How Coridum data are collected, anonymised and published

What we publish

Three kinds of data, each on its own page with a stable number (D-0001, D-0002 …), a version, a date, a downloadable CSV and a citation block:

• Reference tables compiled from primary documents — exposure limits by jurisdiction, evidence summaries. Every row names its source document.
• Aggregated measurements from Coridum devices — for example the distribution of carbon dioxide in offices by month, or the share of monitored rooms above a formaldehyde limit.
• Aggregated user-reported data — answers people give about their spaces, activities and symptoms, combined with measurements. Not yet published; the rules below apply from the first release.

Where the data come from

Devices. Coridum monitors report readings at fixed intervals with a device identifier, a timestamp and the room type the owner assigned. Sensor types, calibration and known limitations are described on each factor's page under “How it is measured”.

User inputs. Questionnaires about the space (size, materials, ventilation, heating, activities), the occupants (age group, sensitivity) and outcomes (symptoms, diagnoses reported by the user). Health-related answers are collected only with explicit, separate consent for research use and can be withdrawn at any time.

Literature. Peer-reviewed studies curated in the Coridum Knowledge Base, cited by number and DOI. Extracted values are recorded with the sentence they come from, the unit, the population and the study design.

Anonymisation and aggregation

• No published table contains data about an identifiable person, household, device or building. Identifiers are removed before analysis; the analysis dataset is pseudonymised, the published dataset is anonymised.
• Minimum cell size. A value is published only if it is computed from at least 20 distinct sources (devices or respondents). Cells below the threshold are shown as “n < 20” and suppressed.
• Geography is published at the level of country or region, never postcode or address. Time is published as month or quarter, never day.
• Rare combinations (for example a specific diagnosis in a specific building type in a small country) are merged into broader categories until the cell-size rule is met.
• We do not publish individual time series, free-text answers or photographs.

Consent and legal basis

Measurements are processed under the contract with the device owner. Answers about health are special-category personal data under the General Data Protection Regulation; they are processed for research and statistical purposes only with the person's explicit consent, kept to the minimum needed, and never used to make decisions about that person. Participants can see what they have provided, correct it and withdraw; withdrawal removes their data from future aggregates. The applicable notice is the Privacy Policy.

Quality control

• Units. Every value carries its unit; conversions (ppm ↔ µg/m³, ppb ↔ µg/m³) use the stated temperature and pressure. Unit errors are the most common defect we find in published studies, so they are checked twice.
• Notation. Quantities, units, numbers and chemical formulae are written according to the SI Brochure (International Bureau of Weights and Measures, 9th edition), the IUPAC Green Book (Quantities, Units and Symbols in Physical Chemistry, 3rd edition, CKB 1705) and the International Vocabulary of Metrology (JCGM 200:2012, CKB 1704); in downloadable CSV files units are coded according to UCUM version 2.2 (CKB 1708). Every page is checked automatically against these rules before it is published; a page that fails is not published. Where a source document states a limit in ppm we keep ppm, noting that 1 ppm = 1 µmol/mol for gases.
• Plausibility. Device readings outside the sensor's physical range, frozen values and readings during known sensor faults are excluded and the exclusion rate is reported.
• Contradictions. Where studies disagree, the evidence table shows both, with the study design and confidence interval; we do not average away disagreement.
• Statistical significance is not the same as relevance. Effect sizes with very wide intervals are reported with the interval, and flagged.
• A sample of extracted values is re-checked by a person against the source document before each release.

Versioning, corrections and citation

Tables are never edited silently. A change of values or method increases the version number and is described in the table's “Version history”. Earlier versions remain downloadable. Errors reported to us are corrected in a new version with the reporter credited if they wish. Cite a table by its stable number, version and date, as shown in its “Cite this page” block. Data tables are licensed CC BY 4.0.

What we do not do

• We do not sell or share individual-level data with third parties.
• We do not link measurements or answers to named people, addresses or employers.
• We do not publish a result because it is interesting if it fails the cell-size or plausibility rules.
• We do not accept payment to rank an intervention or a product in an evidence table.

Version history

• 0.1 (5 September 2026) — draft published ahead of any user-contributed dataset: scope, sources, anonymisation rules (minimum cell size 20, geography and time granularity), consent basis, quality control, versioning. To be finalised before the first release.