Originally published on Medium.
Large dispensers have been on package-level tracing for a year. The rule asks for data exchange, not a recall.
On 27 November 2026, FDA measures which American pharmacies count as small dispensers: twenty-five or fewer full-time licensed pharmacists or qualified technicians at the owning company. Larger dispensers have been under full package-level tracing since the same date last year.
So the drug supply chain is now a year into the most detailed traceability regime the United States has ever required of it. Serialized package identifiers, electronic transaction information, interoperable exchange between trading partners, most of it running on EPCIS.
Here is the question nobody is being asked on 27 November, and it is the only one that will matter on the day something goes wrong.
Can you use any of it to run a recall?
What the rule actually requires
Be precise, because this is where the confidence comes from.
DSCSA requires trading partners to exchange transaction information and statements electronically, to verify product at the package level, and to operate systems that interoperate with everyone else’s. It is a data obligation, and it is a real one. A year in, large dispensers are exchanging that data daily.
What the rule does not require is a demonstration. Nowhere does it ask a company to prove that, given a lot number at nine in the morning, it can produce by nine the next morning a list of every customer who received a unit from it.
Those are different capabilities. The first is a format. The second is a query, run under time pressure, across systems that were built to satisfy the first.
The gap is not theoretical
Two things from the last few weeks make the point better than an argument would.
When FDA wrote to a contract manufacturer in Guangzhou on 27 October, one of its demands was a list of all products distributed in the United States in the last three years, with lot number, quantity shipped and expiry dates, plus the customers and addresses they went to. That is the query. The agency had to ask for it, in a letter, with a fifteen business day deadline, because it is not something a regulator can pull.
And when a toothpaste manufacturer in Haridwar recalled on 10 September, it recalled everything still within expiry that it had shipped across two years, because the records that would have narrowed it were the records under question.
Neither of those firms is a DSCSA dispenser. The point is not the jurisdiction. It is that in both cases the system contained a great deal of data and still could not answer the one question a withdrawal turns on.
Why serialized data does not automatically answer it
A serialized identifier tells you that a specific package exists and that a specific transaction happened. A recall question runs the other way: start from a lot, fan out to every package in it, then to every party that took custody, then to what each of them did next.
That reverse traversal is not what transaction exchange was designed for. It usually crosses several trading partners’ systems, each holding its own slice, in formats that satisfy a standard without being joined up for interrogation. Each party can answer for its own hop. Almost nobody has tested the whole chain end to end against a clock.
The honest version is this. DSCSA has made the data exist. It has not made the answer fast.
Two pharmacies on the same street, two different answers
There is a second consequence of the 27 November measurement that nobody frames as a recall problem, and it should be.
A dispenser owned by a company with twenty-six or more full-time licensed pharmacists or qualified technicians crossed into full package-level tracing on 27 November 2025. One at twenty-five or fewer is exempt from five of those enhanced requirements until 27 November 2027, and FDA urges them to prepare in the meantime but asks them to file nothing.
So from this month there are two populations of American dispenser, side by side, holding the same products. One can in principle answer at the package level. The other has no obligation to, for another year.
Now run a manufacturer’s recall through that.
If you are withdrawing a lot, you have to reach every party that received a unit, and a meaningful share of them are small dispensers with no package-level requirement. Your ability to narrow the recall is capped by the least capable party in the chain, not by your own systems. You can serialize perfectly, exchange flawlessly, and still end up asking a thousand independent pharmacies to go and look on a shelf.
That is not a criticism of the exemption. Small pharmacies had a real case for more time and FDA gave it to them for sound reasons. It is a statement about where your recall width actually lives, which is downstream of you, in other people’s systems, and largely outside your control.
Worth knowing before you describe your traceability programme to a board.
Where the chain usually breaks
When the query does fail, it fails in predictable places.
It fails at the hop into a third-party logistics provider, where the data is held under a different schema and reconciliation is a manual job. It fails where a wholesaler aggregates and the lot-level link is preserved for compliance but not indexed for search. It fails at returns and resupply, where a unit comes back into the chain and rejoins it with a new transaction history. And it fails at the last hop, where the receiving party records the shipment rather than the packages.
None of those is a scandal. Every one is a system doing exactly what it was specified to do. The specification was exchange, and the need is interrogation.
The test to run before anybody makes you
This one costs nothing and takes a morning.
Pick a lot number at random from six months ago. Not a convenient one. Ask your team for a list of every customer who received a unit from it, and time how long it takes to arrive complete.
If the answer comes back in an hour, you are in better shape than most and you should say so in your next audit. If it takes three days and a spreadsheet merge, that is your recall timeline, and you have just discovered it for free rather than during a withdrawal.
If the answer is partial, note exactly where it broke. That hop is the whole problem, and it is usually one trading partner or one system that was never joined to the others.
What a good answer looks like
For contrast, because this piece should not only describe a hole.
The companies that answer the query fast have three things, and none of them is exotic. They hold their own lot-to-customer mapping rather than relying on being able to reassemble it from trading partner exchanges. They have run the query at least once as a drill, with a stopwatch, on a lot nobody chose for them. And they know in advance which hop in their chain is the slow one, so the recall plan routes around it instead of discovering it.
That is a week of work and an owner, not a platform purchase. It is also the difference between a withdrawal that names lots and a withdrawal that names a year.
This is one of ten
That exercise is question four of ten in a short self-assessment we published this week. It works out your recall width: whether a withdrawal would come down to individual units, to lots, to a production window, or to everything still in date.
It takes three minutes. Nothing is collected, there is no sign-up and no form, and your answers stay in your browser.
How wide would your recall be?
The reason it exists is that recall width is a cost figure, and it is fixed long before the recall. On the day a regulator calls, you cannot improve it. The questions are the decisions that set it.
What we are not saying
We are not saying DSCSA was a mistake or that the data is worthless. Serialization is the precondition for everything else, and a supply chain that can identify packages is better than one that cannot.
We are not saying anybody is failing. There is no published evidence that dispensers cannot answer the query, because as far as we can establish nobody has measured it. That absence is the point rather than an accusation.
And we are not saying the thing we build closes this. Unit-level identity tells you what a unit is when you are holding it. It does not assemble a distribution list across four companies’ systems on a Tuesday morning. What narrows a recall is a chain that can be traversed backwards, and that is an integration problem before it is a hardware one.
One sentence to take into 27 November
The rule measures headcount that day. The thing that will be measured on a worse day is how fast you can answer a question the rule never asked you to practice.
Aeroz tracks counterfeit enforcement, recalls and supply chain failures in regulated industries, three times a week. Follow for the next one. aeroz.io · [email protected]