How this is sourced

The only facts on this site

This is an arithmetic tool, and arithmetic needs no sources. Three categories of fact appear here, and nothing else does.

First, molar masses, formulae and chemical identifiers, each read from a named public chemical record and printed with the record’s identifier and the date it was retrieved. Second, the printed calibration of insulin syringe barrels - units per millilitre, capacity, graduation - which is a property of a manufactured device and is checkable by anyone holding one. Third, the result of an automated query against a public drug application register, reproduced verbatim with its date.

No page here reports what any study found, what any amount was, what any compound does, or what anybody should do. Those are not omitted by oversight: a tool that converts numbers has no business asserting them.

Unknown stays unknown

A field that was not sourced reads Unknown. A field that cannot apply reads Not applicable. No value is inferred, no gap is filled with a plausible figure, and no value is borrowed from a similar compound.

The strongest version of that rule is the compound index: compounds whose molar mass was not sourced have no page at all, rather than a page with an estimated figure on it. The index says how many there are and names them.

How the arithmetic is verified

The calculation engine is a separate package with no browser code, no network access and no data in it. It carries 64 automated tests: unit tests, boundary cases, dimensional checks, property tests, and golden reference cases whose expected values were worked out by hand from the identities rather than captured from the implementation. A test whose expected value came out of the code proves only that the code agrees with itself.

Passing tests are weak evidence on their own, because a green suite is equally consistent with a thorough suite and with one that never exercises the arithmetic. So the engine is also mutation tested: a harness introduces 20 deliberate defects of kinds that have shipped in real calculators - an inverted ratio, a multiply where a divide belongs, a unit factor wrong by a thousand, a safety label that overstates what is known - and verifies that the suite turns red for every one. All 20 are currently caught.

The static reference tables on this site are computed by the same code the live calculator calls, and a test asserts cell by cell that the two agree. A printed table that contradicted the tool on the same site would be worse than no table.

Exact arithmetic, and one place where it is not possible

Values are held as exact fractions of whole numbers, not as decimals. Ordinary floating-point arithmetic cannot represent most decimals exactly, and the reconstitution identity chains four reader-supplied figures through a divide, a divide and a multiply, so the error compounds into the digit a reader would act on. Rounding happens once, at the display boundary, half away from zero.

The exception is the half-life curve. The fraction remaining is one half raised to a fractional power, which is irrational for most inputs and cannot be held exactly. Every value on that page is labelled approximate, and the exact figures at whole multiples of the half-life are given separately because those are the cases where an exact answer exists.

What is deliberately not built

There is no account, no saved vial, no history and no sharing, so nothing a reader enters is stored anywhere. There is no feature that accepts a compound and returns a quantity, and no feature that converts a published study regimen into a syringe reading. The second of those is the one worth naming: presenting a study’s reported amount next to a tool that turns an amount into units would build an implied path from "a study did X" to "draw Y", which no disclaimer fully severs. So the two are kept on separate sites, and this one links out rather than reproducing.

Corrections and the audit trail

The data this site publishes is a small, versioned file in a source repository, so every change to a figure is a dated, attributable revision rather than an edit to a database nobody can see afterwards.

A page is wrong if it does not match the record it cites, or if its arithmetic can be shown to be incorrect. Either is a defect and is corrected. An upstream record that changed after the retrieval date printed here is not an error here: the record at source is the current authority and is linked for that reason.

Automation, stated plainly

These pages are generated programmatically from that data file. The explanatory prose was written against the arithmetic it describes, and every number on the site is computed by the tested engine rather than typed in.

No qualified clinical reviewer is recorded against this site, and no credential is claimed anywhere on it. A reader deciding how much weight to give a page is entitled to know how it was produced and what was not done to it, which is why this page exists rather than a page claiming an editorial board.