Computed, not fetched

Most space maps on the web are front ends for somebody else's API. Ask where Mars is, wait for a server to answer, draw the answer.

Mars Deck does not do that. It makes no network calls to draw the map at all.

What that actually means

The engine is called ORRERY, after the clockwork solar systems built in the early 1700s — brass planets on geared arms, turned by a handle, made for people who wanted to understand how the thing moved rather than merely look at it. Ours is the same object in a different material, with Kepler's equation where the gears were.

Planetary positions come from JPL's published Keplerian elements and are solved in your browser, every frame. The 6,200 asteroids, Trojans and trans-Neptunian objects carry real orbital elements from JPL's Small-Body Database — fetched once, on our machine, when the site is built, and shipped to you as a static file.

The practical consequences:

• It loads instantly, because there is nothing to wait for.
• It works with no network at all.
• It cannot be rate-limited, because there is nothing to rate-limit.
• It cannot be killed by somebody else retiring an API.

That last one is not hypothetical. Half the space widgets built in the last decade are dead for exactly that reason.

The part that keeps us honest

Computing your own numbers means you own the mistakes. So the engine is checked against independent astronomy rather than against itself — 547 verification checks, run before every single deploy, which refuse to let a failing build ship.

A few of them:

• Mars oppositions land within 0.8 days of independently published dates.
• The Lambert solver matches the closed-form Hohmann transfer to **five
• The propagator closes a 295-day interplanetary transfer to 1.5 metres.
• The Jupiter Trojans cluster at +62.4° and −62.6° either side of Jupiter,

That last one is the check we trust most. It is very hard to fake by accident.

The honest limits

Positions are accurate to arcminutes, which is far beyond what any screen can show — but this is not navigation-grade software and we will not pretend otherwise. JPL's planetary element set is stated as valid from 1800 to 2050, so that is the range we stand behind. The asteroid field is propagated from a 2026 snapshot without modelling planetary perturbations, so it drifts over years: excellent as a map of the belt, not a substitute for an ephemeris.

Saying so costs us nothing. Being caught not saying it would cost everything.