# Evidence note — shape-controlled gravitational propulsion

This companion accompanies working paper 028, web edition 029. It reports the completed 027 finite study and does not add a scientific run.

## Scope

Newtonian point-primary gravity, three equal onboard masses, static configurations. Thirteen cases compare exact force with point gravity and a leading quadrupole. Three imposed work maps compare forward, reverse and equal-radius paths. Isolated, uniform-field and linear-field controls check the stated no-thrust limits. The work map is not a trajectory. The primary is the external gravitational partner.

## Recorded observations

Swift, C++ and Objective-C each ran once and exited successfully, without retries. C++ compared the producer's TSV and rejected two malformed controls. Objective-C accepted the original JSON and a whitespace reencoding; its inspected passing sequence required fifteen negative mutations to be rejected. Its runtime output is aggregate, not a per-mutation log. The original values are in [the JSON packet](propulsion-evidence.json).

For geocentric R=7e6m, half-span L=100m, mass 1000 kg, the radial and transverse differences are approximately -3.32 and +1.66 micronewtons relative to point gravity. Both total forces remain inward. The synthetic forward work map gives equal shape/center external-potential work of 1.3126816601649605e-5J. Reverse changes the sign; equal radii give zero. No full actuator/structure/attitude/heat model is supplied.

The finite tolerance is abs(actual-expected)<=2e-12*max(1,abs(expected)). Both C++ floating-point types report 53 significand bits on the tested platform; this is not an extended-precision reference or a universal error bound.

## Sources and independence

Landis's [NASA conference record](https://ntrs.nasa.gov/citations/19910012850) directly precedes the broad gravity-coupled mechanism. See the paper's references for access distinctions and the record's publication-date discrepancy. This is a targeted source trace, not a comprehensive novelty or patent assessment.

The requested model labels were gpt-6-astra, gpt-6-sol and gpt-5.6-terra. Shared context/files and unblinded synthesis limit independence; serving builds are unknown. The models and coordinator are not independent scientific certification. The post-study revision is coordinator synthesis.

Novelty, a new force, useful propulsion, a realizable flight cycle and measured human benefit remain unestablished. These open questions are distinguished from the recorded finite calculations.