Does the map get the numbers right? Every gate below runs here, in your browser: a textbook Lambert solution, the Hohmann limit and multi-revolution branches, propagation round trips, the float32 shader against the Float64 solver, the ephemerides against known positions, and the physics of the 2026 Mars window. Then the timings: the GPU redraw, the CPU fallback, the 3D viewer's frame rate and a five-leg gravity-assist optimisation.
| Gate | Measured | Expected | Status |
|---|
Lambert. Example 5.2 of Curtis, Orbital Mechanics for Engineering Students: an Earth-centred prograde transfer over one hour; the textbook velocities are quoted to five figures. The Hohmann case is two coplanar circular orbits at 1 and 1.524 AU joined by a 180° transfer in exactly the Hohmann half period; the chord is then collinear with the Sun and the plane of the transfer is undefined, so the solver takes the ecliptic normal and the analytic Hohmann Δv must come back at both ends. For a longer flight time the same geometry admits one-revolution solutions; both are propagated with the integrator to check that they really reach the target.
Shader against CPU. The shader works in float32 with days since a reference epoch, AU and AU/day, and the Lambert problem scaled by |r₁|; the CPU works in Float64, km and seconds. Two hundred random cells of the current map are compared. The departure and arrival days of every pixel are quantised to 1/64 and 1/32 day, so the shader's stored time of flight is the same float32 number the CPU computes, bit for bit.
Ephemerides. JPL's approximate elements for 1800–2050 are good to arcminutes for the inner planets and a fraction of a degree for the outer ones; the checks here are deliberately loose (a degree, 0.002 AU, 0.2 km/s), and the oppositions of Mars (2025-01-16) and Jupiter (2024-12-07) test that Earth and the planet share a heliocentric longitude on the right day.
Mars window. The minimum departure Δv from a 300 km parking orbit for the 2026 opportunity should fall between 3.5 and 4.1 km/s with a 180–320 day flight, and the window finder, started at the beginning of 2026, should place the first window in October–December 2026 and the next about 26 months later.
What is not tested. The elliptical-capture and aerocapture budgets are simple energy bookkeeping and are not checked against a mission; the gravity-assist optimiser is timed, not validated against a flown trajectory beyond the Cassini demo on the main page; and the N-body check uses point masses at the ephemeris positions, with no relativity, no Moon and no planetary oblateness.