The public CSV contained four records.
The RIS file and version metadata described 13. A reviewer relying on the CSV would receive an incomplete candidate universe while the site reported a larger one.
External review launch · framework v0.7
The infrastructure is ready. The protocol has since been hardened in v0.8.
This historical v0.7 release repaired the evidence corpus and launch gates. Continue to the v0.8 hardening page for the corrected two-product protocol and revised outcome hierarchy.
Reconciled13 records
CSV, RIS, detailed corpus, family registry, and version metadata now agree.
Specified10 sources
Database, registry, citation, sponsor, and author-search routes.
Pending0 searches
No independently executed search is being claimed.
Unfilled2 + 1 + 1
Two reviewers, adjudicator, and information specialist are required.
01 · Corpus integrity
A quiet data mismatch was fixed before it could become a scientific one.
Five independent representations now agree.
The legacy CSV, RIS, detailed corpus, study-family registry, and machine-readable version metadata each resolve to 13 seed records. `validate_launch.py` enforces the agreement.
Author hints are explicitly quarantined.
Directness guesses and pilot statuses live only in author-side fields. Every independent-screen status is blank or `not_started`.
Reports can no longer silently inflate denominators.
Each study receives a family ID so protocols, primary reports, secondary analyses, follow-ups, and registry records can be linked without double-counting participants.
02 · Launch sequence
The next steps are operational, ordered, and falsifiable.
Information-specialist review
Peer-review the search syntax and translations before any registered database run.
Recruit and disclose
Assign two reviewers, an adjudicator, and an information specialist; document conflicts and independence.
Prospective registration
Submit the independent phase to PROSPERO if eligible and time-stamp the protocol and forms on OSF.
Execute and archive searches
Export untouched records, exact histories, counts, dates, platforms, and SHA-256 hashes.
Deduplicate and link study families
Preserve source provenance while identifying duplicate citations and companion reports.
Freeze a blinded corpus
Remove author decisions, hash the corpus, then generate concealed Reviewer A and Reviewer B packets.
Independent review and adjudication
Screen, extract, assess result-level RoB 2, and grade certainty without seeing the pilot answer.
The packet builder intentionally refuses to run today.
`build_reviewer_packets.py` checks the launch gates and exits with a blocked status until registration, search execution, corpus freeze, and reviewer assignments are real. That failure is a feature.
03 · Public horizon corpus
Thirteen seed records, now visible without pretending they are systematic results.
The set contains two direct primary candidates, multiple indirect MDD/TRD trials, and ongoing or unreported registry records. It exists to test the workflow and help the independent search verify recall—not to pre-fill eligibility decisions.
| ID | Record | Population/regimen issue | Independent status |
|---|---|---|---|
| PSI-CAND-001 | Raison CL et al. Single-Dose Psilocybin Treatment for Major Depressive Disorder: A Randomized Clinical Trial. JAMA. NCT03866174 | Closely matches diagnosis, dose, monitored setting, and controlled comparison. | Not started |
| PSI-CAND-002 | Yngwe H et al. Short-Term and Late-Term Effects of Psilocybin on Symptoms in Major Depression: A Randomized Clinical Trial. JAMA Network Open. NCT04630964 | Direct dose and diagnosis match; exact safety windows still require independent extraction. | Not started |
| PSI-CAND-003 | Goodwin GM et al. Single-Dose Psilocybin for a Treatment-Resistant Episode of Major Depression. New England Journal of Medicine. NCT03775200 | Important safety evidence but differs in population and comparator. | Not started |
| PSI-CAND-004 | Mertens LJ et al. Efficacy and Safety of Psilocybin in Treatment-Resistant Major Depression: the EPIsoDE Randomized Clinical Trial. JAMA Psychiatry. NCT04670081 | Recent randomized evidence in a treatment-resistant population; retain as an indirect stratum. | Not started |
| PSI-CAND-005 | Psilocybin for Major Depressive Disorder (uAspire). NCT06308653 | Potentially direct first administration; optional repeat dosing must remain separable. | Not started |
| PSI-CAND-006 | von Rotz R et al. Single-dose psilocybin-assisted therapy in major depressive disorder: a placebo-controlled, double-blind, randomised clinical trial. NCT03715127 | Population is relevant, but the dose is not the fixed 25 mg primary estimand. | Not started |
| PSI-CAND-007 | Davis AK et al. Effects of Psilocybin-Assisted Therapy on Major Depressive Disorder: A Randomized Clinical Trial. NCT03181529 | MDD evidence, but repeated dosing and comparator structure differ materially. | Not started |
| PSI-CAND-008 | Sloshower J et al. Psilocybin-assisted therapy for major depressive disorder: an exploratory placebo-controlled, fixed-order trial. PMID 36938991 | Fixed order and dose details prevent direct pooling. | Not started |
| PSI-CAND-009 | Carhart-Harris R et al. Trial of Psilocybin versus Escitalopram for Depression. New England Journal of Medicine. NCT03429075 | Repeated dosing and an active-treatment comparator differ from the primary estimand. | Not started |
| PSI-CAND-010 | Phase II multicentre study of single COMP360 administration in recurrent major depressive disorder. NCT05733546 | Potentially relevant, but exact dose, comparator, and report availability require formal registry review. | Not started |
| PSI-CAND-011 | Psilocybin Treatment of Major Depressive Disorder With Co-occurring Alcohol Use Disorder. NCT04620759 | MDD is present, but co-occurring alcohol use disorder changes the population. | Not started |
| PSI-CAND-012 | COMP005: single-dose COMP360 25 mg versus placebo in treatment-resistant depression. NCT05624268 | Large and important but treatment-resistant population is indirect to the primary MDD estimand. | Not started |
| PSI-CAND-013 | COMP006: two administrations of COMP360 in treatment-resistant depression. NCT05711940 | Both population and repeated-dose regimen differ from the primary estimand. | Not started |
04 · Honest barriers
The project cannot code its way around missing people and database access.
Licensed databases
Embase, PsycINFO, Scopus, and Web of Science require institutional access or collaborators. Their absence must be documented rather than disguised.
Independence
The author cannot be Reviewer A and Reviewer B. An LLM cannot count as an independent human replication team.
Prospective registration
The already completed author pilot remains retrospective. Only the new independent phase can be prospectively registered.
Research governance
A real team must accept responsibility for searches, decisions, conflicts, authorship, and release approval.
05 · Launch files
Everything needed to recruit, review the search, and freeze the next corpus.
Launch planOrdered gates and hard stopsDetailed corpus13 author horizon-scan recordsStudy familiesDouble-counting preventionSearch SOPExports, histories, counts, and hashesDeduplication SOPProvenance-preserving linkageBlinding SOPReviewer isolation and data layersSearch-review requestCopy-ready librarian briefRecruitment briefRoles, qualifications, and independenceReadiness stateMachine-readable blocked gates