Only 267 of 41,884 have a measured composition.
That is 0.64% of the catalog. The overwhelming majority of near-Earth asteroids have never been spectrally characterized, and hand-curation cannot keep pace with the modern flood of discoveries. Scroll, and watch the model close the gap fivefold.
Real trajectories, validated.
A universal-variable Lambert solver and a porkchop sweep find the minimum-Δv launch window for any target. It recovers asteroid Eros's optimal Δv at 6.10 km/s against the published Benner value of 6.11: screening-grade, but genuinely correct.
Predicting composition from reflectivity.
A class-balanced model reads an asteroid's albedo and orbital dynamics and reduces them to a compositional class with a calibrated confidence. It reaches 82% accuracy against a 72% baseline and expands the characterized population fivefold to 1,354, every prediction tagged, never disguised as a measurement.
Worth, built from what's actually there.
A single metallic near-Earth asteroid can hold more platinum-group metal than humanity has ever mined. Value is assembled from recoverable resources, water, iron-nickel, and platinum-group metals, never asserted as a single number, and always carried with an uncertainty band.
Honest by construction.
Every figure is an estimate with explicit uncertainty and a provenance tag: measured, ML-predicted, or assumed. As certainty drops, the band widens, on screen, in the open. Lodestar decides which asteroid to study or visit next and roughly when. It is a screening and decision-support engine, never flight software, and it never claims to be more certain than it is.