Qualify an exact integer source-bound implementation
Short runtime profiles identify repeated general-fraction construction as a substantial cost. This candidate represents binary inputs as exact integers over powers of two. Arbitrary rational value endpoints remain exact; nonbinary row inputs retain the original fallback. No rounded arithmetic replaces an exact sum, and no physical assumption changes.
Repeat the complete native software-control matrix as a separately frozen implementation qualification: all three candidates, both processing cases, center, stationary-pose, derivative and nonlinear-containment paths. Compare against every saved source-derivative predecessor bound. Require identical centers, all three coefficient matrices, physical bindings, source counts and error-remainder components, not merely containment of sampled points.
Separate finite allowance
- 24 fixed control identities, 78 native oracle calls and 19,632 packet visits.
- 348 propagation bounds and 60,288 logical bound steps.
- 348 source and derivative setups; reserve 7,536 appends for each stream and 7,536 source-body bounds. No LP calls or observed optimizer starts.
- Four workers, one numerical thread each; four threads total.
- The existing 60-second cooperative limit per bound; 600 seconds overall.
- No retries, allowance expansion, process signals, aircraft-original reads or protected ESA observations. Preserve every outcome and charge.
Component tests compare exact-rational outputs, extreme binary scales, both source phases and malformed inputs. Native equality is a separate requirement. Passing this study cannot establish long-history speed, narrower ranges, whole-program roundoff coverage or calibrated observational uncertainty. Those checks need their own finite freezes.