How we source astronomy and space science

"Sourced" is doing a great deal of unexamined work in most science writing. A link at the bottom of an article can point at a paper, at a press release about a paper, at another outlet's article about the press release, or at a university's summary of a result its own researcher was a co-author on. Those are four different levels of evidence and they are routinely presented as the same thing.

Here is the hierarchy this publication works to, and — more usefully — what each level is actually prone to getting wrong.

Level 1 — the primary literature

A peer-reviewed paper, found through ADS [1], with a DOI or a bibcode. This is the strongest thing an article can point at, and it is still not certainty: peer review checks that a result is competently argued, not that it is true. Papers are wrong regularly and are corrected by other papers.

What this level is prone to: over-reading the abstract. Abstracts are written to be interesting; the caveats live in section 4. When this publication quotes a number from a paper, it quotes the uncertainty with it, because a measurement without an error bar is not a measurement.

Level 2 — the mission and the instrument team

NASA, ESA, JAXA, ESO, STScI, NOIRLab and the institutes that build and operate the instruments. Authoritative on their own hardware, their own schedules and their own data, and usually the first to publish a result about them.

What this level is prone to: enthusiasm about its own mission, and the gap between a press release and a paper. An agency announcing a discovery is telling you something true about what its scientists believe. It is not the same as a result that has survived review, and an article that treats it as one has quietly promoted it a level.

Level 3 — the catalogues and archives

SIMBAD [3], VizieR, Gaia, the NASA Exoplanet Archive [4], the Minor Planet Center [5], HEASARC, NED. These are the reference layer astronomy runs on: curated, versioned, and the actual answer to questions like how far away is that star or how many planets are confirmed.

What this level is prone to: being read as more settled than it is. A catalogue value is a best current estimate from a particular data release, and it changes. Two catalogues frequently disagree because they are reporting different measurements, not because one is wrong. Where this publication quotes a catalogue figure, it says which catalogue and which release.

Level 4 — preprints

arXiv, before review [2]. A preprint is a claim its authors are prepared to make in public. That is genuinely valuable — most significant results appear there first, sometimes years before the journal version — and it is not the same as a checked result.

What this level is prone to: being reported as a discovery. Preprints get withdrawn. Results get weaker between the preprint and the paper more often than they get stronger. Every article built on a preprint is labelled PREPRINT at the top, and the word "candidate" is not removed to make a sentence flow better.

What does not count

Other outlets' coverage. Content farms. SEO pages that rewrite an agency release with the numbers subtly altered. Social posts without a traceable source. AI-generated summaries — including of the sources above, because a summary is where the qualifier goes to die.

These are not banned out of snobbery. They are excluded because each one is a rewrite, and every rewrite is a chance for a hedge to be dropped by someone who did not know it was load-bearing.

The distinction that matters most

The single most common failure in astronomy reporting is collapsing this sentence:

A signal consistent with X was detected, though Y and Z remain viable explanations.

into this one:

Scientists detect X.

Both describe the same paper. Only one of them is what the paper says.

Every article here carries an evidence class precisely so that collapse has somewhere to be caught. When you see PREPRINT or INTERPRETED at the top of an article, that is not decoration — it is the publication telling you which of those two sentences it is writing.