Corrections

Product pages compared every item against sachets and 5 ml samples. On beauty pages the box said “Cheapest $0.25”, and an item's position (“#376 cheapest — 83% cost more”) was counted against 21 trial-size packets nobody shops for.

Fixed: Samples are out of the price distribution, so the cheapest, median and ranking now describe full-size products. Their own pages stay reachable but are no longer submitted to search engines.

The price report's CSV export gave percentages as raw fractions — a 13.2% drop was exported as 0.132, so anyone charting our data was off by a factor of a hundred.

Fixed: The export now carries the same numbers as the page.

The home page banner counted more price checks in 24 hours than there are products in the catalogue: retries and per-variant reads were counted as separate checks.

Fixed: The counter counts products checked, which is what the sentence claims.

Page descriptions promised direct-to-consumer brands, video and saved lists on pages that had none of those.

Fixed: Descriptions are generated from what the page actually holds.

Amazon prices were kept for months. The PA-API licence allows 24 hours, and a price older than that is not a price — it is a memory presented as current.

Fixed: Amazon readings older than a day are dropped, and those pages carry no price history chart at all.

Store and brand headlines advertised prices that belonged to samples: “101 picks from $0.25” pointed at a sachet, not at anything a shopper would buy.

Fixed: Headline prices are taken from full-size products only.

Deals, trending and tag pages counted every colourway as a separate product, so a single item on sale in six shades read as six discounts.

Fixed: Variants are grouped; one product counts once.

A flame marker on the first three products in a list implied momentum we never measured — nothing behind it was counted.

Fixed: Removed. Markers now appear only where a number backs them.

Collection pages said items were ranked by how often people saved them. Ranking never used saves.

Fixed: The pages describe the ordering they actually use.

Thirty products were listed under their colour name instead of the product's name.

Fixed: Titles are rebuilt from the store's product name.

One product in five reported the colour of its photo background as the colour of the product, so colour filters returned the wrong things.

Fixed: Colour is read from the item, not from the image as a whole.

The “looks similar” block showed six essentially random products — the visual match it claimed was not being computed.

Fixed: The block shows image-matched items, and says plainly that a visual match is not a match of material, size or quality.

A product page could claim a lowest observed price on the strength of two checks, which is not a record, and the affiliate disclosure ran together into “@alexearns”.

Fixed: A claimed low needs a real observation window, and the disclosure reads as a sentence.

224 products held prices in pounds or euros labelled as dollars: our server sits in the EU and stores answer European visitors in European currency.

Fixed: Price checks ask for the US storefront explicitly, and a reading in an unexpected currency is discarded rather than converted at a guessed rate.