The Embr Exposure Ledger

What the ledger is, and why the join matters

Take a chemical you might worry about at home. Separate public datasets each answer one question about it. The EPA's drinking-water monitoring says whether it shows up in tap water. The CDC's biomonitoring program says whether it shows up in people. Proposition 65 and the EU's chemical law say whether either has flagged it. EPA's screening levels say at what concentration it would matter. The EU's Safety Gate says whether a regulator has pulled a product off a shelf over it. Each dataset lives on its own, in its own format, keyed its own way.

The Embr Exposure Ledger joins them onto a single compound record, matched CAS number first and name second against the Embr Bedroom Chemistry Atlas. A chemical that appears in three of the datasets becomes a three-spoke record. Where a dataset has nothing for a compound, that spoke is marked "no data" rather than dropped or guessed, because an honest gap is more useful than a filled-in guess.

Why is a join worth more than the four datasets it draws from? Because the raw facts are already everywhere, and a large language model has most of them. A join is unique if any single input is unique. Two of these inputs are Embr's own: the curated Atlas that the datasets are matched against, and the Verdict Layer that reads a plain, evidence-tiered verdict across the spokes. That is the part that exists nowhere else, and it is what turns ten commodity datasets into one asset.

The spokes

Each spoke answers one exposure question and carries its source with it, so any figure can be traced back to the agency that published it.

The most-joined compounds

Of the 268 compounds in the ledger, most appear in only one or two datasets. The ones below are the exceptions: the deepest records in the ledger, up to 9 of ten exposure questions answered at once. These are where the join earns its keep. Each links to its full, cited page in the Atlas.

Coverage as of v1.5 (built 2026-08-06): 268 compounds joined across 10 active spokes — 5 nine-spoke, 16 eight-spoke, 17 seven-spoke, 29 six-spoke, 28 five-spoke, 30 four-spoke, 50 three-spoke, 43 two-spoke, and 50 that currently appear in a single dataset. A new spoke does not add pages; it adds a column, so a two-spoke compound can become three-spoke as the datasets grow.

A second worked example: US production volume against US body burden across 43 chemicals.

How to cite this dataset

The ledger is free to use, including commercially. If you publish a figure from it, cite the version — the numbers change as agencies release new data, and a version makes your citation reproducible.

Eyjolfson, K. (2026). Embr Exposure Ledger (Version 1.5)
[Data set]. Embr Sleep. https://embrsleep.com/exposure-ledger
Frozen copy of this version: https://embrsleep.com/data/ledger-v1.5.json
Built 2026-08-06 (build ee55d2f5). Accessed 2026-08-06.

Every version is frozen at its own URL, so a citation to v1.5 keeps resolving to the numbers you actually quoted after the live file moves on.

Built from this

The Dose Gap — the annual artefact derived from this ledger: the compounds already measured in people for which no inhalation reference value has been published. It ranks the absence of dose-response rather than hazard, and says on the page why that distinction matters.

ClearSupply — the same join, pointed at a product instead of a compound. The regulatory spoke carries REACH SVHC listings, CLP Annex VI harmonised classifications and the POPs Regulation Annex I prohibitions, which are three of the four legs of the EU's ESPR definition of a substance of concern. From 2027 those have to be declared, by name and concentration, in a Digital Product Passport for textiles sold into the EU.

Licence

The compilation is released under CC BY 4.0. That covers what we made: the record linkage, the resolver decisions, the Verdict Layer, and the provenance stamps on every value.

It does not cover the underlying government data, and it does not need to. EPA, CDC, Statistics Canada, ECHA and California OEHHA figures are public records; you need no permission from us to use them. Each spoke carries its own source and link so you can cite the agency directly, which for a primary figure is the better citation anyway.

What we ask in return for the compilation is a link back, and that you do not present a spoke value as something the agency did not say. Every caveat in the section below travels with the number.

Changelog

What changed between versions, so a citation to an older version still means something.

• v1.5 2026-08-02 — Widened the environment spoke from one UCMR cycle to all five (2001–2025), doubling it from 17 compounds to 34 and retiring it as the weakest spoke. It had been PFAS-only because UCMR 5 is PFAS-only; cycles 1–4 add flame retardants, chlorinated solvents, metals and pesticides — among them BDE-47, BDE-99, BDE-100, hexabromobiphenyl, NDMA, 1,4-dioxane, 1,3-butadiene, chromium, cobalt, manganese, perchlorate and haloacetic acids. A compound measured in more than one cycle now keeps every cycle under `cycles`, each with its own reporting level, rather than being collapsed into a single number that would imply one current measurement. Every result carries the minimum reporting level, and a zero-detection result is labelled `detection_limited`: UCMR 2 found no BDE-47 in 4,140 systems at a reporting level roughly a thousand times above where PBDEs occur, which measures the method and not the water. 257 to 268 compounds; ten active spokes unchanged.
• v1.4 2026-07-25 — Added the enforcement spoke — has a regulator ever acted? — pairing EU Safety Gate (RAPEX) alerts with US CPSC recalls (105 compounds); this is the first spoke that records action taken rather than presence measured. Added two regulatory sub-sources: EU CLP Annex VI binding harmonised classifications (127 compounds, 79 CMR) and the EU POPs Regulation Annex I, the one instrument in the ledger that is an outright ban (22 compounds). 254 to 257 compounds; nine active spokes to ten.
• v1.3 2026-07-20 — Added REACH Annex XVII restrictions (44 compounds) and Annex XIV authorisation with sunset dates (9 compounds) to the regulatory spoke. The lookup can now distinguish identified, restricted and prohibited-unless-authorised.
• v1.2 2026-07-20 — Added the commerce spoke (EPA TSCA CDR production volumes, 168 compounds). Expanded the body spoke with Statistics Canada CHMS and a full harvest of CDC's exposure-report tables (US coverage 21 to 80 compounds). Added EU REACH SVHC listings to the regulatory spoke (54 compounds). 124 to 244 compounds; four active spokes to six.
• v1.1 2026-07-13 — Per-value provenance: every figure now states what statistic it is, the period it covers, and when it was captured. Ledger stamps its own version, build date and git SHA.
• v1.0 2026-07-06 — First release. Four spokes: EPA UCMR 5 drinking water, CDC NHANES biomonitoring, California Proposition 65, and curated interventions.

Method, and what the ledger does not claim

The datasets are copied verbatim from their public sources and joined by record linkage: CAS number first, chemical name second. A match that fails returns nothing and is surfaced honestly; it is never patched with a guess. Nothing in the ledger is invented — not a value, not a CAS, not a verdict, not a listing.

Read these caveats before you quote a figure

• Occurrence is not risk. A chemical being detected in water, or measurable in blood, says it is present. It does not by itself say it is harming anyone at that level. Dose, duration, and who is exposed all matter.
• A null spoke means "no data," not "safe." When a dataset has nothing for a compound, that spoke is blank. That is a gap in what has been measured, not a clean bill of health.
• A Proposition 65 listing is a legal fact, not a risk verdict. It records that California has listed the chemical for cancer or reproductive harm under state law. It is not a statement about the exposure you personally face.
• The evidence tiers travel with each claim. An intervention marked Peer-reviewed rests on published journal literature; one marked Regulatory rests on an agency or standards body. The tier is part of the data, so you can weigh it. The full definitions live on the methodology page.