Editorial and verification policy

What every tool on ScienceQuest is built from, how it is checked, and what happens when it turns out to be wrong. This describes the process as it actually runs, including where it stops.

The short version

Every calculator and every simulator names the source of its equation, directly under the equation. 173 of the 218 tools on the site do; the rest are reference tables, visualisers and practice sets with no single equation to cite.
Physical constants follow the 2019 SI and CODATA 2022. Atomic weights are the IUPAC CIAAW 2024 abridged values.
Calculations are held to known values by automated tests, and a correction comes with a test that keeps the error from returning.
The clinical calculators are study tools. Each one says so beside the tool and names the equation it uses.
Every tool page has a Report an error link. The corrections section below says what happens next.

Where the equations come from

An equation on a page is a claim, so every one carries its provenance. The citation sits under the equation rather than in the prose, because the question it answers, which equation is this, is asked while looking at it. For several quantities that question matters: there are four common QT corrections, four body surface area formulas and three estimates of kidney function, and they give different numbers for the same patient.

Citations take one of two forms, chosen by what is true of the equation. An eponymous or empirical result names its authors and year, as in “Bazett (1920), Fridericia (1920), Hodges (1983), Sagie (1992)”. A definition or a standard result names itself or the syllabus that states it, as in “Definition of density” or “Definition of magnification, AQA GCSE Biology specification 8461, section 4.1.1.5”. Where an author cannot be established, the descriptive form is used on purpose rather than a guessed name.

No citation carries a link yet. A reference such as “Cockcroft and Gault (1976)” can be looked up as written, and a link that resolved to the wrong paper would look like evidence while being none. Links will be added one at a time, where the canonical source has been confirmed by hand.

Where the numbers come from

Physical constants are the 2019 SI definitions, which fixed the Planck constant, the elementary charge, the Boltzmann constant and the Avogadro constant as exact values. So the gas constant used here is 8.31446261815324 J/(mol·K), the exact product of two exact quantities, rather than a rounded figure from a textbook. Constants that are still measured, such as the gravitational constant and the vacuum permittivity, are the CODATA 2022 values, checked against NIST’s listing and printed with their uncertainty.

Atomic weights for the 118 elements come from the IUPAC Commission on Isotopic Abundances and Atomic Weights, using the 2024 abridged standard values. A standard atomic weight is defined only for an element with a characteristic isotopic composition on Earth. Technetium, promethium and every element from polonium upward, except thorium, protactinium and uranium, have none, so they have no standard atomic weight at all. For those the mass number of the longest-lived isotope, or of one of them where CIAAW lists several, is used, and the page says so rather than presenting a nominal figure as a measured one. Isotopic compositions are NIST’s, credited on the element pages.

Derived constants are derived in the code, not pasted in. The centrifuge conversion factor, for instance, is computed as (2π/60)² / (100·g) rather than written as the rounded 1.118 × 10⁻⁵ that circulates in protocols, so it carries full precision and its origin can be audited. Other reference data is compiled from named sources: Schott’s data sheets for optical glass, Kaye and Laby’s tables for photoelectric work functions, and the figures the CRC Handbook, Atkins and Petrucci share for standard reduction potentials, which are quoted to two decimal places because that is the precision at which those sources agree.

How the tools are checked

The calculations are covered by thousands of automated tests. They are written to catch the errors that a quick look at a page would miss, and they check in several different ways, because each catches something the others do not.

Known values. Results are compared with figures that do not come from this code: textbook landmarks, published reference values and answers worked out independently. A sodium of 140 mmol/L with a glucose of 90 mg/dL and a BUN of 14 mg/dL must give a calculated osmolality of 290 mOsm/kg. The kidney function equations must reproduce the reference values they are routinely demonstrated with.
Structural checks. Identities that must hold whatever the arithmetic, so a transposed coefficient fails even where one point value happens to pass. At a heart rate of exactly 60 beats per minute, every QT correction must return the QT unchanged.
Round trips. A calculator that can be solved for any of its variables is solved forwards and then backwards, and must return the value it started from.
Second implementations. Some results are checked against code that shares nothing with the site’s own: the expression parser against the mathjs library, significant figure rounding against decimal.js, and primer melting temperatures against an independent script cross-checked with primer3-py and Biopython.
Edge cases. Inputs that have no answer are asserted explicitly, so total internal reflection past the critical angle or a division by zero returns a stated reason rather than a number.

The worked examples, 130 of them across 51 calculators, are not typed in. Each one declares its inputs, and the working on the page is produced by running the calculator itself. Its answer is also pinned in a test, so a change that would alter a published answer fails the tests instead of quietly republishing a different number. The practice sets work the same way: every answer is computed by the calculator the question came from, so the two cannot disagree.

Each of the 56 simulators carries a Model and assumptions box stating the numerical method, what the model assumes and, where it has one, the point at which it stops holding. Where that box gives a numerical error, the figure was measured when the site was built, by running the model at its step, at half of it and at a quarter of it, so it comes from the solver that is actually shipped rather than from a claim written beside it.

Every change to the code the site is built from is also run through automated checks: type checking, the full test suite, a build, and audits of the structured data, typography, security policy and accessibility of the built pages.

Who checks the tools

ScienceQuest is an independent project, and the checks described here are its own. They are automated and repeatable, and where this page says a result is checked independently it means checked by a method or a program that shares no code with the tool. No outside body reviews the tools before they are published, and no endorsement by one is claimed anywhere on the site. That is also why there are no star ratings or review counts in the structured data: there are no reviews to count.

Clinical calculators

Some tools produce numbers that look exactly like the ones a clinician acts on. They are written for learning, and each carries the same standing caution beside the tool: it is not clinical decision support, reference ranges are typical rather than your laboratory’s, units must be checked before a number is trusted, and, where the formula was derived in a particular population, which population that was. The caution also names the equation the page uses, and each clinical calculator names its source under the equation like every other tool. The same caution appears on the practice set that draws questions from them, and a shorter form travels with each of them when it is embedded on another site.

The tools that carry it:

Dates on the pages

A tool page’s “Last updated” date, its structured data and its sitemap entry all come from the page’s own history in version control, from one lookup, so they cannot disagree. Where that history cannot vouch for a date, the page is left undated rather than given the date it happened to be built.

Corrections

If a result looks wrong, it might be. Constants are revised, formulas get misapplied, and a tool that shows its working makes its own mistakes easier to spot. Reports of errors are the most useful thing anybody can send.

Every tool page has a Report an error link beneath the citation block. It opens an email to hello@sciencequest.in with the page address already in it. Add the values you entered with their units, the answer the tool gave, and the answer you expected with where that figure comes from. Where the tool has a share link, send that too, because it encodes every input. The contact page has the same address for anything else.

When a report turns out to be right, the tool is fixed and a test is added that holds the corrected behaviour, so the same error cannot come back unnoticed. Two corrections show what that looks like:

The pH calculator’s weak acid solver left out water’s own ions, so a very dilute weak acid came out basic: acetic acid at 10⁻⁸ M gave pH 8.00 where the right answer is 6.98. It now solves the full charge balance, and a test compares it with values solved independently of the calculator.
The anatomy explorers credited their meshes under the licence BodyParts3D carried before its publisher relicensed it to CC BY 4.0 in 2025. The credit was corrected on every page and in every model file, and a test now checks the credit the pages print.

A corrected page’s “Last updated” date moves with the fix. Corrections are not listed separately.

Advertising

The site shows advertising from Google so that the tools can stay free, as the privacy notice explains. Embedded tools carry no advertising at all, so an embed is safe to put in front of a class.

Images

Images the site generates are free to reuse under CC BY 4.0, with credit and a link. Images drawn from the 3D anatomy models must credit BodyParts3D as well. The image licence sets out what is covered and exactly what to write.