Methods
How accurate are the numbers this site computes? Its transit, orbit and radial-velocity models and its sampler are compared here with twelve other codes and with a reference computed independently of all of them. The other codes’ answers are stored; this site’s side of those four comparisons is recomputed in your browser each time the page opens. The other models (phase curves, moons, N-body dynamics, photometry, catalogues) have their own tests, summarised in section 5 from stored runs.
Loading the stored references and recomputing…
1The short version
Four things are tested: how much a star dims when a planet crosses it; where the planet is on its orbit, and when; how fast the star moves toward and away from us; and whether a fit returns the right values with honest error bars.
Every code gets identical inputs. The answers are compared with a reference computed independently of all of them (the orbit in 30-digit arithmetic, the blocked light by numerical integration), against tolerances fixed in advance.
2Where this code is weaker, or wrong
3Where another code is the outlier
When codes disagree, the reference shows which one is off. Five cases where it was not this one:
4The four comparisons
Each gives a number to remember and one chart; the rest opens on request. The equations behind them are in the Methods tab: transit model, orbits, radial velocities, nested sampling.
4.1The dimming of the star
The fraction of starlight a planet blocks as it crosses the star, for about forty combinations of planet size, path, orbit shape and limb darkening. An error here becomes an error in the planet’s radius.
4.2The orbit and its timing
Where the planet is at a given time (Kepler’s equation), when it passes behind the star, and how long transits and eclipses last. Elongated orbits are the hard case: a solver that stops early puts the planet in the wrong place.
4.3The star’s velocity
The speed of the star along our line of sight as its planet pulls it around. Velocity codes are known for differing by a sign or by 180° in the angle ω, so the conventions are tested along with the numbers.
4.4Fits and their error bars
Four simulated data sets are fitted with this site’s sampler and, independently, with emcee, dynesty and UltraNest. The sampler should not matter, and a 68% interval should hold the truth 68% of the time.
5Checks elsewhere in the project
The rest of the site has its own tests against independent software, summarised here from their last stored run.
6What has not been tested
Rerun the validation
Every number of the Validation tab, from stored references and this repository.