How the figures are made
Three sources and no fourth. Nothing here is quoted, borrowed, averaged or estimated.
Educational arithmetic only. Not medical advice and not a dosing protocol.
The three places a figure here can come from
The reader entered it. Every worksheet takes its quantities from its own form, which is why none of them can originate an amount, an interval or a half-life.
It is printed on the barrel. The five calibrations, their capacities and their graduation intervals are properties of manufactured objects, visible to anybody holding one, and they are the only facts about the physical world this site asserts.
It was computed at build time by the same functions the worksheets call. Every cell in every table under by half-life, by interval and background is generated that way, so a table cannot disagree with the worksheet it should match.
The parameter values, and why they are not data
Ten half-life values and six intervals appear in the static page families. They were chosen to span the ranges this arithmetic is useful over - a couple of hours to a fortnight, and eight hours to a week - spaced roughly logarithmically because that is how published figures in this area are distributed.
None of them is attributed to any substance and none is presented as typical. They are points at which a function was evaluated so that its shape is visible without filling in a form, which is a different thing from a table of facts.
How the arithmetic is checked
The calculation engine is dependency-free, has no network access and holds no data. The exact half holds quantities as fractions of whole numbers and rounds once, at the point of display. The approximate half is computed in double precision and every value it returns carries a flag saying so, which is what the labels on the figures are generated from.
It is covered by a unit test suite and by mutation testing: a harness introduces deliberate defects one at a time - an inverted ratio, a draw count rounded up instead of down, an exponent the wrong way round, a safety label that overstates what is known - and fails if the tests do not notice.
A separate suite runs over the built site and checks what unit tests cannot: that every page carries the disclosure sentences, that no internal link points at a page that was not built, that no amount is stated for anyone, and that no page here links to any other site in the same network.
None of that is clinical review. It is evidence that the software does what it says, which is a smaller and different claim.
The date in the footer
The date at the foot of every page is the date a committed data snapshot was last matched against its sources. It is a real event recorded in a file in the repository, not today's date printed to look current.
A stamp that moves on every build is a build timestamp dressed as a review date, and it tells a reader nothing. This one does not move unless something was actually checked.