The evidence policy

Two of these do most of the work in practice. Typical characteristic covers nearly everything anyone wants to say about a category of drink, and it is the type most often written as though it were a fact: farmhouse cider is not dry, it is commonly dry, across a range wide enough that the exceptions are not exceptions. And measurement is the type that carries the whole context model below.

The last of them, structured inference, is what CiderHQ derives for itself — a blend direction, a pairing match, a serving temperature where no source states one. The rule for those is that they must be explainable: the reasons that produced the result have to be displayable, or the result is not shown. How each kind is produced is set out in the methodology.

Needs verification is the useful one. It lets a contributor record something they have encountered — a figure in a book, a claim made in a forum, a number repeated in three places with no attribution — without either publishing it or losing it. It sits in the data, invisible to readers, until someone checks it against a source of adequate authority or establishes that it cannot be checked.

Disputed is different, and it is published. Where credible sources genuinely disagree, or where the correct answer legitimately varies by context, CiderHQ states the disagreement rather than picking a side and presenting it as settled. A page that explains why two authorities differ is more useful than one that quietly follows whichever was consulted last.

Most sites that show sources show a list of links. A link tells you nothing about whether anyone opened it. The commonest form of citation dishonesty on the web is not fabrication; it is a bibliography assembled from search results and presented as reading.

CiderHQ refuses that trade. Where a source has been identified as the body that defines a subject but nobody has yet worked through the document, it is recorded as registered, its access date is left null, and the page says so. That is a true statement about the state of the work, and it is more useful to a reader than a fabricated access date would be. It tells you exactly where the authoritative answer lives, and it tells you that CiderHQ has not yet been there.

It also constrains what may be published. A source in this state can support a statement that a body is the authority for something. It cannot support a quoted figure, a specific threshold or a passage attributed to it — those require the passage to have been located and dated. So the honesty is not merely declarative: the register’s own state field limits what the site is allowed to say.

The consequence is visible on the source register, where the consultation state and date of every source are shown as recorded. Where a retrieval was wanted and did not happen, the attempt is recorded as a gap along with what CiderHQ publishes instead — which is usually nothing.

Consider a single question: how much acid is there in Dabinett juice?

The Long Ashton Research Station analysed cider fruit systematically from 1903 until it closed in 2003, at a site in Somerset, in a maritime English climate, using the analytical conventions of its day — total acid reported as malic acid in percent weight per volume. Cornell’s cider programme runs cultivar trials in New York State, in a continental climate with hotter summers and colder winters, and reports juice chemistry in the modern convention of grams per litre.

If the two disagree, neither is wrong. Acid falls as fruit ripens further, and a warmer season ripens fruit further; the same cultivar on a different rootstock, in a different soil, picked a fortnight later, genuinely carries a different acid load. A Somerset figure from the 1950s and a New York figure from the 2010s are two measurements of two different things that happen to share a cultivar name.

So CiderHQ does not average them, and will not. Averaging destroys the only information that made either number usable: where the fruit grew and when it was measured. What it produces instead is a figure with a false air of authority — one number, three decimal places, no context, endlessly copyable. That single number is how a historic Somerset analysis becomes a universal constant in every subsequent article, and it is the failure mode this entire model exists to prevent.

What the site does instead is show them side by side, each with its location, period, sample count, method and source, and let the spread be the answer. Where several analyses exist, the page says how many there are and states plainly that they are not averaged. Where only one exists, the page says that too — a single historic figure is a single historic figure, not a property of the fruit.

A conversion looks harmless and is not. Percent weight per volume is what historic UK juice analyses used for acid and for tannin; grams per litre is the modern convention. The arithmetic between them is trivial, but performing it silently means that a figure a reader traces back to a 1950s table no longer looks like the figure in that table, and the reader cannot tell whether the difference is a conversion or a discrepancy. Storing what the source reported keeps the audit trail intact.

Two of these units exist purely to stop rounding from destroying meaning. Micrograms per litre is kept apart from milligrams per litre because most aroma-active compounds and most contaminant limits live at that scale, and a threshold of 0.05 mg/L rounds away entirely at the precision milligrams are shown to. Parts per million is kept apart from milligrams per litre even though they are numerically equal in dilute solution, because sources use both and the reader should see which was used.