Data and provenance on AsteriaStar

The stellar radius on a typical astronomy website is a number. On this platform it is a number, the catalogue it came from, the data release, the method that produced it, and a confidence grade — and if it was derived rather than measured, the formula and the inputs that produced it.

That is more expensive to build and much more expensive to maintain. Here is what it buys.

A number you can argue with

If a value is wrong, provenance tells you where it is wrong. Was the source wrong? Was the wrong release used? Was the derivation applied to inputs it does not hold for?

Without provenance, a wrong number is simply a wrong number, and the only available response is to replace it with another one and hope. It also makes disagreement legible: two catalogues reporting different radii for the same star are not a contradiction to be resolved by picking one, they are two measurements, and a platform that records which is which can show both.

An honesty envelope on anything live

Live data has a failure mode static data does not: it goes stale, and stale data looks exactly like fresh data. So every live value carries a status — live, recent, delayed, forecast, computed, stale, unavailable, provider_error — the provider's own timestamp, and the moment we fetched it [1].

Three rules follow, and they are the entire discipline:

• A provider that cannot be reached shows nothing. Not the last value dressed as current. Not

zero. Nothing, plus the reason.

• "Live" means a real observation inside the provider's own publication cadence. A six-hour-old

measurement is labelled as six hours old, however dynamically the page was rendered.

• Unavailable never becomes zero. "We could not reach JPL" and "there are no close approaches"

are different facts, and a platform that renders them identically is lying by arithmetic.

That last one sounds obvious and is the single easiest mistake to make. data?.length ?? 0 is a completely ordinary line of code, and it silently turns an outage into a reassuring statistic.

What each source is for

Providers are not interchangeable, and the platform records what each is authoritative about. NOAA's Space Weather Prediction Center [2] is the operational authority on geomagnetic conditions and is not a research catalogue. NASA/GSFC's eclipse canon [3] is a fixed computation — published once as a NASA Technical Publication and unchanged since, which is why nothing about it is ever called live, and why it is cached for a week rather than polled.

A provider used outside what it is authoritative for is a bad citation even when the number is right.

Measured, not asserted

The platform states uncertainties on computed values, and those statements are checked on every build against independently published instants — NASA's lunar-phase tables, the US Naval Observatory's season times, JPL Horizons positions. If a computation drifts past what the pages claim, the build fails.

This matters more than it sounds. The alternative — writing "accurate to about fifteen minutes" in a caption and never checking — is how a platform ends up publishing a confidence it has no evidence for. Twice during that work the check caught errors that were invisible in review: a frame mismatch putting an equinox nine hours late, and an approximation that placed Earth's aphelion a day and a third early. Both were obvious in measurement and invisible in code.

What it does not buy

Provenance does not make a value correct. A carefully sourced number from a catalogue that turns out to be wrong is still wrong.

What it buys is the ability to find out — and to say, precisely, what would have to change for the answer to change. For a reference platform, that is the difference between being useful and being merely confident.