The number that is not an offer

Tomas Lindqvist writes about subscription pricing and unit economics. Updated September 2026.

Every figure in this corpus was read by a person and given a kind: a standing rate, a “from” price, a first-month price, a prepay total, a one-off purchase. The kind is what makes a figure comparable to another figure. On 57 of the 312 published NAD+ prices — from 26 of the 106 sellers on this roster — no kind could be decided, because the page does not say.

These are the most misleading numbers in the data. A reader meets a dollar sign on a seller’s NAD+ page and reads it as what the thing costs. On these 57 that reading is wrong, and on 9 of them it is not even a NAD+ price.

Where these came from, and what nine of them are not

This site publishes three lists of what it could not do. The refusal ledger holds 55 figures that state a quantity and still cannot be divided. The dose-line ledger holds the sellers whose quantity is a volume, a ceiling or a schedule. This is the third and the earliest: figures that never reach either list, because before you can ask what a price buys you have to be able to say what the price is.

The 11 reasons a figure has no kind

Each of the 57 was ruled on individually, and each names the clause it fell to. The clauses are ours, written for a reader; the words being ruled on are the sellers’. Where a figure had more than one thing wrong with it, the first clause that applies is the one recorded, so two people counting by hand get the same answer.

The two largest clauses are the same failure at different sizes: a page printing more than one number for one product and labelling none of them. Between them they account for 23 of the 57. A human reader usually assumes the higher number is struck through and the lower one is current. Sometimes it is. The pages that say so are not in this list — a figure the page itself labels as the old price is in the anchor clause instead, which is a different and much more honest thing for a seller to do.

All 57, in the sellers' own words

Every one of the 57, grouped by clause, with the seller’s own sentence for the figure. The quotation is the seller’s, verbatim, and it contains the figure it is offered as evidence for — asserted for all 57 before this page renders. The ruling beside it is ours.

Why a parser cannot fix this

The obvious objection is that most of this could be automated. Take the lower of two adjacent numbers; ignore anything labelled “regular price”; skip lines mentioning a second drug. That would produce a figure for nearly all 57 of these, and it would be wrong often enough to poison everything downstream — without ever failing, because a number is a number and arithmetic on it always succeeds.

Take the two-number rule. On one seller here the pair is bmiMD’s block of 12 figures, where taking the lower of each pair gives a coherent-looking set of prices the page never offers. On another, the same page is served from two URLs with different numbers in the block. On a third, the identical pair of numbers appears on the sermorelin and glutathione tiles too, because it is template copy rather than a quote about NAD+ at all. A rule that matches one shape finds one shape, and a roster is exactly where that bites.

So each of these was read and ruled on by hand, and the ruling is recorded per figure rather than derived. The counts on this page are computed from those rulings; the rulings themselves are not computed from anything.