The cheapest listing in our index was not a listing at all

The vendor directory now prints the spread across the whole index — the cheapest per-milligram figure we hold and the dearest. When that went live the low end read $0.002. That is not a plausible price for anything on this site, so we went looking for the row behind it.

It was a bottle of MK-677 capsules, titled 'MK677 Capsules 12.5mg'. Our parser reads a count out of a title when the title states one: '60 capsules', '10 vials'. Against that title it found the digits 677, immediately followed by the word 'Capsules', and read a count of six hundred and seventy-seven. Twelve and a half milligrams times six hundred and seventy-seven is 8,462 mg, and $169.99 divided by that is two cents a milligram.

The digits were the compound's name. MK-677 is what the thing is called.

What makes this worth writing up rather than quietly fixing is that the guard for it already existed. Capsules with no stated count are supposed to be held back for a person to read, precisely because a bottle's price against one capsule's strength is meaningless. That check never fired, because the count did not look unstated — it looked stated, confidently, and wrong.

A second listing showed the same class of problem from the other direction. One vendor sells the same blend twice, once described as '(10 vials)' and once as '(10 btls.)'. We counted the first and not the second, so the index carried both readings of one product: $33.50 a milligram and $335.00 a milligram. The abbreviation was simply not in the list of words that mean a container.

Both are fixed, and both now have tests naming the listing that produced them. The reason for saying so publicly is the one that governs the rest of this site: a per-milligram index is worth exactly as much as its worst silently-wrong row, and a number with two decimal places in a column of other numbers gives a reader no way to tell the difference. If we only corrected these quietly, the honest position — that a price here is a reading of a vendor's own words and readings can be wrong — would be something we assert rather than something we demonstrate.

One practical note for anyone reading prices in the meantime. Existing rows do not re-read themselves when the parser improves; they are corrected by a separate pass over the catalogue. Until that has run, the affected rows carry their old figures.