At two German university hospitals, 144 people were sorted into three groups. Every one of them was in a depressive episode that had already resisted at least two antidepressants from different drug classes. One group took 25 milligrams of psilocybin, one took 5 milligrams, one took a vitamin. Six weeks later the trial asked who had cut their score in half on the Hamilton depression scale, which runs from 0 to 52. Among the 142 whose results could be analyzed, the paper counts 8 of 47 responders on the full dose, 6 of 48 on the low dose, and 5 of 47 on the vitamin.
Three people is the whole margin, and it is far too small to tell us anything in either direction. That is EPIsoDE, a phase 2b trial published in JAMA Psychiatry in March, and it missed its primary endpoint, the one result a trial commits to in advance and is judged on. It landed a month after Compass Pathways announced that both of its far larger phase 3 trials had hit theirs. The two outcomes look like a contradiction. They are not, and what they share matters more: every psilocybin result here, the hits and the miss alike, was measured against a comparison group that could tell it was the comparison group.
Neither program handed anyone a mushroom: both gave a standardized synthetic dose under clinical supervision alongside psychological support, which in EPIsoDE meant seven preparatory and integration sessions running to fourteen hours in all. Start with what Compass has, from the company's own announcements, which are a company describing itself. In COMP005, reported in June 2025, a single 25 mg dose beat placebo by 3.6 points on the MADRS, a different depression scale running from 0 to 60. The plausible range around that figure runs from 1.5 to 5.7 points. In COMP006, announced this February, two doses beat a 1 milligram dose by 3.8 points. Note that second comparator: the widely quoted 3.8 is not against a placebo, it is against a sliver of the same drug.
Both results were highly unlikely to be chance, and both came in a genuinely sick population; in COMP006, current depressive episodes had been running over three years on average. So: real, reproduced twice, and small. Compass CEO Kabir Nath called it "a remarkable achievement for the field of psychiatry," though the company's own claim is narrower than the coverage around it: the first classic psychedelic to post results like this twice, not a settled approval. Jerry Rosenbaum, who directs Massachusetts General Hospital's Center for Neuroscience of Psychedelics and had no part in the trials, told the medical news site STAT the data "probably meets the bar for approval. It doesn't shout out to you that this is miraculous."
Compass set the target itself, in public, before the trials read out
The gap between Nath's remarkable achievement and Rosenbaum's not-miraculous is mostly about who chose the measuring stick. Compass counts a response when a patient's score drops by 25 percent or more. As Drug Discovery & Development reported in February, that is a lower bar than the 50 percent cut normally used in antidepressant trials. It is the bar EPIsoDE used. It is also the one Newcastle University's Hamish McAllister-Williams applies to the March results, concluding that the treatment is "not a panacea with many patients not achieving at least a 50% improvement in symptoms." His comment came in a roundup of expert reaction gathered by Britain's Science Media Centre, which also solicited the comments from Anthony Cleare and Robin Carhart-Harris quoted below. On Compass's measure, 39 percent of the two-dose group responded, against 25 percent for the single dose in the earlier trial. Overall remission rates, the number a patient actually cares about, were still unreleased for both trials as of that February report.
The target was also named in advance by the party being measured. As the trade publication Psychedelic Alpha reported, Chief Medical Officer Guy Goodwin had said earlier that year that he and his colleagues "would be very pleased to see effects of over 3 on the MADRS scale," adding, "We think that is clinically significant, anything above that is a bonus." He meant the difference between the active and placebo arms, so it is the 3.6 that answers him, and it clears his line while its lower bound does not. A Deutsche Bank analyst put commercial relevance nearer 4.5. This is not bad faith. It is worth knowing that the ruler and the thing being measured came from the same building.
EPIsoDE was built to detect an effect nobody had grounds to expect
EPIsoDE was run by the Central Institute of Mental Health in Mannheim on German federal research money, with no Compass involvement, though independent is not the same as unconflicted: the Usona Institute is listed as a collaborator, and several authors declare fees, grants or shareholdings involving Beckley Psytech, MindMed and the clinic group OVID.
Its authors are unusually honest about why it missed. They powered the trial expecting the kind of effect seen in an early open-label study, an expectation they now call an overestimate, and the calculation was not based on a prespecified minimum important difference, the smallest change agreed in advance to matter to a patient. They call that "an omission for a phase 2b trial." A study built to catch a large effect will miss a small one. Their secondary measures did show a clinically meaningful improvement, but once the primary comparison fails, what follows is exploratory rather than confirmatory. Their verdict, in full, is that "while overall this constituted an inconclusive trial, these results add to the existing evidence on the potential of psilocybin treatment for depression." Inconclusive is not the same as negative, and they are careful to say so.
The safety signals are real, and they cut both ways
Nor is this side-effect free. In EPIsoDE, reports of suicidal ideation on dosing days were higher on the full dose, 4 percent against 1 to 2 percent in the comparison groups, and there were two serious adverse reactions after it, one a case of hallucinogen persisting perception disorder, a visual disturbance that outlasts the drug. Compass's earlier trial in the New England Journal of Medicine recorded suicidal ideation, behavior or self-injury in all dose groups. Against that, the independent board monitoring the phase 3 trials found no meaningful imbalance in suicidality between arms, and COMP006's one serious event of suicidal behavior occurred in the 1 milligram group. Most other effects, headache, nausea and visual hallucinations, arrived on dosing days, and the vast majority resolved within a day.
Almost everyone could tell what they had taken
EPIsoDE tried harder than most to keep patients guessing, using both a vitamin and a low psilocybin dose as comparators, and then did the thing almost nobody does, which is check whether the blinding held. It did not. 86 percent of participants correctly identified the 25 milligram dose. The vitamin group also did worse than placebo groups usually do, which Anthony Cleare of King's College London calls a "know-cebo" effect, the disappointment of not getting the interesting drug making people worse. Cleare has taken consulting fees and honoraria from Compass Pathways among others, and grant funding from Beckley Psytech. EPIsoDE's authors call their sample clinically representative, and suggest that characteristics specific to treatment-resistant depression may also account for the low comparison-group response.
Published the same day in the same issue, a meta-analysis by Zachary Williams, Balazs Szigeti and colleagues pulled that thread properly. If psychedelic trials are unblinded anyway, they argued, compare them against ordinary antidepressants given openly, so both sides know what they took. Across 24 trials they detected no difference between the two. More telling: blinding shifted the results for ordinary antidepressants and did not shift them for psychedelic therapy, which is what you would expect if those patients already knew. The psychedelic side rests on 8 trials and 249 patients against nearly 8,000, so this is a difference not detected rather than a sameness demonstrated.
Follow that back and it reaches Compass too. If patients on 25 milligrams of psilocybin can reliably tell, then 3.6 and 3.8 points are also gaps measured against people who knew they had drawn the short straw, and part of each gap may be the comparison group sinking rather than the treated group rising. How much is genuinely unsettled, and the strongest check so far cuts the other way: EPIsoDE's authors looked at whether knowing had moved their own week-6 results, and say their data suggests it played a negligible role, while adding that the broader question cannot be settled from what they measured. That is not a reason to dismiss the phase 3 results, which are real and were reproduced. It is a reason to stop reading them as though the comparison were clean.
Robin Carhart-Harris of UCSF, among the field's most prominent researchers, goes further and rejects the reasoning outright. He co-authored Compass's own earlier trial and advises Atai-Beckley. Expectations are a poor predictor of who responds, he says, and in his own trial pitting psilocybin against escitalopram, a common SSRI antidepressant, the patients who expected the most from psilocybin did the worst. That trial was phase 2, 59 patients, and enrolled people with major depression rather than the treatment-resistant kind. Of comparing separate trials he says, "It's proposed as comparing apples with apples when it is more like comparing apples with oranges," and he does not stop at the caveat: "I would therefore advise against drawing conclusions from the results of this meta-analysis."
He may be right that the method cannot carry the claim. But his own preferred alternative is the one nobody has delivered: psilocybin has never been tested head to head against an existing treatment at phase 3 scale, in treatment-resistant patients, powered to settle which is better. Until someone runs it, the honest position is that psilocybin does something real for some people who have run out of options, that the effect is smaller than a decade of coverage promised, and that some unmeasured share of it may come from knowing you took it.
None of this is available to anyone. Psilocybin remains a Schedule I drug in the United States, the most restricted category, so even approval by the FDA, the American regulator, would not put it in a clinic without a separate rescheduling decision from the Drug Enforcement Administration. Compass expects to complete its submission late this year, and its July update reports responders holding their benefit, on average, through at least 26 weeks. An average is not a floor, and some will have relapsed sooner. Compass has reported remission among the patients who had already responded, but not the plain figure for everyone who entered. For a drug positioned to change psychiatry, the share of all comers who got fully well rather than somewhat better is the number to ask for first.
