One certifier has an index gap, and none offers a route a buyer would find

Two things about certification databases are true and checkable. One certifier's search cannot see some of its own live certificates. And none of the three offers a route a buyer would find to the complete set — every database can be enumerated, but only by a method you would discover from the network traffic.

This page previously argued something broader: that certifier search engines return wrong answers in general. That was too strong, and one of the cases supporting it was our own bug. We have retracted it below rather than quietly editing it. Queried properly, NSF’s and WQA’s search returns what they hold — we tested 160 products drawn from their complete sets and every one came back.

A live certificate its own search cannot find

IAPMO R&T publishes certificates through two systems. One serves a certificate by its id; the other is the search box. They disagree. Each certificate below is served as active and expiring in 2029, and returns zero rows in the search — including a search for its own file number.

These are certificates for a mainstream reverse osmosis system from a mainstream manufacturer. Listed, live, and invisible to the database's own search.

We did not go looking for this. We found it because we happened to hold certificate ids from earlier work and re-checked them. Without those ids we would have concluded the product was uncertified and published that.

Retracted: WQA’s keyword search works

What we published. That WQA’s keyword search missed LG Electronics despite LG Electronics holding 5 listings, and that enumeration was needed to find them. We offered it as evidence that certifier search is unreliable.

What is actually true. WQA’s search works. Searching it for LG Electronics returns thirteen listings. Our enumeration found 5. The search was more complete than we were.

What went wrong, twice, and both ours. The zero was a parsing bug: WQA returns results nested, with capitalised keys, and we read them flat and lowercase, which yields empty strings rather than an error. A working search looked like an empty one. The 5 was an incomplete enumeration: we browsed a hardcoded list of nine standard ids when WQA has nineteen, and among the ten we skipped was NSF/ANSI 401, holding 396 listings — including the two that make thirteen rather than 5.

A second claim falls with it. We recorded that WQA issues no NSF/ANSI 401 listings and used it to argue that a certifier can be silent about a standard it does not issue. WQA issues 401. We had read our own gap as theirs.

One thing survives. Searching WQA for that filter’s part number does return zero, and that is the correct answer — it holds no WQA listing. We reached a true conclusion through a broken instrument, which is worse luck than being wrong, because nothing prompted us to check.

We are publishing this rather than editing it away because the argument of this page is that verification is harder than it looks and that we do it carefully. A page making that argument cannot carry an unretracted false accusation against a certifying body. We got it wrong in exactly the way we warn others about: we concluded absence from an instrument that was not capable of finding the thing.

The standard everyone tells you to look for was never a standard

NSF/ANSI 473 is cited everywhere as the PFAS standard — by retailers, by review sites, and in manufacturers' own marketing. There is no NSF/ANSI 473 and there never was. It was NSF P473, a protocol, not an ANSI standard, and it no longer exists either.

A buyer following the common advice searches for a listing no certifier issues, finds nothing, and concludes the product is uncertified. The advice is not merely out of date. It names a category that cannot be looked up.

What to look for instead. PFAS reduction is certified under NSF/ANSI 53 (carbon and anion exchange) and NSF/ANSI 58 (reverse osmosis). On the listing, look for a Total PFAS claim, or PFOA and PFOS claims individually on an older listing.

Three further things we measured that are not widely stated:

• NSF/ANSI 401 does not cover PFAS. It covers emerging compounds and pharmaceuticals. NSF's complete 401 listing set contains no PFAS claim of any kind, and a 401 certification is frequently described as if it were one.
• Total PFAS covers seven compounds, not six. NSF's own listing footnote adds perfluorodecanoic acid (PFDA) to the six the usual summaries name.
• No reverse osmosis system in NSF's set currently holds a PFAS claim. NSF/ANSI 58 permits one; its complete listing set carried zero on 2026-09-21. Every PFOA, PFOS and Total PFAS row is under 53.

The two certifiers also word the claim differently, so a matcher built for one misses the other: NSF prints Total PFAS Reduction and PFOA Reduction; WQA prints Total PFAS and Perfluorooctanoic Acid (PFOA).

Pending. NSF task groups are working to update NSF/ANSI 53 and NSF/ANSI 58 to align with EPA's 2024 final PFAS rule, with a ballot pending. We re-check these entries every 30 days.

This one caught us too. Our own category filter tested for a 473 listing, which would have matched nothing at all, forever, while looking like it worked.

A certifier's silence about a standard it does not issue

Corrected 21 September 2026. This case originally used WQA and NSF/ANSI 401 as its example. WQA issues 401. The principle holds; the example was wrong and is replaced with one verified from the certifier’s own interface.

NSF’s drinking water treatment unit database has no NSF/ANSI 372 category. Its standard dropdown offers 42, 44, 53, 55, 58, 62, 177, 401, CSA B483.1, P231, P72, P508 and ASSE 1087 — and no 372. NSF does certify to 372, in a separate components database that is not the one a filter buyer searches.

So “we searched NSF and found no 372 listing” says nothing about a product. It says that database does not issue 372 listings. Counting that silence toward an absence lets a confident finding rest on a category that does not exist in the place you looked — which is precisely the error we made about WQA and 401, turned on ourselves.

Before concluding a product is not certified to something, check that the database you searched issues that something at all.

How to check a certification properly

Each database can be enumerated completely. None of these routes goes through the keyword search.

Then run a control query before trusting any result: search for a product you know is listed. If the control comes back empty, the database is not answering and no result from it means anything.

Two traps in the controls themselves. IAPMO's model search matches model strings, so a brand name returns zero there. And no single search mode is correct for every product: some brands are listed under a contract manufacturer's legal name and are invisible under their own, while others are their own listee and invisible under a model search. Search every mode and take the union.

What we changed because of this

We had published that a particular filter holds no listing in any of the three certifier databases. That claim rested on three negatives, one of which we then found to be unreliable, so we downgraded it to “we could not find a listing” — a fact about our search rather than about the product.

We have since earned the stronger claim rather than assumed it. NSF’s and WQA’s complete sets were enumerated, and IAPMO’s certificate id space was read end to end with nothing left unread. The filter appears under none of them, and no certificate anywhere in IAPMO is held by that manufacturer. That is an absence we may state. The first claim was right by luck and the second is right by evidence, and only the second is worth publishing.

Something better came out of it. Aftermarket replacements for that filter are certified — six active IAPMO certificates cover them, two carrying NSF/ANSI 53, health effects rather than taste and odour. The manufacturer’s own part is the uncertified one.

A negative is recorded at one of three strengths.

• Holds no listing. Every certifier's complete set was enumerated and the product is not in it.
• We could not find a listing. Every search came back empty and every control passed, but at least one certifier was searched rather than enumerated. A fact about our search, not about the product.
• We could not check. A database was unreadable or a control failed. This says nothing.

Only the first may be stated as absence. We also stopped counting a certifier’s silence toward an absence when that certifier does not issue the standard in question: NSF’s treatment-unit database returning no 372 listing says nothing, because that database has no 372 category.

The weaker claim is less satisfying to publish. It is the true one.

Every figure on this page was read from the certifier's own interface on 2026-09-21 and can be re-run. Corrections are welcome and we will publish them. Contact.