source steps, nine fork attempts, 66 compared tail frames and six parent-isolation checks.
- Total: 3,132 source steps, 162 fork attempts and 1,188 tail frames.
- Zero geographic integrations, optimizer starts, aircraft reads or fits.
- No retries or expansion; shared freeze, journals and sealed outputs.
Passing qualifies these typed source-copy operations only. Full native reuse still requires geographic, attitude, force-integral, common-clock and conditioning state equivalence. It must not replace a native bound until that gate passes. The 53-stretch evidence, physical bounds and prediction targets are unchanged.