About this project
The Fadiman
Knowledge Atlas
A proposed living knowledge infrastructure for 58 years of recorded public output — a 1968 KQED documentary through to April 2026 — by the person who coined the Fadiman Protocol, popularized modern microdosing, and has been professionally involved with psychedelics since 1961.
The problem this exists to solve
James Fadiman has been thinking, speaking, writing, and teaching about psychedelics since 1961. He was a graduate student at Stanford when his former Harvard undergraduate adviser — Richard Alpert, who later became Ram Dass — introduced him to psilocybin in Paris. He guided Stewart Brand's first LSD trip in 1962. He participated in some of the last legal psychedelic research sessions before the government shut the field down. He was there at the beginning. He watched the entire field get buried. And he lived long enough to watch it come back.
The catalog includes Fadiman's 2011 and 2025 books and his 2019 paper with Sophia Korb. The paper describes an uncontrolled, self-selected reporting project involving more than 1,000 people in 59 countries. It does not establish efficacy, safety, or suitability for any person. Primary record: PubMed 30925850 ↗
He is 87 years old. He is doing two to three recorded interviews per month. He is still explaining the same foundational ideas to new audiences, because there is no other mechanism for those audiences to find what he has already said.
The oldest entry is Drugs in Our Culture, the 1968 KQED documentary he scripted and introduced. The newest is a 61-minute YouTube interview published April 25, 2026. That is 58 years of recorded thought on a single manually maintained page with no search, no topic index and no transcripts — and, as of this link check on 29 August 2026, 50 entries whose recording has been taken down and 8 more with no public URL at all.
Sources on this page were reviewed on 17 August 2026 and every link was re-checked on 29 August 2026. The public prototype helps people locate where historical topics appear in a catalog of recordings. It does not consolidate clinical answers, evaluate an individual's circumstances, or recommend a substance, dose, schedule, taper, interaction, or treatment. A catalog match is a discovery aid—not evidence that a statement is accurate.
What has been done — and what has been deliberately held
The complete public archive has been catalogued entry by entry, and every link in it has been checked. The gap analysis has been run. The three books have been mapped as a separate evidentiary layer. The annotation schema has been fully designed and specified. The technical infrastructure — Vercel, Supabase pgvector, AssemblyAI, OpenAI embeddings, RAG query interface — is already live and running on other active projects.
The step that has not been taken is ingestion: downloading the audio, running transcription, putting his words into a system. That step has been deliberately withheld. The material is his. Beginning without his awareness felt like the wrong place to start something designed to honor sixty-plus years of work.
What the scope work covered: 132-entry archive catalogued; 10 verified gaps found; books mapped; 20+ academic papers inventoried; 12-field annotation schema specified; 8-layer technical architecture written with cost estimates; intellectual timeline 1961–2026 reconstructed.
Corpus search demo
This demonstrates how the Atlas will answer questions about Fadiman's body of work — grounding every answer in the catalogue with source citations. The full build would use transcribed audio and embedded text; this demo uses Claude's knowledge of Fadiman's published statements together with the catalogue entries above.
Live annotation demo
Paste an archival excerpt below. Local, deterministic heuristics flag literal terms and return matching catalog source IDs. This aid does not infer biological mechanisms, verify evidence, assess risk, or provide medical analysis.
34-year output timeline
Every catalogued entry plotted by year. The two surge periods — 2011 (The Psychedelic Explorer's Guide launch) and 2025 (Microdosing for Health launch) — are visible. Click any bar to see entries from that year.
Corpus explorer
Filter, search, and sort all 132 catalogued entries. This is the data layer the Atlas indexes.
| # | Date | Type | Host / Interviewer | Show / Organization | Runtime | Co-guest | Link |
|---|
The complete archive inventory — all 132 entries
Every entry from jamesfadiman.com/watch-hear.html, catalogued in full. Entries are linked where a URL was confirmed. Dead and missing links are flagged. This table is the inventory receipt — proof that the scoping work was done at the level of individual entries.
| # | Date | Type | Host / Interviewer | Show / Organization | Runtime | Co-guest | Status |
|---|
Gap analysis — verified appearances not on the archive
The archive is curator-maintained, lags publication by months, and structurally omits third-party uploads, conference re-uploads, and premium/gated content. Cross-referencing against YouTube, Spotify, Apple Podcasts, IMDb, and Netflix confirmed the following. Each entry has been individually verified.
| # | Title / Episode | Host / Channel | Platform | Date | Runtime | Reach & Verification |
|---|---|---|---|---|---|---|
| G1 | How to Change Your Mind — Chapter 1: "LSD" | Dir. Lucy Walker; Michael Pollan host | Netflix (worldwide) | Jul 12, 2022 | ~55m | Netflix global release, dubbed into 10 languages. Fadiman credited as "Self — Psychologist" on IMDb. He describes the 1966 creativity study: "we had 48 problems and 48 satisfactory solutions." Almost certainly his highest-viewership appearance. Confirmed via IMDb full cast listing. |
| G2 | "The Genesis of Microdosing" — Ep. 116 of The Psychedelic Podcast | Paul Austin / Third Wave | YouTube + Apple + Spotify | Feb 28, 2021 | ~1h 27m | Third Wave is one of the highest-traffic psychedelic media platforms. Covers origins of the Fadiman Protocol, citizen science methodology, and Your Symphony of Selves. YouTube URL live and confirmed. |
| G3 | "The Power of Less: Microdosing for Mood, Energy, and Clarity" — Ep. 318 | Paul Austin / Third Wave (w/ Jordan Gruber) | YouTube + Apple + Spotify | Aug 26, 2025 | ~1h | 2025 book-launch interview. The linked source contains health-related discussion that the Atlas has not independently validated. Published four months after the book; URL confirmed live. |
| G4 | "Microdosing (Part 1) — James Fadiman and the Fundamentals" | Greg Kubin & Matias Serebrinsky / Business Trip (PsyMed Ventures) | Spotify / Podcast | ~2020–2021 | — | Episode on a psychedelic-entrepreneurship podcast. Topic descriptions are catalog metadata, not validated health claims. Referenced in the linked episode material. |
| G5 | "Microdosing Q&A with James Fadiman" | Lynn Marie Morski / Psychedelic Medicine Podcast | Apple Podcasts + Spotify | Aug 19, 2020 | ~1h 32m | Listener Q&A format containing health-related historical discussion. The Atlas does not validate or operationalize that discussion. Distinct from the Dec 2020 Morski/Bramlage episode; linked source URL confirmed. |
| G6 | MINDSET Lecture Series: "Microdosing: Uses, Benefits, and Latest Research" | MINDSET Professional Education | YouTube | Jun 8, 2022 | ~1h | "Designed for a professional audience of researchers, clinicians, and practitioners." Distinct register from general public interviews — clinical framing throughout. YouTube URL confirmed live. |
| G8 | "Microdosing: A Conversation with James Fadiman" — SAND18 audience Q&A | Science and Nonduality (SAND) | YouTube | 2020 (upload; SAND18 footage) | — | Audience Q&A companion to G7, uploaded separately. "James Fadiman takes questions from the audience at SAND18 US." Neither this nor the main talk appears on the archive. |
| G9 | "Try Psychedelics. Access Transcendence." — Big Think premium segment | Big Think | YouTube (Big Think Membership) | Feb 19, 2024 | ~6m | Premium Big Think content — distinct from the two 2023 Big Think reels already on archive (entries #27, #28). Produced separately for the membership tier. YouTube URL confirmed live. |
| G10 | "Microdosing Results, Limitations, and Emerging Questions" | OPEN Foundation (European psychedelic research org) | OPEN Foundation website + YouTube | ~2020 | — | Europe's leading psychedelic research organization. A dedicated Fadiman webinar for a European research audience — distinct framing from North American appearances. Event page confirmed on open-foundation.org. |
| G11 | Inside LSD — National Geographic documentary | National Geographic | Television | 2009 | ~50m | National Geographic television documentary. Fadiman appears discussing 1960s problem-solving research. Only known pre-2011 mainstream television appearance — predates the archive's systematic coverage. Referenced in Fadiman's own materials. Not on archive. |
The annotation schema — evidence-boundary specification
This proposed schema records what a source explicitly says. It must preserve the source date, speaker, quotation span, and review status. Public tooling does not translate historical terminology into dosing guidance, infer biology, or convert a topic mention into an indication.
| Field | Type | Controlled values | Key logic |
|---|---|---|---|
| substance | enum | LSD · psilocybin · MDMA · ibogaine · cannabis · ketamine · DMT · 5-MeO-DMT · mescaline · multiple · general_unspecified | Match compound or unambiguous synonym. Lower confidence when discussed via quoted third party. |
| historical_dose_language | verbatim text span | source wording only · unspecified | Preserve only the language present in the source. The public Atlas does not map amounts to bands, schedules, or recommendations. |
| mechanism_language | verbatim text span | explicit wording · none stated | Record explicit source language only. Do not infer a biological mechanism or scientific validity. |
| health_topic_mention | controlled topic + text span | topic named by source · none | A mention is not an application, indication, diagnosis, or treatment claim. |
| evidence_quality | enum | peer_reviewed_study · fadiman_citizen_science · anecdote · personal_experience · expert_opinion · speculation · historical_account | Most important field for scientific credibility. fadiman_citizen_science is distinct from generic anecdote. Default to human review when confidence <0.7. |
| claim_type | enum | empirical_claim · recommendation · definition · historical_narrative · personal_anecdote · philosophical · hedge · opinion | Classify the speech act, not the content. |
| confidence_flag | enum (speaker certainty) | high_certainty · moderate · low_speculative · explicit_uncertainty · disclaimed | Distinct from evidence_quality — this is Fadiman's expressed certainty. Lexical hedging detection. Critical for trajectory visualization. |
| temporal_marker | enum | early (1961–1979) · middle (1980–2009) · late (2010–2026) | Derived primarily from source date. When recounting early events in a late interview, tag both recording_era and referenced_era. |
| audience_type | enum | general_public · researchers · clinicians · practitioners · founders · students · policy_legal | Infer from host/venue metadata then confirm from register of language. |
| protocol_specificity | enum | specific_protocol · general_observation · comparative | Concrete numbers/schedules → specific_protocol. Comparative: Fadiman vs. Stamets vs. intuitive. |
| citation_present | boolean + span | true · false | TRUE when a named study, author, journal, or dataset is referenced. "Studies show" → FALSE. |
| contradiction_flag | enum + linked IDs | none · intra-corpus_tension · evolution · self-correction | Computed in second pass via nearest-neighbor embedding. Always human-reviewed. Genuine evolution tagged evolution, not contradiction. |
Technical architecture — 8-layer specification
Layer 1 — Ingestion
YouTube/Vimeo (bulk, idempotent): yt-dlp -x --audio-format mp3 --audio-quality 0 --write-info-json --embed-metadata --download-archive archive.txt -o "%(upload_date)s_%(uploader)s_%(title).80s_%(id)s.%(ext)s" — --download-archive makes re-runs idempotent, critical for CI/CD. Podcast RSS: Python feedparser, deduplicate on guid. Books: EPUB → ebooklib + BeautifulSoup. Diarization: AssemblyAI speaker_labels=True as primary — bundles diarization + transcription + word-level timestamps in one API call.
Layer 2 — Transcription
Whisper large-v3 via faster-whisper (CTranslate2, ~4× speed) — required for period audio (1992 APA recording, 1997 transpersonal, Russian TV overdub, documented audio dropouts in the 2018 Toronto webinar). AssemblyAI API alternative: $0.17/hr (base + diarization) → ~$20–21 for full corpus. Modal A10G GPU for Whisper: ~$15–20. Speaker ID: longest-cumulative-talk-time rule + pyannote voiceprint reference from 3–4 known-Fadiman clips.
Layer 3 — Chunking
Hybrid semantic-within-speaker-turn. Target: ~400 tokens (300–500 band), 15% overlap. MinHash/SimHash deduplication for repeated stories (the "48 problems, 48 solutions" session appears across dozens of interviews). story_cluster_id links instances for evolution-layer comparison.
Layer 4 — Vector Database
OpenAI text-embedding-3-large (3072-dim, $0.13/1M tokens, <$5 one-time for full corpus). Supabase pgvector — keeps vectors, annotation metadata, and auth in one Postgres instance, enabling SQL WHERE filters combined with vector search. HNSW index: m=16, ef_construction=64.
Layer 5 — Annotation Pipeline
A future annotation pipeline may propose structured metadata, but no generated health or scientific interpretation may be auto-accepted. Source identity, date, quotation span, reviewer, limitations, and review status are required; health-related interpretation requires appropriately qualified independent review.
Layer 6 — Query Interface
Proposed retrieval: hybrid keyword and vector search with metadata filters, source IDs, timestamps, and an extractive-only result mode. Health-related queries return source passages and limitations—not generated answers. Topic maps may aid discovery but must expose source coverage, uncertainty, and review status. Rate limits, audit logs, and abuse monitoring are required before any inference endpoint ships.
Layer 7 — Evolution Layer
Every chunk carries a source date. "How has Jim's thinking on X changed from 2011 to 2025?" → filtered retrieval bucketed by year, structured timeline with inflection points and confidence trajectory. Topic emergence detection: first-occurrence by source_date per topic cluster. All driven by SQL aggregations over the same chunks table.
Layer 8 — Infrastructure & Deployment
Next.js on Vercel · Supabase pgvector · Upstash Redis (query caching TTL 24h, rate-limiting) · OpenAI · AssemblyAI · Modal (GPU transcription) · Cohere (reranking). Daily Vercel Cron re-scrapes watch-hear.html + runs yt-dlp --download-archive + checks RSS; new GUIDs auto-enqueue full pipeline.
Fadiman & Korb (2019). An uncontrolled, self-selected reporting project with more than 1,000 participants in 59 countries. It can describe reported experiences; it cannot determine efficacy, safety, causation, or individual suitability. PubMed 30925850 ↗
Bershad et al. (2019). A placebo-controlled low-dose LSD study in healthy volunteers with a narrow experimental scope. It is not treatment evidence and does not support personalized use. PubMed 31331617 ↗
Polito & Stevenson (2019). An observational study of self-reported microdosing. Observational associations do not establish therapeutic benefit or safety. PubMed 30726251 ↗
Scope rule. Atlas records and search results are historical discovery metadata. Health-related statements remain attributable to their original speakers and dates; they are not endorsed or converted into advice here.
No current procurement budget or vendor-price estimate is represented on this page. Service prices, model availability, quotas, and terms change. Any future build must obtain dated quotes and complete security, privacy, accessibility, and legal review before procurement.
Who scoped the prototype
© 2026 Scoped by R.N. Collins in Honolulu, Hawaii. An interdisciplinary background in neuroscience, law, science communication, and software helps identify where source attribution and professional review are required. Credentials do not make the Atlas a clinical authority, and no individual credential substitutes for independent medical, legal, privacy, security, or accessibility review.
Every statement the finished system produces should stay linked to its original recording, speaker, date, quotation span, and review status. The public prototype deliberately limits itself to catalog discovery and deterministic language flags.
Health, scientific, legal, and safety claims require review by appropriately qualified independent professionals. The system must surface uncertainty and disagreement instead of generating a clinical conclusion.
This is an archival discovery tool, not medical guidance. It is not a dose calculator, contraindication screen, interaction checker, diagnosis tool, treatment tool, or emergency resource. It does not personalize output and must not be used to decide whether, when, or how to take, stop, combine, or change any substance or medication.
Catalog records and heuristic flags may be incomplete, outdated, or wrong. Follow source links, check publication dates, and seek appropriately licensed professional care for health decisions. In an emergency, contact local emergency services.
Where things stand and what comes next
This public build is a catalog, filterable explorer, timeline, deterministic term flagger, and local source-ID search prototype. It does not contain full transcripts, a production ingestion pipeline, verified semantic search, or a clinical knowledge system. The larger ingestion and review system remains unbuilt.
The step that has not been taken is ingestion: downloading the audio, running transcription, putting his words into the system. That step has been deliberately withheld. The material is his. Beginning without his awareness felt like the wrong place to start something designed to honor sixty-plus years of work.
To take this further — with his input on what he most wants it to do — the build is ready to begin. If the timing is not right, the scope work and infrastructure exist either way.
R.N. Collins · collins.ra@northeastern.edu · linkedin.com/in/rn-collins · Honolulu, HI