The archive, by niche

What these four numbers are, and are not

These four come from one function, saude_do_arquivo(), and the table shows what it returns — nothing derived, nothing inferred.

• Recording since and most recent window are window labels, and a window is rounded to the whole hour. The rounding is what makes a re-run in the same hour a no-op, and it also erases the real interval: in this archive, one pair of consecutive windows was 12 minutes apart and another was 96 minutes apart, and both read as one window. The exact instant is recorded, in a field called capturado_em, and this page does not show it.
• Windows in 24h is a count of distinct windows, not of rows — and a window only exists here once something was recorded in it. An hour in which the collection was attempted and brought nothing leaves no window behind, so this number counts what arrived, not what was tried.
• Items in that window is one window's count, not an average, and not a measure of size.

Why a niche has fewer sources than we collect

A source only appears under a niche once someone has read the titles it returned and confirmed they are about that niche. Of the 27 (niche, source) pairs with enough data to judge, read on 2026-09-30, sixteen were rejected — in one of them the niche term collided with the Hindi word for mother-in-law and returned 188 items of an Indian television drama.

Each niche page lists its own rejected sources and what each one actually returned. That check is a person reading titles, not something saude_do_arquivo() does: the function counts windows and items, and never judges whether content matches a niche.

Nothing in the archive is deleted. There is a retention job, and it touches only tables that hold an IP address, an e-mail or a funnel event — never the observations. A window recorded in September is still readable today, which is the whole point of keeping it.