Methodology

How a figure gets onto this site, and what we refuse to do to put it there.

What an observation is

The unit of this site is not a price. It is an observation: this parameter held this value, observed at this time, at this URL, whose content had this SHA-256 fingerprint. We archive the fingerprint, which is what turns a claim into evidence. It lets us show what a source said on a given day, after that page has changed or disappeared.

An observation is immutable. We never edit or delete one. The database enforces this with triggers that reject any update or delete, so it holds even against our own mistakes. Corrections are made by adding, never by rewriting.

How a change is recorded

A change is never typed in by a human or generated by a model. It is derived from two successive observations that differ. Before it is published it must pass four filters: confirmation by a second observation unless the source is level 1, a materiality threshold of 0.5%, comparison at identical unit and currency, and a redesign guard that suspends writing when more than 30% of a source's parameters move at once, because that usually means the page was rebuilt rather than repriced.

How we rank sources

A level 5 figure is a lead, not a fact. It tells us where to go and verify, and it flags disagreements. Right now this rule holds back 17 465 parameters, which we hold but do not publish. We would rather show an empty cell than a confident wrong one.

What the language model is allowed to do

No fact on this site is produced by a language model. Not a price, not a quota, not a date, not a limit. A model is used to translate, to write the surrounding context, and to classify, always working from figures that were already verified. It dresses the value, it never touches it.

This is enforced mechanically, not by good intentions: every generated text is scanned for numbers and dates, and any figure that does not appear in the verified input is rejected and the text regenerated.

Freshness, including when it is bad

Each parameter has an expected re-check cadence. When one has not been re-verified within three times that cadence, the page says so, in place, rather than hiding it. Showing our own weak spots is what separates a source from an aggregator.

How we collect

Only public sources, never behind authentication, a paywall or a captcha, and never by circumventing a technical protection. Our crawler identifies itself as BertrandeBot with a link back to this page, reads at a low rate, and we publish facts with a link to the source rather than republishing anyone's content.

On terms of use we owe you a precise statement rather than a comfortable one, so here is the uncomfortable version. We do not treat robots.txt as our rule. It is a convention addressed to crawlers that copy pages into an index. We are not that: we read a small number of public price pages, at a low rate, and we publish figures with a link back rather than republishing anyone's content. The owner of this site took that decision on 2 September 2026 and stands behind it. We report what each source's robots.txt says on the sources page, because it is a fact about that source and you are entitled to it, but we do not present it as a permission we obtained.

What we never do, and this is not a matter of judgement: read anything behind authentication, a paywall or a captcha, or circumvent a technical protection of any kind. And we stop tracking any source immediately on request from its publisher, without argument and without asking why.

Beyond that, the written terms of each source are read, dated, fingerprinted and given a verdict, and those verdicts are published on the sources page, including when the verdict is unfavourable to us. A blanket claim that we respect every source's terms would be easy to write and impossible for you to check, which is the opposite of what this site is for.

If you publish a source and want us to stop tracking it, write to contact@bertrande.com and we will act on it without argument.