The data, downloadable
Everything this site renders is served as plain files from stable URLs, rebuilt by the same pipeline that rebuilds the pages. Counts below are live as of the current build (Index v2026.11.18); recompute any number on this site from these files.
| Dataset | Rows | Rows come from (share · source last read) | Download | License |
|---|---|---|---|---|
| Clinics & treatment centers Ketamine, esketamine and psilocybin treatment sites, geocoded, with published contacts and a legal-status class on each row we could classify. A stated legal basis appears on the licensed and authorised-prescriber sites. lic carries a state licence as the register lists it: a Colorado healing center's number and status, or an Oregon licensee's OHA licence type (Oregon publishes active licences only). |
2,248 | HealingMaps 87.6% · last read 2026-09-02 source is blocking us Agent sweeps (/agarikon-sweep) 8.1% · last read 2026-09-09 behind schedule Colorado Natural Medicine Division + DORA licence lookup 2.5% · last read 2026-11-17 ClinicalTrials.gov 1.1% · last read 2026-10-16 last run failed Oregon Health Authority — consenting licensees 0.8% · last read 2026-11-16 |
JSON · CSV | HealingMaps: Directory listings; not relicensed Agent sweeps (/agarikon-sweep): Published pages only; source URL on every row Colorado Natural Medicine Division + DORA licence lookup: Colorado public records ClinicalTrials.gov: U.S. Government, public domain Oregon Health Authority — consenting licensees: Oregon public records; per-field directory consent honored |
| Oregon facilitators Psilocybin facilitators licensed by the Oregon Health Authority, the people who run a client's preparation, administration and integration sessions. The list is OHA's public licensee directory, which carries only licensees who agreed to appear in it. Each row has the name and licence, plus a website where the facilitator gave one. We leave out the phone and email the directory shows: most facilitators work alone, so those reach the person directly. The directory gives no address either, so facilitators are listed in the index under Clinics and never mapped. |
344 | Oregon Health Authority — consenting licensees 100% · last read 2026-11-16 | JSON · CSV | Oregon public records; per-field directory consent honored |
| Trial sites One row per registered-trial location worldwide. unnamed=1 marks a location the registry published with no facility name — the row belongs to its study and is not a distinct named site. |
4,528 | ClinicalTrials.gov 100% · last read 2026-10-16 last run failed | JSON · CSV | U.S. Government, public domain |
| Studies One row per registered clinical trial (deduplicated NCT records). |
1,604 | ClinicalTrials.gov 100% · last read 2026-10-16 last run failed | JSON · CSV | U.S. Government, public domain |
| Papers Published psychedelic-medicine literature with citation counts, journal, topic tags, the substances studied (drugs) and open-access status (oa). rt=1 marks a work OpenAlex lists as retracted and rn=1 a notice that retracts one; both sort last and count toward no author's totals. Works outside the index's scope never enter it: exclusions.json counts them under “Publications: scope”. |
32,289 | OpenAlex 100% · last read 2026-11-04 | JSON · CSV | CC0 |
| Retreats & legal access Retreat centers and legal access points, with the law each works under where we could establish it. |
217 | HealingMaps 53.5% · last read 2026-09-02 source is blocking us Agent sweeps (/agarikon-sweep) 46.5% · last read 2026-09-09 behind schedule |
JSON · CSV | HealingMaps: Directory listings; not relicensed Agent sweeps (/agarikon-sweep): Published pages only; source URL on every row |
| Researchers & Authors One row per person in the entity graph: authors with 3+ papers in this corpus, PubMed researchers with 3+ psychedelic-therapy papers since 2015, NIH principal investigators, trial investigators and contacts, and 45 authors listed here before who have since fallen under three papers, kept so their links still work. papers, fa, sa and cit count the person's papers in this index (retracted works and notices left out) and those papers' citations, not a whole career. inst is the latest institution on their papers and inst_year the year of the paper that names it. lat/lon sit near that institution's ROR location (gp=institution) when inst_year is 2015 or later; a person known only from an older affiliation is listed but not mapped. roles says which of these the person is. aka lists other spellings of the name, for search. alias keeps every earlier id, the old author-… ids included. |
9,100 | OpenAlex 80.5% · last read 2026-11-04 PubMed — researchers 17.3% · last read 2026-11-04 NIH RePORTER 2.2% · last read 2026-11-16 |
JSON · CSV | OpenAlex: CC0 PubMed — researchers: NLM terms; MEDLINE metadata public NIH RePORTER: U.S. Government, public domain |
| Institutions Organisations the entity graph links something to: a person's latest affiliation, a trial they sponsor or collaborate on, an NIH grant, a trial site they host. Most are ROR records, and lat/lon are ROR's location for them. The rest are sponsors, collaborators and grantees that ClinicalTrials.gov or NIH RePORTER names and ROR does not list; they have no location. counts gives how many people, trials, grants and trial sites; aka lists acronyms and other names, for search. |
2,328 | ROR (Research Organization Registry) 93.1% · last read 2026-11-04 ClinicalTrials.gov 4.8% · last read 2026-10-16 last run failed NIH RePORTER 2.1% · last read 2026-11-16 |
JSON · CSV | ROR (Research Organization Registry): CC0 ClinicalTrials.gov, NIH RePORTER: U.S. Government, public domain |
| Grants NIH RePORTER projects found by a title-and-abstract search for psychedelic substances and ketamine, one row per core project: institute, activity code, grantee (org), total award in US dollars (usd), first and last fiscal year (fy0, fy1), active=1 while funded, and the principal investigators (pis). lat/lon sit near the grantee's ROR location where the entity graph matched one. The search also returns abuse and addiction research that names those drugs. |
407 | NIH RePORTER 100% · last read 2026-11-16 | JSON · CSV | U.S. Government, public domain |
| Policy & bills US state bills returned by LegiScan keyword searches for psilocybin, psychedelic, MDMA, ibogaine and entheogen, by status tier. LegiScan's relevance is query-relative, so full-text hits with no psychedelic term in the title come through too. psy=1 marks the rows whose own text is psychedelic-specific — filter on it rather than assuming the whole set is. |
1,537 | LegiScan 100% · last read 2026-11-17 | JSON · CSV | CC BY 4.0, attribution required |
| Industry Suppliers, funders, developers, training programs and the organizations around the field. Where the entity graph links an organization to SEC or DEA records, cik and tickers give its SEC filer number and stock tickers, and dea what DEA registration notices list it as (bulk manufacturer, importer); lic carries an Oregon manufacturer's or laboratory's OHA licence. aka lists acronyms, tickers and former names, for search. |
167 | Agent sweeps (/agarikon-sweep) 100% · last read 2026-09-09 behind schedule | JSON · CSV | Published pages only; source URL on every row |
These files were built on 2026-11-18; each source is read on its own schedule, shown in the table below. The CSV tables carry no email addresses. Where a source publishes contact details, the record's own page in the index shows every working one it lists, and the JSON files carry them because the index reads them to draw those pages. No phone or email of an Oregon facilitator is published anywhere here.
When each source was last read
This build ran on 2026-11-18. A build date can't tell you whether a source is current: a source that stopped answering stays stale however often the index is rebuilt. This table shows when each upstream source was last read successfully.
| Source | Last read successfully | Expected |
|---|---|---|
| HealingMaps | source is blocking us 2026-09-02 · 76 days ago | weekly |
| CMS Medicare Part B by provider and service | 2026-09-03 · 76 days ago | quarterly |
| Agent sweeps (/agarikon-sweep) | behind schedule 2026-09-09 · 69 days ago | monthly |
| ClinicalTrials.gov | last run failed 2026-10-16 · 32 days ago | daily |
| OpenAlex | 2026-11-04 · 14 days ago | monthly |
| PubMed — researchers | 2026-11-04 · 14 days ago | monthly |
| ROR (Research Organization Registry) | 2026-11-04 · 14 days ago | monthly |
| Oregon Health Authority — consenting licensees | 2026-11-16 · 2 days ago | weekly |
| Federal Register — DEA production quotas | 2026-11-16 · 2 days ago | weekly |
| Federal Register — DEA registration notices | 2026-11-16 · 2 days ago | weekly |
| NIH RePORTER | 2026-11-16 · 2 days ago | weekly |
| SEC EDGAR full-text search + submissions | 2026-11-16 · 2 days ago | weekly |
| LegiScan | 2026-11-17 · yesterday | weekly |
| Colorado Natural Medicine Division + DORA licence lookup | 2026-11-17 · yesterday | weekly |
Source health computed 2026-11-18 09:55 by the pipeline's own monitor, which is the one definition of “fresh” this site uses. A source counts as behind when an output is older than twice its expected cadence, or when its last run failed.
Citing this index
The index is archived on Zenodo and has a DOI, so it can be cited in a paper, a protocol or a policy submission the same way any dataset is. Cite the concept DOI: it always resolves to the newest release. Each deposit also has its own DOI, but quoting one pins a reader to a snapshot that later builds have moved past. Then give the build stamp and the day you retrieved the data:
Counts change from one build to the next. The build stamp says which meta.json a figure came from, and that file is what makes it reproducible.
Every count on this site can be recomputed from these files. Find an error and the fix lands in the next build; a changed facility row shows up in changes.json under its stable key. That is the standing invitation, not a slogan.
The downloadable mirror
Most of the files described here are also published as a git repository:
github.com/erichmayne/Agarikon-index-data.
It carries the data/ JSON, the source CSV tables behind it, the exclusion log and
SOURCES.md, with email addresses removed. Clone it if you want the index in one
download rather than file by file, or if you want its commit history.
Entity detail: the shards
The dataset files above are the index: enough about every row to list it, filter it and put it on a map. Detail that only matters once you open a record — a paper's full relationships, an author's works — lives separately, in 256 shard files under /data/e/.
A shard also holds a record's contacts as its source lists them, where the index files keep
one: phones, emails and webs on a clinic, retreat or
organisation; contacts (name, role, phone, email) on a study or trial site, with a
study's officials; emails on a person.
A shard is a JSON object keyed by entity id, not an array, so a lookup is a key access rather than a scan:
{"generated": "…", "entities": {"<eid>": {…, "edges": {…}, "links": {…}}}}
The shard for an entity is md5(eid), first two hex characters — hashed rather
than taken from the start of the id, because upstream identifiers cluster (OpenAlex work ids
run roughly in issue order) and a prefix split would produce a handful of enormous files and
hundreds of nearly empty ones. Hashing gives 256 files between 112 KB and
178 KB.
eid is not the same as id. id is the routing key this site's own links
use. eid is what the graph joins on: the native upstream identifier wherever one
exists — an OpenAlex work id, a ClinicalTrials.gov NCT number — and x:<kind>:<slug>
where none does. Both ship on every row in the index files, so you can go from a list row to
its detail without a lookup table.
Each entity carries edges (neighbour counts) and links (every
neighbour's eid, by kind). The counts are denormalized deliberately: a link that says
“12 papers” before you click it is worth more than one that makes you fetch to find
out it leads nowhere. 37,564 of 54,769 entities currently carry at least one edge: the
entity graph's links, each matched on a shared identifier, an exact name, or a name plus a second
signal, and co-authorship from OpenAlex. lb marks the links that rest on a name:
x for an exact name, t for a name plus a second signal. Coverage is a
statement about what we can prove, not about how connected the field is.
The ?v= parameter is the build stamp. A shard URL with a version
is immutable, so it can be cached indefinitely; the next build simply asks for a different
URL. Omit it and you still get the current shard, you just lose the cache guarantee.
Also served
These files also sit at stable URLs. They are not datasets in the table above, but nothing this site serves should be undocumented:
| File | What it is | Rows |
|---|---|---|
| meta.json | Build stamp, per-dataset totals and mapped counts, the country figure and its definition, distinct-operator counts, and the legal-basis provenance split. Every count this site prints is read from here — it is the file to recompute against. | — |
| quotas.json | DEA aggregate production quotas by substance and year, parsed from the Federal Register, with the document URL on every point. Drives the quota chart. | 154 |
| zips.json | US ZIP centroids from GeoNames, used only for the “trials near me” distance filter. Loaded on demand, not part of the index. | 41,488 |
| index.json | Machine manifest: every dataset, its row count and its JSON and CSV URLs, with the build stamp. Start here if you are consuming the index programmatically. | — |
| jurisdictions.json | What the law says, one line per place and drug, from the data agent's product/entities/jurisdictions.csv: each line in its source's terms, the statute, regulator or gazette page it comes from (src), any licensed program and its page, and the day it was checked (as_of). The pages state the law from these lines and no others; a place without one is named as unverified, with no claim about its law. | 26 |
| vocab.json | The search vocabulary, from the pipeline's shared drug and place vocabulary: 41 street, brand, code and plant names for the drug labels (synonyms), 8 condition abbreviations (conditions), the drug labels themselves (labels) and US state codes (us_states). Search uses it to widen a query; no record is tagged from it. | — |
| /data/ids/ | Exact-id lookup in 256 files: a DOI, PubMed id or ORCID iD mapped to the id of its record. A key reads doi:<DOI, lowercased>, pmid:<number> or orcid:<iD>, and its file is the first two hex characters of md5(key), as for the shards. Contact email addresses are here too, but only as em:<SHA-256 of the address, lowercased>, mapped to every record that lists the address; no address is published in these files. Search fetches one only for a query that looks like an id or an address. | 51,985 |
The honesty files
changes.json — the entity diff between the last two refresh runs, over treatment and research facilities, with stable keys. Zero-change days are common and real: the facility sources move on weekly and monthly cadences, and an honest diff says so instead of inventing churn. Dataset-level additions (a new well landing, say) show up in each build's counts. A machine manifest of everything here lives at /data/index.json.
coverage.json — the coverage ledger: what was swept, the spot-check confidence per sweep, and what has not been verified. Also readable as plain text at /data/coverage.txt, and in the live index under “How it's built”.
exclusions.json — every gate that removes rows, with the rule, the count and what enforces it. The numbers below are read from that file at build time.
What was removed, exactly
The relevance purge, with denominators, recomputed on every build: 839 of 2,443 harvested studies (34.3%) and 1,032 of 5,673 site rows (18.2%) were set aside as anesthesia or analgesia research that shares only a molecule with this field. Before that, 980 directory listings tagged only as wellness, GLP-1 or peptide clinics never entered the index, and 36 facility rows were quarantined at merge with written reasons. Trials match on the intervention field only, so artifacts like “DMT” as disease-modifying therapy can't enter (93 before the rule, 0 since).
Studies the classifier kept but whose title or conditions name an anesthesia or procedural term. Retained deliberately: many are psychiatric trials that merely mention a procedural agent (a psychedelic masked with propofol). The exclusion count above is what was removed, not a claim that the category is now empty. About 147 such studies are still served.
The scope rules for papers removed 22,492 works: 13,236 abstract-only mentions · 6,291 with no psychedelic term · 1,082 humanities framing · 1,510 microdosing · 373 PCP or MK-801 as the only drug.
Licensing, honestly
Provenance is mixed, so one blanket license would be wrong in both directions. Federal data (ClinicalTrials.gov, CMS, the Federal Register) is US-government public domain. Legislative data comes from LegiScan under CC BY 4.0 — attribution and a link to the LegiScan record render on every bill card and detail view. OpenAlex metadata is CC0. Directory-compiled rows are assembled from public listings whose underlying rights remain with their publishers; our compilation, structure and enrichments are released CC BY 4.0. When in doubt, cite the upstream source a row links to.