How we build this library
Every number on this site can be traced back to a sentence in a specific document, on a specific page. This page explains how that chain is built, who checks it, and where we use automation.
The chain
Indoor environments are described here as a chain: a source releases a factor, the factor causes an outcome, an intervention reduces it, a standard sets a limit, and a method measures it. Every page enters that chain at one of those six points, and links to the neighbouring ones.
Documents
The library holds primary sources — peer-reviewed studies, standards, agency reports and guidance. Each document is filed once and receives a permanent number. We do not republish the documents themselves: we index them, cite them, and link to the publisher's own copy or its DOI.
Findings
A Finding is one result taken from one document. It records the verbatim sentence the result comes from, the page it appears on, what was measured, in what population or setting, and the numbers with their units and uncertainty.
A Finding is taken whole or not at all. If a number in the original sentence cannot be carried across into a field of its own, the Finding is rejected rather than summarised — a rejected row keeps its number and the reason for rejection, so the decision stays on the record and cannot quietly reappear later.
Observations
An Observation is a result from our own measurements rather than from someone else's document. It is recorded the same way and is subject to the same checks, and it says which dataset and which method it came from.
Claims
A Claim is a synthesis: what several Findings and Observations, taken together, support. Each Claim carries a grade from A to D for the strength of the evidence behind it, and the language of the claim has to match that grade — a grade D claim may not be written as if it were settled. Every Claim also has to record the evidence that points the other way. A Claim that cites no counter-evidence is rejected.
When new evidence supersedes a Claim, the old one is not deleted. It is marked as superseded, keeps its address, and points to the Claim that replaced it.
Glossary entries, articles and data tables
A glossary entry is published only once there is verified evidence behind it. We deliberately do not publish the full vocabulary as stub pages: a term with nothing but a definition adds nothing that is not already available elsewhere.
An article walks one path through the chain and has to name the evidence it rests on. A data table — for example exposure limits by jurisdiction — is generated from the underlying records, carries a version number, and states when each value was last checked against its source.
Where automation is used, and where it is not
Reading a document and proposing which sentences contain results is assisted by software, including a language model. That step only proposes; it does not publish.
Nothing reaches this site from an unverified row. Automated checks run before every release and stop it outright if a page cites an identifier that does not exist, cites evidence that has been superseded or retracted, links to a page that is not there, or writes a quantity, unit or chemical formula in a form that does not follow ISO 80000-1, the SI brochure and IUPAC conventions.
Where a person checks
Every Finding is checked by a person before it appears on the Site. Nothing is published from an unverified record, and a record which cannot be verified is rejected rather than softened.
We are introducing clinical review for Articles which enter the chain through a health outcome. Until that reviewer is in place, an Article of that kind carries a note at the top of the page saying plainly that it has not been reviewed by a clinician. We would rather say so on the page than let a reader assume a review which has not happened.
Corrections
If a number, a citation or a statement is wrong, please tell us at coridum@coridum.com and say which page it is on. We check it against the source it came from and correct it. The identifier of every record stays the same after a correction, so the correction is visible rather than hidden.
What we do not do
- We do not give medical advice, and nothing here is a diagnosis.
- We do not restate a limit without naming the document it comes from. Where a value has not been verified against a primary source, the table says so.
- We do not republish the full text of the sources we cite.
- We do not recycle identifiers. A number, once issued, belongs to that record permanently.