MED01 Research Brief — Historical Evidence Landscape: Medical Cannabis & Cannabinoid Research, 1960s → 2026

MED01-P2 — Modern Clinical Development, 2000 – 2017 (part 2 of 3) · CORE BRIEF (Stage 1)

FROZEN — DO NOT MODIFY

Version v1.0 · FROZEN 2026-10-03 — authorized by Danny after ChatGPT and Gemini editorial passes and confirmation checks. Frozen together with the Probe Appendix (brief-P2-probes.md, v1.0). Neither file is edited again; later clarifications go to corrections.md, identically to both researchers. Researchers receive this file as a git .md attachment, not a Drive export. The P1 brief (brief.md, v1.0 FROZEN 2026-09-20) is not modified by this file. MED01-wide evidence search cutoff (inherited, unchanged): 2026-09-20.

Governing protocol: PROTOCOL-MEDICAL v0.5 BASELINE (git 7beaff6) · base PROTOCOL v1.4 Independence rule: ChatGPT and Gemini receive this brief with substantively identical instructions. Neither sees the other’s pass until both are filed (§M1 firewall). Not medical advice. This round describes a research literature. It does not evaluate any treatment for any reader.

0. How MED01 is organized (parts and stages)

• Each part has its own frozen brief, its own two independent passes, its own firewall, verification, and cross-audit, and reaches PART LOCK before the next part’s brief is frozen. MED01 itself stays UNLOCKED under PROTOCOL-MEDICAL §M10 until P1–P3, the final synthesis, the plain-English derivatives, and the required audits are all complete.
• One MED01-wide evidence search cutoff: 2026-09-20, inherited unchanged from P1.
• All parts share one ledger (P2 claims continue from MED01-C041), the shared SOURCE_LIBRARY, and one final synthesis. P2 also gets a short interim synthesis.
• The window boundaries are administrative, not findings (§3, F1 asks where the literature itself turns).

0a. Two stages per pass

• Stage 1 — core map (this document). Answer §1–§5 from the literature. You do not have the Probe Appendix at this stage. Your Stage 1 output is filed verbatim before Stage 2 begins.
• Stage 2 — probe addendum. After Stage 1 is filed, you receive the frozen Probe Appendix (identical for both researchers) and answer it as a separate addendum. The addendum may point out omissions in your Stage 1 map, but it does not revise the filed Stage 1 output.
• The two stages exist so that the core map is built from the literature, not from a list of topics handed out in advance.

0b. How to run this pass

• Work in a fresh session with this brief as an attachment. Do not browse any shared drive or project folder, and do not draw on earlier conversations about this project. Live web, PubMed and registry searching is expected.
• State at the top: brief version, the model/product you are, the date(s) run, and whether live browsing was available.
• The researcher’s job is to find, read, and report. Formal risk-of-bias instruments, evidence-extraction files, S-IDs and claim IDs belong to the verifier.

0c. Boundary rules

• Publications. Each publication belongs strictly to the part containing its earliest public publication date for the cited version (electronic ahead-of-print or print issue, whichever is earlier, as recorded by PubMed, Crossref or the publisher): on or before 1999-12-31 → P1; 2000-01-01 to 2017-12-31 → P2; 2018-01-01 onward → P3. When electronic and print dates fall on different sides of a boundary, record both and flag the source as a boundary case. (The §3A counts use PubMed’s own publication-date index and may differ slightly at the edges; that is accepted and noted.)
• Pre-2000 studies inside period reviews. A review published 2000–2017 is a P2 source for what it concluded. Primary studies it synthesizes that were published before 2000 remain P1 evidence: report the review’s conclusion and note its underlying study dates, but do not extract those older studies as new P2 findings.
• Non-publication processes (applications, regulatory proceedings, programmes, litigation, studies completed but unpublished): allocate each dated milestone to the part in which it occurred. Earlier milestones may be stated only as the minimum context needed to understand a P2 event. For any process still open on 2017-12-31, report its documented state on that date and hand later milestones to P3. No single event has to count as the process’s “resolution.”

1. Research question (P2)

Between 2000 and 2017, what was the medical and scientific research literature on cannabis and cannabinoids — who was studying what, in which populations, with which products, routes and study designs, under which legal, regulatory and programme conditions — and what did the evidence syntheses of that period conclude by the end of 2017?

This is an evidence map, not an efficacy review. The unit of analysis is the literature. Claims are about what was studied, how, where, by whom, and what was concluded at the time — not verdicts on whether cannabis works for anything today. Present-day efficacy questions are handed off to later condition-specific rounds.

2. Scope

• In scope: human clinical and observational research; preclinical and mechanistic findings that redirected human research (labeled non-human, §M3); development, regulatory review and approval of cannabinoid medicines; medical-cannabis programmes as conditions of research; government, professional-body and evidence-assessment reports; trial registration and reporting; harms research; geography and funding.
• Out of scope: any present-day verdict on efficacy or safety; recreational-use epidemiology except where it is the harms literature of the period; cultivation; policy advocacy; legalization history except where it changed research conditions. Research published after 2017 is out of scope as evidence of what was known by end-2017. Later reviews, histories, corrections and re-analyses may be used only as retrospective sources (rule 6a) — for chronology, source discovery, integrity status or later reinterpretation — clearly separated from the period evidence.
• Populations / interventions / outcomes: not restricted — recording which ones the period studied is the point.
• Jurisdictions: no restriction. Do not treat the United States as the default.
• Literature window: primary sources published 2000-01-01 → 2017-12-31 (§0c).
• Evidence search cutoff (§M12a), MED01-wide: 2026-09-20. It governs retrospective sources; primary sources are bounded by the window.

2a. Minimum context from P1 — not to be re-researched

P1 (1960–1999) is locked. The points below are given only so that P2 events that respond to them can be understood. Do not re-verify P1 claims or re-research P1 questions; P1’s open items are resolved through P1’s own correction log. Do not use this section to choose the topics or weighting of your P2 map.

• A US single-patient compassionate-access programme for smoked cannabis had been closed to new applicants in 1992. Its pre-2000 history and patient counts are P1 matters; any 2000–2017 publication or milestone concerning it is P2 material.
• A US national-academy evidence review (1999) and a UK parliamentary committee report (1998) had each recommended further controlled research. Their content is P1; any 2000–2017 response or follow-up is P2 material.

3. Key questions

A. Volume and shape. How much was published, in which fields, journals and countries, and how did that change? Report the four canonical counts (§4a), one per period: 2000–04, 2005–09, 2010–14, 2015–17. Use published bibliometric analyses for country and field breakdowns, cited like any other source. Never smooth, interpolate or estimate a count. Say what counts cannot capture (non-English and non-indexed literature, conference abstracts).

B. Science that redirected the field. Which chemical, pharmacological or biological findings of 2000–2017 changed what human researchers could ask? For each: what was shown, in what system (in vitro / animal / human), and what it did not show at the time. Include drug-development programmes targeting the endocannabinoid system that reached humans, and their outcomes.

C. What was studied in humans. Which indications and outcomes were studied, in roughly what order, with which designs, sample sizes, products, routes, doses, comparators, durations and outcome measures? Report only what contemporaneous sources identify as motivating, enabling or constraining that research; where they are silent, say the reason is uncertain. Name each condition-level handoff to a later round.

D. Products and regulatory decisions. What was actually administered, and how did that change? For each qualifying regulatory decision identified in your documented regulator searches: regulator, jurisdiction, event date, procedure, indication as worded, pivotal trials, scheduling changes, and any later regulatory or reimbursement review in the window. Approval is not efficacy evidence (§M8).

E. Study quality and reporting. What designs dominated, and how did that change across the window? For randomized trials: size, blinding (and its integrity — psychoactive interventions unblind easily), comparator, parallel / crossover / other design, duration, attrition, whether the reported primary outcome matches the registered one. Do not retro-grade every study; characterize the body and describe the design features of the specific studies you lean on. For named medicines and keystone trials, list registered trials (ClinicalTrials.gov, WHO ICTRP, EU registers) and whether results were published or posted.

F. Conditions of research. F1 — natural phases: where does the 2000–2017 literature itself show turning points, and on what evidence? F2 — law, access and supply: which scheduling, licensing, supply, programme and funding conditions governed who could study what, by jurisdiction and date? Separate documented effects on research from asserted ones. F3 — geography and funding: where was the work done and who paid for it?

G. Harms research. How were dependence and withdrawal, psychosis and other psychiatric outcomes, cognition and adolescent exposure, driving and psychomotor impairment, pregnancy and lactation, respiratory and cardiovascular effects, acute toxicity and poisonings, and drug–drug interactions studied in this period, in which populations and with which designs? Record harms reported inside therapeutic trials (adverse events, serious adverse events, withdrawals, interactions, special populations — §M4) to the same standard as efficacy. Any statement about the balance of attention between harms and benefits needs a stated empirical basis; otherwise say it cannot be characterized.

H. What the syntheses said. Which systematic reviews, meta-analyses, national evidence assessments, regulatory assessment reports, guidelines and professional-body statements of 2000–2017 assessed this evidence? Report what each concluded as that body’s conclusion, attributed and dated, with its certainty or strength-of-evidence ratings where it gave them, and the designs it rested on — never as this program’s finding.

I. Synthesis targets. (1) An era table: indication × design × product/route × approximate volume, with period sub-columns where the shape changed. (2) A list of areas for which no human study was identified in the documented searches, kept distinct from areas found only in non-human evidence (“none identified” ≠ “none exists”). (3) Open threads handed to P3, including processes still open on 2017-12-31 (§0c). (4) End-of-era evidence-state statement — describe what the identified period evidence could support by end-2017, without creating a new certainty scale. Group findings descriptively: replicated or controlled human evidence identified · preliminary human evidence, or non-human evidence only · claims appearing in period sources for which this search did not identify traceable support. State the underlying designs and products for each, and attribute formal certainty ratings only to the period syntheses that issued them. This describes 2017, not today.

Scope guard. MED01 maps how each evidence area evolved and what level it reached by 2017. It does not perform the condition-specific review. Where an indication has many trials, characterize the body (how many, which designs, which products, what the period’s syntheses concluded), use survey lines, give full extraction only for the few studies you lean on, and name the handoff.

4. Evidence standards (PROTOCOL-MEDICAL §M3–M8 — binding)

1. Every source gets a source type and a study design (§M3): primary | sr-ma | narrative-review | guideline | official × rct | interventional-nr | observational | case | animal | in-vitro | mechanistic. For a review or report, state the designs it rests on (e.g. rests on: rct ×9, observational ×4).
2. Histories and narrative reviews are secondary. They support “author X says,” not the underlying fact. For a keystone event or study, cite the primary document as well.
3. Two tiers of study description.
   • Full extraction is required only for (a) a primary study whose own result directly supports a clinical-finding-of-the-period or harms-research candidate claim, and (b) a study you explicitly designate keystone because your narrative materially relies on its result. A synthesis-level, literature-shape, programme or regulatory candidate does not trigger extraction of every study a review cites. For dense evidence bodies, use survey lines except for the few primary studies materially relied on.
   • Full-extraction fields — trials: population (condition, diagnostic criteria, age, key inclusion/exclusion) · N randomized / analyzed · product (named), cannabinoid content and ratio · dose and titration · route · duration · comparator · primary outcome and timepoint · comparative result with interval · adverse events, serious adverse events, withdrawals · registration (rule 10a) and primary-outcome match · funder.
   • Full-extraction fields — observational studies: population and setting · exposure definition · comparison group · outcome and how it was ascertained · follow-up / time window · principal unadjusted and adjusted estimates where reported · major adjustment variables · funder.
   • Survey line for the wider literature: [Author, Year | design | N | population / indication | product / route | primary outcome | comparative result | AEs / withdrawals | registration or "none identified"].
   • Gap codes: not applicable — design where a field does not apply to that design; not reported only after reading the full text including supplements; not accessible to me when you could not; abstract read incomplete when an abstract display was truncated or failed to load. Never leave a field out; never guess.
4. Comparative results only (§M4 controlled-effect rule). For a controlled study, give the comparative estimate appropriate to the design (between-group for parallel trials; between-condition for crossover). Change from baseline alone is labeled as such; for uncontrolled or open-label studies say “uncontrolled.”
5. No drift (§M7, D1–D14). Animal ≠ human. Association ≠ cause. Symptom relief ≠ disease modification. Statistically significant ≠ clinically meaningful. Short-term ≠ long-term. “No harm reported” ≠ safe. Trial product and dose ≠ retail product. 5a. Product and route specificity. Name the exact studied product or formulation and route whenever the source permits. Distinct molecules, formulations, plant preparations, routes, doses and populations are not interchangeable. A result for one is never worded as a property of a broader product class, or of “cannabis” or “cannabinoids” generally (D4, D5, D6, D9, D10).
6. Historical-claim discipline. Distinguish documented (a dated primary document shows it) from commonly stated (repeated without a traceable source). Label each. 6a. Historical-source rule — no hindsight. Mark every publication or dated report, by the date of the cited version, period (2000–2017), prior (pre-2000; context only) or retrospective (2018 through the 2026-09-20 cutoff). All reviews, meta-analyses and evidence assessments published 2000–2017 are period sources for what they concluded at the time. Retrospective sources may verify chronology, identity and regulatory context and locate period literature, but any claim about what was known, supported, uncertain or accepted by end-2017 must trace to period evidence. Where later work changed the interpretation of a 2000–2017 finding, report the contemporaneous state and the later reinterpretation separately, each dated. 6b. Evolving records (trial registrations, regulatory dockets, legal records, web pages that are updated): do not force them into the three labels unless they have a discrete issued version. Record instead: record type · event or milestone date · cited-version or last-update date · retrieval date.
7. Harms get the same rigor as benefits (§M4 harm domains; §3G). Every efficacy finding you report from a trial carries that trial’s adverse-event and withdrawal data, or a gap code.
8. Regulatory statements name regulator, jurisdiction and event date (§M8), with a document identifier (assessment report, label, notice of compliance, Federal Register or Gazette citation, statutory instrument). When stating a status rather than describing an event, also give an explicit as-of date.
9. Publication integrity. Check and flag any source that is retracted, corrected, withdrawn or under an expression of concern (journal page, PubMed, Crossref).
10. Identifiers must resolve — never from memory. Give DOI and/or PMID for every journal source. Open each identifier live (doi.org, PubMed) and copy the resolved title next to it. If an identifier does not resolve to the paper you mean, or you could not open it, say so and give full journal/volume/page instead. Never complete or reconstruct a PMID, DOI or registry number from memory — a wrong-object identifier is a citation failure even when the paper exists. 10a. Trial registration. For every trial, give each verified registration identifier found (NCT, ISRCTN, EudraCT or other), opened live. If none is identified after searching, write “no registration identified in documented search” and name the registries searched; a missing registration is not itself a citation failure.
11. Quote by copy, not recall. Text in quotation marks must be copied from a source you opened in this session. Otherwise paraphrase and mark it as paraphrase.
12. No advice, no present-day efficacy verdicts.

4a. Search requirements (§M12b)

• Minimum sources: PubMed/MEDLINE · Cochrane Library · ClinicalTrials.gov and WHO ICTRP (EU trial registers where reachable) · the record of the relevant regulator for every regulatory decision you report · government and international-body archives relevant to your jurisdictions · at least one bibliometric analysis covering the window, if any exists.
• Core search concepts: cannabis · marijuana/marihuana · cannabinoid(s) · named plant cannabinoids · named pharmaceutical and investigational cannabinoids · endocannabinoid · therapeutic use · clinical trial / randomized · adverse effects / harms — combined with date limits 2000–2017; plus unrestricted-date searches for retrospective reviews of the period.
• Canonical volume counts (§3A). Both researchers use this identical PubMed term (the same expression the verifier used for P1, so the series is continuous): ("Cannabis"[Mesh] OR "Cannabinoids"[Mesh] OR "cannabis"[tiab] OR "marijuana"[tiab] OR "marihuana"[tiab] OR "cannabinoid"[tiab] OR "cannabinoids"[tiab]) URL-encoded, with only the dates changing per period (START/END = 2000/01/01–2004/12/31, 2005/01/01–2009/12/31, 2010/01/01–2014/12/31, 2015/01/01–2017/12/31): https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&term=(%22Cannabis%22%5BMesh%5D%20OR%20%22Cannabinoids%22%5BMesh%5D%20OR%20%22cannabis%22%5Btiab%5D%20OR%20%22marijuana%22%5Btiab%5D%20OR%20%22marihuana%22%5Btiab%5D%20OR%20%22cannabinoid%22%5Btiab%5D%20OR%20%22cannabinoids%22%5Btiab%5D)&datetype=pdat&mindate=START&maxdate=END&rettype=count For each period report: the full clickable URL with dates filled in · the count · the run date · PubMed’s query translation (shown on the PubMed web page for the same search, or in the ESearch QueryTranslation field when rettype=count is removed). Any other queries you run are exploratory, reported in the search log, and are not the canonical counts. A count without a working URL is reported as unreproducible.
• Search log (required, at the end of the pass): tools and databases actually used · exact query strings (as ESearch URLs where PubMed) · dates run · result counts · what you could not access. If you relied on training knowledge rather than a live search for any item, say which.
• This is not a systematic review and must not be described as one.

5. Required output format (Stage 1)

1. Narrative answer to A–I, in that order, with inline citations.
2. Candidate claims — numbered CAND-01…, normally 20–30; hard maximum 35. The maximum is not a target: prefer fewer well-supported, non-duplicative candidates that together cover the A–I map. Each: one-sentence claim · claim type (literature-shape | discovery | clinical-finding-of-the-period | product-regulatory | programme | harms-research | synthesis-conclusion | absence) · supporting source(s) with identifiers and resolved titles · source type × study design · period | prior | retrospective · population / product / route / dose / comparator / outcome where the claim concerns a study · your confidence · what would change your mind.
3. Era table (I-1) and not-identified list (I-2).
4. Regulatory decisions table: product · composition · route · regulator · jurisdiction · event date · indication as worded · pivotal trials · document identifier.
5. Reference list with identifiers; for each: verified live (resolved title copied) | from memory — unverified · access level full text | abstract only | secondary report · period | prior | retrospective (or the rule 6b fields).
6. Self-flagged weak points: citations you are unsure of, claims resting on a single secondary source, counts you could not reproduce, abstracts read incompletely, boundary cases (§0c).
7. Search log (§4a).
8. State at the top which brief version you worked from (§0b).

6–7. Archive probes and public claims → Probe Appendix (Stage 2)

Not part of Stage 1. Delivered after your Stage 1 output is filed (§0a).

8. What “done” looks like

A reader finishes P2 knowing what the 2000–2017 literature consisted of — how much, which designs, which products and routes, which populations — which cannabinoid medicines reached regulatory decisions, by whom and on what trials, what the period’s own syntheses concluded with their certainty ratings, how harms were studied, and which common claims about the period are documented. The ledger gains roughly 25–40 verified claims (MED01-C041 onward; sources continue the shared S-series from S-308). Nothing in it reads as a statement about what cannabis does for a patient today.

9. Known limits of this round as a protocol test

P1 exercised citation identity, the two-axis source rules, historical discipline and the cross-audit, and only lightly the trial-level controls. P2 is their first real test: RoB 2 variants (parallel and crossover), the design-aware controlled-effect rule (§M4), the registry cross-check for named products and keystones, the harm domains in therapeutic trials, JBI appraisal of observational cohorts, and AMSTAR 2 for the period’s systematic reviews — all applied by the verifier, not the researchers. Expected strains: the volume of the window, the temptation to perform condition reviews (scope guard), regulatory records spread across many agencies, and many sources being accessible only as abstracts (§M9 cap). New in P2 and to be assessed in the MED01 postmortem: whether the two-stage pass (§0a) works as a guard against anchoring, and whether the extraction triggers and candidate cap are workable at this volume.

Editorial history (pre-freeze)