Naming the Hazard Is the Easy Part

Every page on this site can tell you what a compound is, where it comes from, whether it has been measured in human blood, and whether a regulator has flagged it. That is the easy half. The half readers actually want is the next sentence: so what do I do about it?

We can answer that for 33 of the 251 compounds in the Exposure Ledger. Not 33 because the other 218 are safe. Thirty-three because that is where we could find a specific action, with a measured effect, attached to a source we could name.

This piece is about why that number is so small, and about which part of the problem is ours to fix.

What we can and cannot answer

The intervention data behind this site is fifteen curated actions — things like reverse osmosis, a mattress encasement, a gas-phase carbon filter — each tagged with the compounds it affects, the size of the effect, and where that claim comes from. Inverted across the atlas, those fifteen actions reach 33 compounds. Inferred

Set that against the compounds a regulator has already flagged. Eighty-nine of our 251 are Proposition 65 listed. Fourteen of the eighty-nine have an intervention we can source: arsenic, benzene, formaldehyde, lead, toluene, chloroform, carbon monoxide, four phthalates, two PFAS, and one flame retardant. The remaining seventy-five are compounds we can tell you are hazardous and cannot tell you how to avoid.

The same gap shows up from the other direction. Seventy-three compounds in the ledger have been measured in the general population through national biomonitoring. Fifty-three of those have no intervention attached.

The part that is our fault

Some of this is simply work we have not done. Fifteen curated actions is a small number, and adding more is a matter of effort rather than discovery. Anyone reading this should treat the coverage figure as a description of a young dataset, not a survey of what humanity knows. We would be overstating our own findings if we let that ambiguity stand, and the number is published precisely so it can be checked.

But the shape of what we have collected is more interesting than the size, and it was not a choice. Of the fifteen actions, seven are water filters. Two are dust control, two are air filtration, three are behavioural, one is about body burden.

Water is over-represented because water is the one place where somebody else already did the hard part. A filter can be tested against NSF/ANSI 58, a standard that specifies what reduction has to be demonstrated and how. That gives a number to cite. Spec

There is no equivalent for a mattress, a carpet, or a bedroom. No standard says what a dust-control routine has to achieve. So the honest entries in those categories are hedged, and they read like it: a mattress encasement is logged as partial on the strength of one Canadian cohort finding lower DEHP in settled dust; a gas-phase carbon filter is logged as partial and inferred, because it adsorbs some VOCs and then saturates.

Two of the fifteen entries are not remedies at all. A standard carbon pitcher is logged as doing nothing for PFAS. Boiling water is logged as actively worse, because it concentrates what it cannot remove. Those stay in the dataset deliberately. An intervention list that only contains things that work is a shopping list, and the two entries most likely to save a reader money are the two that say this does not do what you think.

The part that is not

The rest of the gap is not ours, and it is worth being precise about how confident we are.

Search the literature for the health effects of a flame retardant and you get thousands of papers. Search it for a tested way to reduce exposure to that same flame retardant in a home and you get dozens. Run the same comparison for phthalates and the ratio narrows but does not reverse. We are reporting the direction of that difference, not a ratio, because keyword counts are a crude instrument and a differently-worded query returns a different number. The direction, though, is not subtle: when we searched for flame-retardant intervention studies, most of what came back was mechanism and toxicology research, not trials of anything anyone could do. Inferred

This is not an accusation of laziness. Hazard research is publishable, fundable, and often mandatory before a chemical can be regulated. Intervention research is slower, needs human participants in real homes, produces messier results, and rarely lands in a high-impact journal. The incentives point one way.

The consequence reaches further than advice. A 2026 study of pregnant women in Atlanta measured 113 analytes across 12 chemical groups and opened by noting that only a fraction of chemicals used in commerce in the United States (US) are biomonitored nationally and have undergone comprehensive safety review, closing with a call for national surveillance to cover emerging exposures1. If a compound is not measured in people, no intervention study can show that reducing it changed anything. The remedy gap sits downstream of a measurement gap. Peer-reviewed

What listing a chemical actually does

It is worth being clear about what those eighty-nine listings are, because the word “listed” does a lot of quiet work.

Proposition 65 requires a business to give a clear and reasonable warning before knowingly exposing someone to a listed chemical2. That is the mechanism. It obliges disclosure, not removal, and it says nothing about what a person should do once warned. Regulatory

This is not a criticism of the statute, which does the job it was written to do and does it in public. But a warning transfers a decision to the reader without transferring the information needed to make it. That is the space this site sits in, and the fourteen-of-eighty-nine figure is a fair measure of how much of it we have managed to fill.

What would close it

Three things, in the order they would matter.

Measurement first. National biomonitoring is the bottleneck the Atlanta authors name, and it is upstream of everything else. A compound nobody measures in people cannot have an intervention proven against it.

A standard for soft furnishings. Water interventions are well evidenced because NSF/ANSI 58 exists to test against. Nothing plays that role for a mattress or a bedroom. Until something does, advice in this category will keep resting on single cohort findings and inference, and it will keep being tagged that way here.

More curation, honestly labelled. The part we own. We will keep adding interventions, including ones that turn out not to work, and the coverage figure will keep being published whether or not it flatters us.

Until then the useful thing we can offer is not a longer list. It is knowing which of the two situations you are in: a compound where somebody has actually measured what helps, or one where the honest answer is that nobody has checked. Both appear on the compound pages. The second is more common, and pretending otherwise would be the one failure this publication could not come back from.