Why every chart here is frozen

11 September 2026 · Analitik

A chart wired straight to a live API looks like the responsible choice. It is not, for anything with prose attached to it.

Write “prices rose 19% since March”, publish it beside a live chart, and walk away. Six months later the sentence is still there and the chart is showing something else. Nothing broke. No error fired. No test failed. The article is simply now wrong, and it became wrong quietly, in a way nobody is monitoring.

So every post here that cites a number commits a snapshot — a JSON file captured at publish time and versioned alongside the markdown. The prose describes that snapshot. The reader gets a visible badge saying when the figures were pinned, plus a link to the live source if they want today’s numbers instead.

The cost is one file per post. What it buys is that a piece can be wrong when published, but it cannot become wrong afterwards without someone editing it.

The second rule: the newest week is usually a lie

Weekly collection is not instantaneous. The most recent period in almost any rolling dataset is partial — collection is still in flight, late submissions have not landed, and the aggregate is computed over a smaller sample than the weeks before it.

Sorting by date and taking the top row is therefore not “the latest data”. It is the least complete data, presented with the most authority.

Every snapshot here records a settled_through date and an explicit excluded_weeks list with the reason each week was dropped. The chicken price piece excludes two weeks on sample-size grounds: one with 869 reporting premises against a typical 1,150, and one with 3,649 observations against a typical 6,000.

Both would have shown a price move. Neither move was real.