What it says about substances on the market today
Two lists over one set: the 260 substances that hold an EU approval today and have not been withdrawn. One ranks them for withdrawal risk, the other screens them for PFAS formation, and fourteen substances appear on both.
The screen covers 242 of the 260. The remaining 18 have no molecular formula in PubChem to read, so they are neither flagged nor cleared.
What it says next
The benchmark grades this model against withdrawals that already happened. This list is the opposite, and carries a different status: these are substances it ranks as concerning now, so none of them has been graded yet. The decisions they anticipate have not been made. What stops that being unfalsifiable is that each entry carries the expiry date of its own EU approval, which is the day the Commission has to rule on it. Every line here can be marked right or wrong on a date already in the calendar.
The ranking is a modelled three-year hazard among substances currently approved and not yet withdrawn, not the “will this ever be withdrawn” score the first benchmark used. That older target was answered largely by approval age, so it mostly returned old approvals. Under this one, approval age is the baseline rate and the evidence does the discriminating: the two substances that most obviously did not belong near the top, a fatty-acid soap and acetic acid, fall from 5th and 62nd to 27th and 192nd.
When this gets marked
Every EU approval runs to a fixed expiry date. On that date the Commission has to rule on the substance: renew it, restrict it, or let the approval lapse. That ruling is what marks the entry right or wrong, and 64 of them fall due by the end of 2027.
An expiry passing is not automatically a verdict: the Commission often extends an approval by a short procedural step while an assessment finishes. Those are recorded as still open rather than counted either way.
Which crops they are approved for
32 of them are in plant protection products currently approved in Sweden. Each bar is the share of that crop's approved products carrying at least one.
This is the catalogue, not the fields: what a grower could legally buy, not what was sprayed or on what area.
The grey line sits at 22%, which is what the figure becomes when every approved Swedish product is counted at once instead of crop by crop. It is the level to read each bar against. They run from 13% for oilseed rape to 31% for strawberry, a narrow band around that line, so these substances are spread across Swedish agriculture rather than concentrated in any one crop.
How much is actually sold
24 of them have recorded Swedish sales. Tonnage alone is unreadable, because sales are skewed across three orders of magnitude, so each carries its rank among every plant protection active sold here in 2024.
Click a substance for its deadline, its full crop list and the Commission's own record.
National tonnage, not tonnage per crop, because nobody publishes the second in a form that joins to a substance ranking. Sweden's per-crop survey reports by pesticide type rather than by substance and was last run for 2021; Eurostat states it has never been able to publish comparable EU use statistics at all. A fix was legislated in 2025, with publication from 2030. Until then the country is the finest unit available.
No product or brand is named, and none will be. At the cutoffs old enough to have resolved, a little under half of a top-100 band was never actioned, so naming commercial products against it would put specific companies on a list carrying that much error. Crop is the finest granularity published. Generated 2026-07-26.
The cohort test
On 20 November 2025 Kemikalieinspektionen opened a reevaluation of 6 plant protection substances that break down into the PFAS compound TFA and reach groundwater. Those 6 were chosen by a regulator, on a dated public decision, which is what makes them a test rather than an anecdote: asked of one substance, any ranking puts it somewhere and the answer can be read either way.
The shaded band is the published top 100.
0 of 6 reach that band, where a random draw would have placed 2.3. The median sits at the 65th percentile, so this is not a near miss. The withdrawal model does not find them.
The structural screen, over the same population and answering the other question, holds all 6 of them. Its shortlist is below, with this cohort held out as its check.
The screen that does find it
TFA comes from trifluoromethyl groups. That is chemistry, not regulation, so it can be read off a molecular formula. A formula is public, free from PubChem, and the same at every cutoff, which means this can be run over the whole approved population without a model and without any way to leak.
What the cohort check does and does not show
Kemikalieinspektionen chose its 6 because they form TFA, and TFA comes from trifluoromethyl groups. A rule that reads those groups was never going to miss them. Finding all 6 shows the rule is implemented correctly and applied to the right population. It is not evidence of a discovery, and any figure quoting the odds against it would be dressing a tautology as a result.
The check that carries weight is the other one. EFSA's own degradation records name TFA parents independently of KEMI and of this screen. 1 of the 1 in the approved population is flagged here. So is the reverse test: 6 substances carry three or more fluorines without a CF3 group matching, and they are tracked as possible holes in the rule rather than quietly dropped.
What the screen is actually for is the 20 substances below that nobody has opened a file on. Those are a prediction, made from data that has been public for decades, and they are unconfirmed.
The whole shortlist
Ranked by fluorine payload and Swedish sales volume, combined by a rule written down in the source rather than learned from the substances it is checked against. Click a row for its formula and uses.
What the shortlist is and is not
A CF3 group means a substance can yield TFA. Whether it does depends on where that group sits and on degradation pathways this rule does not model, so the screen is built wide on purpose: 20 of the 26 carry no published TFA finding from any regulator. A bound that misses real cases is worthless, and one that includes some innocents is merely wide.
Two checks, neither of them an input. Kemikalieinspektionen named six TFA-forming substances on 2025-11-20, and all 6 are here. Separately, EFSA's own degradation records already list TFA as a metabolite for flutolanil, which the rule also flags without being told.
6 substances carry fluorine in some other arrangement, almost all of them difluoromethyl, and are excluded. Those degrade toward difluoroacetic acid, a related concern and a different compound. A further 18 have no PubChem structure and are left out of the population rather than counted as clean.
Reproduce with pipeline/35_run_tfa_screen.py. The rule, the weights and the held-out cohort are in screen/tfa.py. Structures are cached in the repository, so the screen runs offline and scores the same molecules a reviewer can read. Generated 2026-07-27.