Methodology: how we research and build

This page describes how a number gets from a primary source onto a GLP Ladder page, and how a calculator gets from a formula to a published tool. It is meant to be checkable: if we say we do something, you should be able to see the result on the page.

Primary sources only

We take facts from the original publisher of the fact, not from a summary of it.

We do not use forums, vendor pages, social media posts or other tool sites as sources for a number. Secondary articles may help us find a primary source, but they are never the citation.

How we verify a fact

1. Quote against the source. Before a number is used, we read it in the source itself and record the exact value, unit and timepoint.
2. Date it. Every source carries its publication date and the date we accessed it. Both appear in the page’s source list.
3. Keep the context attached. A weight-change figure travels with its timepoint, its arm and its analysis. A dose travels with who was given it and under what protocol.
4. Leave out what we cannot source. If we cannot tie a number to a primary source, it does not appear on the site.

Estimand labeling and cross-trial caveats

An estimand is the exact question a trial analysis answers. Two figures from the same trial can differ because they answer different questions.

• The efficacy estimand asks what happened if everyone stayed on treatment. Sponsor press releases often lead with it.
• The treatment-regimen estimand, often analyzed as intention-to-treat, asks what happened to everyone randomized, whether or not they stayed on treatment. Journals often lead with it.

We state which estimand a number uses and do not mix the two in one comparison. We also do not present a result from a selected subgroup or an extension as the general result of a trial.

Comparing numbers from different trials is not the same as a head-to-head trial. Populations, trial length, background therapy and analysis methods differ. Every comparison page on the site says so, and a gap between two figures is described as a gap between two trials.

How the calculators are tested

Each calculator is built from a stated formula and checked before it ships.

• Unit tests with published test vectors. Tests use worked examples taken from labels and trial protocols, plus hand-checked arithmetic for concentration and unit conversions.
• Formulas shown. Reconstitution and unit tools show concentration as vial milligrams divided by water volume, and units as dose divided by concentration, times 100 on a U-100 syringe.
• Defaults without scripts. Default tables render without JavaScript; scripts only add live recalculation.
• Model limits stated. The plotter models a relative amount from a simple one-compartment model, not a blood level. Its assumptions, half-lives and limits are in the model card.

Calculator results depend on the numbers you enter. We test the arithmetic, not the inputs.

Update cadence

We review pages when something that affects them changes: a trial readout, a publication, a label revision, an approval, or a manufacturer program change. The updated date on a page changes only when the content changes. It is never stamped automatically, and we do not change dates to look fresh.

Corrections

When a reader or an internal check finds an error, we compare the page with the primary source, fix the page, and change the updated date. Where a correction changes the meaning of a page, we say what changed on that page. To report a problem, use the contact page. Our wider commitments are in the editorial policy, and the full source list is on the sources page.

Limits of this method

Primary sources can be wrong, change after we read them, or be revised. A press release is not peer review. Investigational compounds have results that are not final. We label these limits on the page where they apply and prefer saying “not reported” to filling a gap with a guess.