Who makes the psilocybin in a clinical trial?

Across the 328 psilocybin trials in our corpus, 26 carry a compound supplied by someone other than the trial’s sponsor, and seven branded product codes turn up in those records. Six resolve to an owner; the seventh is held, because the sponsor names on the trials carrying it point to more than one legal entity and our rule refuses to choose.

Two words need pinning down. A code belongs to an organization when exactly one company-type sponsor appears among the trials carrying it, which is our own corpus attesting our own corpus with nothing else behind it. “Supplies” means the registry record for that trial carries the organization’s compound and nothing further, which is no evidence of a contract, a payment or a manufacturing arrangement. It is a name in a field.

Five of the seven supply only trials they also sponsor. Two do not. Usona’s eleven supplied trials and its four sponsored ones are disjoint: no trial appears in both.

In the intervention field of a ClinicalTrials.gov record, the drug in one of those eleven trials is named like this:

The answer is typed by the sponsor into a field anyone can read, and what does not exist is a way to ask for it. Our own published trials extract carries 47 fields per row, six of which describe the sponsor and none of which describes the product, so you can ask who funded a psilocybin trial six ways and cannot ask what was in the capsule at all. ClinicalTrials.gov, the European Union’s CTIS and the UK’s ISRCTN each record the intervention as free text, where a product name turns up when it turns up at all; in one ISRCTN record that field runs to 8,119 characters.

On 29 April 2026 Red Light Holland Corp. completed its acquisition of Filament Health Corp. by statutory plan of arrangement under British Columbia’s Business Corporations Act, and Filament became a wholly owned subsidiary. The supplier of sixteen trials changed hands, the sponsor strings on those trials stayed exactly as they were, and a sponsor-indexed view of the field recorded nothing at all.

Our own merged corpus put Filament at eleven, and it was wrong. Five of the sixteen name PEX010 only in their European record, and our merge kept the American one, which names psilocybin generically, so the product name died inside our pipeline before anything downstream could read it. Re-deriving from the raw files on 1 September 2026 recovered all five, and on 5 September 2026 we changed the merge to keep every record’s intervention names and added a check that fails the corpus build if one goes missing again. All sixteen survive.

The five were pmg-tr-00168 (Brugmann University Hospital), pmg-tr-00181 (Northern Stockholm Psychiatry), pmg-tr-00189 (University Hospital, Ghent), pmg-tr-00291 (Nantes University Hospital) and pmg-tr-00311 (ARQ National Psychotrauma Centre).

Identity is the other half of the difficulty: the name Filament Health Corp. sits on two British Columbia companies, one active on the register and one inactive, and our record has not established which of them signed the April arrangement.

A registry record names what a trial intends to give. It never says whether material moved. Of the 328 trials, 292 produce no candidate supply edge at all, and 271 of those name psilocybin with no product name attached, so part of what is measured here is which organizations brand their material. No primary-source verification pass sits behind a single edge, and concentration here may be entirely ordinary: pharmaceutical-grade psilocybin is difficult to make and tightly scheduled, and a few producers serving many academic trials is what an early, heavily regulated supply chain looks like.

None of this speaks to any product’s quality, safety or effect, and nothing changes who is eligible for a trial or how anyone joins one. No organization named here is accused of anything. Every string behind the finding sits in a public registry record, typed there by the trial’s own sponsor. A nonprofit supplying eleven trials is a fact about research infrastructure, two organizations is a thin layer, and 8% is a floor. The question “who made this” has an answer in the record, and no sponsor-indexed view that will return it.