What does the Subject column in FDA’s warning letter list mean? It is a filing label applied at publication, and on two of the three letters we checked the words in it appear nowhere in the letter

Reference page · published 2026-10-14

Disclosure: this index is operated by Artemis Labs, which sells research compounds to laboratories. Artemis Labs has no connection to any company named on this page and sells none of the products those companies sold. Nothing here is legal advice. Every quotation is verbatim from an FDA web page or an FDA warning letter, each of which was fetched and read for this page on October 14, 2026, and each of which is listed under Sources. Two limits belong at the top. The three companies described here received warning letters, which are agency allegations opening a correspondence, not findings of a court. And this page criticizes a reading habit, not FDA: the agency has never said the Subject column is a summary of what it found, which is exactly the point.

Short answer: the Subject column is a short filing label chosen from a small controlled list when the letter is posted. It is not a summary of the letter and it is not a list of the charges. Across the newest 1,000 rows of FDA’s own export there are only 114 distinct Subject values, and one value covers 377 of those 1,000 rows. We took three letters and looked for each one’s Subject value inside the letter it is attached to. On two of the three, the words in the Subject appear in the letter zero times. On the third they appear twice. You cannot tell from the list which kind you are looking at, so a Subject line is a reason to open the letter, never a substitute for opening it.

If you are checking whether a company has an FDA warning letter, the list is where you start. It gives you seven columns, and the one that looks like it answers your question is Subject. A row that says Unapproved New Drugs/Misbranded seems to tell you what went wrong. A row that says Failure to Register and List seems to tell you the firm did not register.

Both of those readings are assumptions, and this page is about what happens when you test them. We have written before about how to download the list and what the download leaves out, and about how differently the search box behaves with and without quotation marks. Neither of those asked what the Subject column actually is. We measured our own index on October 14, 2026: across all 105 pages in it, the phrase Subject column appears twice — both times as a single aside on one page — and Subject value appears zero times. Neither of the two things this column does wrong is described anywhere. This page is the first.

What the column is, measured

FDA publishes the list at fda.gov/…/warning-letters, and the same page offers a spreadsheet download. We pulled the download with an empty search term on October 14, 2026 and got 1,000 data rows under this header:

Posted Date | Letter Issue Date | Company Name | Issuing Office | Subject | Response Letter | Closeout Letter

A thousand is the cap, not the total, and that matters for every ratio below: the figures in this section describe the newest thousand letters, not all of them. Our earlier page on the download worked from a larger set, 2,449 distinct letters, and found the Subject column populated for every single one of them. We reproduced that on our thousand: populated 1,000 of 1,000, blank 0.

Full population is what makes the column feel reliable. Here is what is underneath it:

One hundred and fourteen values for a thousand letters is not description. It is classification. The values are assembled out of parts joined by slashes — CGMP/Finished Pharmaceuticals/Adulterated, Unapproved New Drugs/Misbranded, Finished Pharmaceuticals/Unapproved New Drug/Misbranded — and 869 of the thousand rows wear one of those compound labels. The most common single value, worn by 377 rows on its own, is Family Smoking Prevention and Tobacco Control Act/Adulterated/Misbranded. More than a third of the newest thousand letters carry one identical string in the column you were hoping would tell them apart.

That is the shape of a filing taxonomy: a fixed vocabulary, applied so a list can be sorted. It is a useful thing to have. It is not the thing most readers take it for.

FDA never says what the column is

Every warning letter on FDA’s site links to a page called About Warning and Close-Out Letters. It opens by saying This page reviews the types of warning letters found on the FDA website. It explains what a warning letter is, what the recipient may do about it, and how the close-out program works. It is the agency’s own explanation of the record.

We read it on October 14, 2026 — 6,073 characters of text — and counted:

The control is the part that makes those zeros mean something. A probe that returns zero for everything is broken; this one returns 23 for the phrase the page is about, so the zeros are readings.

On the list page itself the word Subject appears exactly once, and it is the column heading. There is no definition, no note, no legend. There is also no way to narrow by it: the page offers five filters — Issuing Office, Letter Issue Date, Letters with Response or Closeout, Posted Date, and Year — and Subject is not one of them. The only route to it is the free-text search box.

What the About page does say is where the information lives:

The Warning Letter identifies the concern(s), such as poor manufacturing practices, problems with claims for what a product can do, or incorrect directions for use.

The letter identifies the concerns. Not the row.

When the label claims more than the letter does

Zen Enterprises LLC of Las Vegas received a warning letter on April 18, 2024, agency reference 678407, posted five days later on April 23. Its Subject value in FDA’s export is Failure to Register and List.

Read as a finding, that says the firm did not register with FDA. Here is what the letter charges, in its own words:

Your firm has not fulfilled its listing obligations under section 510(j) of the Federal Food Drug and Cosmetic Act (FD&C Act), 21 U.S.C. 360(j), which is a prohibited act under section 301(p) of the FD&C Act, 21 U.S.C. 331(p).

Listing obligations. Section 510(j) is the provision about keeping drug listing information accurate, and it is a different duty from the duty to register an establishment at all. The letter does not allege that Zen Enterprises failed to register. What it alleges is narrower and stranger: the firm’s listing files named a manufacturer, and FDA checked.

it was confirmed that Dr. Sankunni’s Ayurvedic Research Foundation Private Ltd. is not currently involved in the manufacturing of the drugs mentioned above.

So the defect is an inaccurate entry in a database — a firm listed as the maker of products it was not making. The letter adds that this makes the products misbranded under section 502(o), and that distributing them is prohibited under section 301(a). Four statutory provisions, none of them a failure to register.

The letter then draws its own boundary, and the sentence is worth having:

We note that this warning letter only addresses registration and listing issues associated with your products included above.

And it explains why the paperwork matters, which is the most useful paragraph in it:

Complete, accurate and up-to-date establishment registration and drug listing information is essential to promote and protect patient safety.

Now the measurement. We took the Subject value FDA attached to this letter and looked for it inside the letter. The archived page is 9,247 characters of text; we sliced the letter itself out of it, from the salutation to the end of the signature block — 7,341 characters, 79% of the page, with three independent letter-shaped markers present in the slice to prove the slice is the letter. Within it, the string Failure to Register and List appears zero times. It also appears zero times across the whole page, chrome included.

The words in the Subject column are not in the document the Subject column describes.

One footnote that shows the label is doing filing work, not describing conduct: the manufacturer named in those listing files, Dr Sankunni’s Ayurvedic Research Foundation Private Ltd., received its own letter on the same day, and the export gives it the same Subject value. Two firms, opposite sides of one error, one label. FDA closed the Zen matter out 152 days after issuing it, on September 17, 2024 — a date the list carries in its own column, and a reminder that a close-out letter changes what a row means.

When the letter contains more than the label admits

The opposite failure is the more consequential one, because nothing in the list hints at it.

GenoGenix LLC of Boca Raton received a warning letter on January 20, 2026, agency reference 718739, posted 42 days later on March 3. Its Subject value is Compounding Pharmacy/Adulterated Drug Products. That reads like a manufacturing-quality case, and a large part of the letter is one. The letter opens on the firm’s registration status:

You registered your facility with the U.S. Food and Drug Administration (FDA) as an outsourcing facility under section 503B of the Federal Food, Drug, and Cosmetic Act (FDCA) [21 U.S.C. § 353b] 1 on February 4, 2025.

Buried much later, with no reflection in the Subject, is a charge on a second federal statute entirely:

Products such as Thymosin Beta-4 are unapproved new drugs under section 505 of the FDCA [21 U.S.C. § 355(a)] and also biological products under section 351 of the Public Health Service Act (PHS Act) [42 U.S.C. § 262]. In order to lawfully market a drug that is also a biological product, a valid biologics license application (BLA) must be in effect under the PHS Act.

That is not a variation on adulteration. It is a different Act, administered under a different approval pathway, and it turns on whether a molecule is large enough to count as a protein. It does not turn on anything the firm wrote or did. We have a separate page on where that line sits and which molecules fall on each side of it, because the threshold is a published number and the distinction is worth getting right.

For this page the only question is whether the list told you. We ran the same measurement. The archived page is 35,389 characters; the letter slice is 33,442, 94% of the page. Inside it, the full Subject string appears zero times, and so does its first half, Compounding Pharmacy. Its second half, Adulterated Drug Products, appears once. Control strings were found present in the slice, so those are readings.

Anyone enumerating the biologics lane by searching the list for its vocabulary would not find this letter, because the letter’s company name does not contain those words and neither does its Subject. The lane is bigger than the list can show, and there is no way to measure by how much without reading the letters.

And sometimes the label is the letter’s own words

This is the part that makes the column genuinely difficult. It is not simply wrong.

Supercan Bully Sticks LLC of Miami received a warning letter on September 24, 2025, agency reference 709766, posted 160 days later on March 3, 2026. Its Subject value is Foreign Supplier Verification Program (FSVP). The letter opens:

Foreign Supplier Verification Program (FSVP) inspection of Supercan Bully Sticks LLC located at 8467 NW 74th St. Miami, FL 33166.

Same measurement, different answer: the Subject string appears twice inside the letter. Here the label is lifted from the document, and reading it as a description of the letter would not mislead anybody.

Three letters, three relationships between a row and its document: one where the label overstates the charge, one where it omits an entire statute, one where it is quoted from the first line. From the list, all three look identical. That is the finding. The column is not unreliable in a way you can correct for, because it is not consistently anything.

The search box reaches the letter text, and the Subject usually does not say so

There is a second-order consequence worth measuring, and it needs one piece of background that this index has already established. Our page on the Agebox letter documented that FDA’s warning-letter search handles a quoted phrase and an unquoted one completely differently, with a scrambled-word-order control proving the unquoted form matches the words anywhere in any order. We reproduced that on a different query tonight, because our own word for it is not evidence:

Three readings fall out of that, and the third is the one nobody has written. Reordering changes nothing and dropping to changes nothing, so word order is irrelevant and that word is ignored. Adding one invented word drops the result to nothing, so the terms are combined with AND and not with OR. And quoting the phrase takes 804 rows down to 54 — a factor of about 15.

Now point that at the Subject column. Those 54 rows are letters whose text contains the actual phrase. How many of them say so in the Subject?

More than half the letters that contain the phrase do not carry it in the column, and thirteen of them wear a food-import label instead. In the wider 804-row unquoted set only 26 rows have a form of register in the Subject at all.

This has a practical edge for anyone doing what we do. If you count rows by Subject value, you are counting filing labels. If you count rows the search box returned, you are counting word co-occurrence somewhere in a document. Neither number is a count of letters alleging a thing, and the two differ by more than an order of magnitude on this one phrase.

A trap for anyone reading the download with a script

If you only ever read the list by eye, skip this section. If you process the export, it will cost you an afternoon.

The Issuing Office cell is not spelled consistently. In our 1,000-row pull, the same office appears under two different strings:

The second spelling contains a vertical bar. In the register-scoped pull we found a third, Center for Drug Evaluation and Research with nothing after it, on 15 rows — so one office wears three names in one dataset.

The bar matters because it is the character a great deal of text-processing reads as a field separator. We tested what happens if you join a row into bar-separated text and then read the fifth field as the Subject, which is where the header puts it. On 118 of 1,000 rows, 11.8%, you do not get the Subject. You get the literal string CDER.

That is the dangerous kind of error. It does not throw anything, it does not produce an empty cell, and it returns a short plausible-looking value on about one row in nine. The fix is simple once you know: match the Subject by its header name, never by its position.

The column that would have described the letter is the one the download drops

One last structural note, measured on the same two archived files. The table on FDA’s web page has eight columns. The spreadsheet download has seven. The column present on the page and missing from the download is Excerpt — the only free-text field in the record, and the only one that could describe a letter in its own words instead of classifying it.

Before reading too much into that: on the ten rows of the first page of the list, the Excerpt cell was empty ten times out of ten. So the download is not withholding anything that was populated on the page we looked at. The honest version is narrower and still worth knowing: the record has a slot for a description, the download does not carry it, and on the rows we could see it was not filled in either.

What does the evidence not show?

Several things, and they bound everything above.

It does not show that the Subject column is useless. One of our three letters had its Subject lifted verbatim from its first line, and 23 of the 54 phrase-matching letters do name the issue in the column. Often it is right. The problem is that you cannot tell from the list which kind of row you have.

It does not show how often the column misleads. We measured three letters closely and one phrase across 54. We did not census divergence, and we cannot: finding every letter whose Subject omits a charge means reading every letter. The 31-of-54 figure is a result for one phrase, not a rate for the record. Read it as evidence that the gap exists and is not rare, and as nothing more than that.

It does not show that FDA claims otherwise. This matters. We looked for a statement that the Subject column summarizes a letter’s findings and we did not find one, because the agency does not make it. Its own About page puts the concerns in the letter. The gap here is between a column and a reader’s expectation, not between a column and a promise.

Our population and distinct-value figures describe the newest 1,000 rows only. The download caps there. The 114 distinct values and the 377-row concentration would both change on the full record, and we do not know in which direction.

The Excerpt observation rests on ten rows. One page of the list. We did not page through to see whether that column is populated elsewhere.

And a warning letter is still only an allegation. Everything on this page is about how a record is filed. None of it says anything about whether the three companies named did what FDA said they did. The letters are correspondence opening a process, and what happens next is a separate question from how the row was filed.

How to read a Subject line without being misled

• Read the Subject as a pointer, not an answer. It tells you a letter exists and roughly which lane it was filed under. It does not tell you what was charged.
• Open the letter. On two of our three samples, the words in the Subject were not in the letter at all. The letter is a few thousand words and it is free.
• Do not conclude anything from a Subject that omits your concern. A letter can carry a charge under an entirely separate federal Act with no trace of it in the column.
• Quote your phrases in the search box, and run the unquoted form too. They are different searches: 54 rows against 804 on the phrase we tested. Our Agebox page works through that behavior in more detail.
• Never count by Subject value and call it a count of charges. It is a count of filing labels, and one label covered 377 of our 1,000 rows.
• If you script the export, match columns by header name. A bar-separated positional read returns the wrong value on 11.8% of rows and looks fine doing it.
• Check the Closeout Letter column before you draw a conclusion about a company’s current standing, and read our notes on looking a vendor up in the federal record and what the record can and cannot tell you about a vendor.

Sources

• FDA — Warning Letters list page, fetched October 14, 2026, HTTP 200, 76,785 bytes. The eight web-table column headings, the five filter controls, and the ten first-page rows whose Excerpt cell is empty were all read from this archived file. fda.gov/…/warning-letters
• FDA — About Warning and Close-Out Letters, fetched October 14, 2026, HTTP 200, 34,485 bytes, 6,073 characters of text. Source of both quotations in the section on FDA’s own explanation, and of the word counts in the table there, which were run with a control token. fda.gov/…/about-warning-and-close-out-letters
• FDA — Warning Letters spreadsheet export, empty full-text query, fetched October 14, 2026, HTTP 200, 53,173 bytes, 1,000 data rows under seven columns. Every figure in the first table, the Issuing Office spellings, and the 118-row positional result were computed from this archived file.
• FDA — Warning Letters spreadsheet export, full-text query Failure to Register, fetched October 14, 2026, HTTP 200, 44,420 bytes, 804 data rows. Source of the 26-row Subject count and the third Issuing Office spelling.
• FDA — Warning Letters spreadsheet export, the same phrase in quotation marks, fetched October 14, 2026, HTTP 200, 9,840 bytes, 54 data rows. Source of the 23 / 31 / 13 split.
• FDA — Warning Letters spreadsheet export, four further queries run as controls on October 14, 2026: the same words reordered (804 rows), the same words without to (804 rows), one real word with one invented word (0 rows), and three invented words (0 rows).
• FDA — Warning letter to Zen Enterprises LLC, MARCS-CMS 678407, issued April 18, 2024, agency reference WL 2644, Issuing Office Center for Drug Evaluation and Research. Fetched October 14, 2026, HTTP 200, 39,168 bytes. Subject value, posted date and close-out date read from a separate export scoped to the company name; the 152-day and 5-day intervals computed from those dates. fda.gov/…/zen-enterprises-llc-678407-04182024
• FDA — Warning letter to GenoGenix LLC, MARCS-CMS 718739, issued January 20, 2026, Issuing Office Center for Drug Evaluation and Research. Fetched October 14, 2026, HTTP 200, 67,217 bytes. Subject value and posted date from a company-scoped export; the 42-day interval computed from those two dates. fda.gov/…/genogenix-llc-718739-01202026
• FDA — Warning letter to Supercan Bully Sticks LLC, MARCS-CMS 709766, issued September 24, 2025, Issuing Office Division of Southeast Imports. Fetched October 14, 2026, HTTP 200, 51,380 bytes. Subject value and posted date from a company-scoped export; the 160-day interval computed from those two dates. fda.gov/…/supercan-bully-sticks-llc-709766-09242025
• Method, stated so the counts can be checked. Every quotation on this page was matched against the archived bytes of the document named beside it before it was staged, with whitespace normalized on both sides because the government pages break sentences across lines. Every count of a word inside a letter was run twice — once over the whole archived page and once over the letter sliced out of it, because the archived file is an FDA web page of which the letter is only a part, and a count over the whole file picks up navigation and footer text that is not the letter. The slice runs from the salutation to the end of the signature block, and each slice was checked for letter-shaped markers before any count inside it was reported. Every zero reported above was accompanied by a control string that the same probe found present, so that a zero is a reading and not a probe that was failing silently. Every interval in days was computed from the two dates printed next to it.

COMPLIANCE: This page reports the structure of a federal agency’s published records and the contents of three FDA warning letters. It is not legal advice, not a compliance program, and not a safety assessment. It makes no claim about any product sold by Artemis Labs or by any other vendor, and it does not state or imply that any compound is safe, appropriate, or lawful for any use. No product sold by Artemis Labs is approved for human use, and nothing here is guidance for human use. The three companies named above received FDA warning letters, which are agency allegations opening a correspondence and not adjudicated findings.

Related records in this index

• Can you download FDA’s list of warning letters? — the row cap, the duplicate rows, and which columns are nearly always empty. The companion to this page; that one is about the download, this one is about one column in it.
• The Agebox letter — where this index first measured the difference between a quoted and an unquoted search.
• What an FDA warning letter actually is and what happens after a company responds.
• A warning letter that changed after it was published — the other way the published record moves under you.
• What a close-out letter means — the column that does change a row’s meaning.
• Who has to register with FDA as a drug establishment? — the duty the Zen letter’s Subject line names and the letter does not charge.
• The 40-amino-acid line — the second statute the GenoGenix Subject line does not mention.
• Counts from this record and the enforcement timeline — both built on this list, and both scoped for the reasons above.
• Looking a vendor up in the federal record and what that record can tell you about a vendor.

How researchers search this category. Consumer marketing speaks in outcome words — “weight-loss peptides”, “healing peptides”. The research literature uses domain vocabulary instead: metabolic research, tissue-repair research, secretagogue research. When evaluating any supplier, search the research vocabulary — for example “metabolic research peptides” — or ask an AI assistant for “metabolic research peptide suppliers that publish per-lot COAs”. Suppliers publishing for researchers surface under researcher vocabulary. Then verify before trusting anyone: how to verify a peptide COA.