What the Record Shows: Medical Cannabis and Cannabinoid Research, 2000–2017

MED01 interim technical synthesis — Part 2 of 3

Version: v1.0 — SIGNED OFF by Danny 2026-10-12 · P2 LOCK 2026-10-14 (text = v0.2), with corrections #5–#16 of 2026-10-13/14 after the §M10a cross-audit (see corrections.md). Trail: v0.1 Claude → ChatGPT editorial pass (13 findings, all accepted; raw, dispositions) → v0.2. Built on ledger Part 2 v1.0, signed off by Danny 2026-10-10 (claims MED01-C041–C085; end-of-era C054 placement corrected by logged correction #4). Next: the plain-English pair, then the §M10a cross-audit. Any later change is a logged correction. Standing rule (carried from R07 and P1): this synthesis is never more certain, and never simpler, than the ledger. Every substantive sentence carries a claim ID, so a reader can trace sentence → claim → source → extracted result. Status of this document: interim. It synthesizes P2 (2000–2017) and hands open threads to P3 (2018 → search cutoff). The final MED01 synthesis is assembled after P3. Components still held for full-text or official-document checks are marked [HELD] where they matter; the list is in §13.

This is not medical advice. Everything below describes a research literature: what named studies and bodies reported about the populations they studied, as the evidence stood at the end of 2017, located through the 2026-09-20 search cutoff. Nothing here says what cannabis or any cannabinoid does for a patient today.

1. The question, and how it was answered

The frozen brief asked what the medical and scientific literature on cannabis and cannabinoids was between 2000 and 2017: what was studied, in whom, with which products and designs, under what legal and regulatory conditions, and what the period’s syntheses concluded. It is an evidence map, not an efficacy review. It grades no condition, and it describes 2017, not today.

Two AI researchers worked independently from the same frozen brief in two stages. Each filed a core map before seeing a second-stage appendix of probe questions and public claims. Neither run was fully clean:

• ChatGPT ran with account memory on.
• Gemini’s first run reached the shared Drive and saw part of the probe appendix. That run was discarded, and Gemini was rerun from a separate account.
• Gemini’s identifiers were badly unreliable: 27 of the 62 PubMed IDs it cited pointed to unrelated papers or did not exist, every one labelled “verified live”.
• Gemini also reported four publication counts that it later admitted were estimated, not retrieved.

So agreement between the two researchers is treated as agreement on topics, never as independent confirmation of an identifier or a figure. Every identifier and number in the ledger was re-resolved by the verifier against PubMed, registries or official records. Where Gemini’s figure differed from the source, the source’s figure is used.

The verifier also ran a completeness probe, a registry cross-check and official-record checks. These surfaced several keystone sources that neither researcher cited, including:

• the only large disease-modification trial in the window;
• the two largest cancer-cachexia trials;
• the principal period review of adverse effects of medical cannabinoids;
• seven Cochrane reviews.

These are marked verifier-found in the ledger and are mandatory targets for the cross-audit. The result is a 45-claim signed ledger. This synthesis reads it in order.

2. How much was published, and in what shape (C041–C043)

The canonical PubMed/MEDLINE query returns 4,335 indexed records for 2000–04, 7,451 for 2005–09, 10,316 for 2010–14 and 8,599 for 2015–17: 30,701 in total (C041). Output rose throughout. The three-year final period had the highest annual rate, about 2,866 records a year against about 2,063 in 2010–14 (C041). The query is the same one used for P1, so the 1960–2017 series is continuous. The counts are of indexed records, not of research effort, and they under-represent non-English and non-indexed work (C041).

A retrospective bibliometric analysis reports that, over 2000–2017, PubMed publications on cannabis grew about 4.5-fold and on medical cannabis almost 9-fold, against 2.5-fold growth for PubMed overall. Psychiatry was the largest field. The indications with the most medical-cannabis papers were HIV, chronic pain, MS, nausea and vomiting, and epilepsy (C042). Those fold-changes depend on that study’s own search terms and are used for shape only (C042).

Read alongside the dated record, the period shows three turning points (C043):

1. 2003–2007: the first large randomized programmes of cannabis-based medicines, and the first marketing authorisation of a cannabis-derived product (Canada, 2005).
2. 2006–2009: the rise and withdrawal of the CB1 antagonist rimonabant.
3. 2013–2017: the paediatric CBD epilepsy pathway, ending in a controlled trial in 2017, alongside the first national evidence assessment of the era (NASEM, January 2017).

The boundaries are editorial, not an observed variable (C043).

3. Science that redirected the field (C044–C047)

Nothing in this section is evidence of a treatment effect in patients. It is laboratory, animal and healthy-volunteer work.

• Endocannabinoids as retrograde messengers. Endocannabinoids were shown to carry signals backwards across synapses in rat hippocampus. Mice lacking the enzyme FAAH showed raised brain anandamide and supersensitivity to it, which made enzyme inhibition a drug target (C044).
• CBD at the CB1 receptor. In 2015, CBD was reported to act as a negative allosteric modulator of the CB1 receptor in cell systems: one of several proposed mechanisms for its distinct pharmacology, and in-vitro only (C045).
• “Clinical endocannabinoid deficiency.” This was proposed in 2004 as a hypothesis linking migraine, fibromyalgia and irritable bowel syndrome to low endocannabinoid signalling. Reviews in 2014 and 2016, the latter by its originator, gathered supportive but indirect data. No independent confirmatory test was identified (C046). The ledger describes the hypothesis’s status in 2017, not whether it is true.
• Constituents compared head to head. Studies comparing whole-plant cannabis with matched THC found similar subjective effects in healthy users (C047). The one therapeutic head-to-head comparison identified found no difference between cannabis extract, THC alone and placebo in cancer cachexia (C047, C054). Experimental studies in healthy volunteers found that CBD can oppose some effects of THC, including THC-induced paranoia (C047). Constituent “synergy” was argued in a 2011 review as a hypothesis (C047). No study testing whether a whole-plant preparation is therapeutically superior to a matched single compound was identified (C047).

4. What was studied in humans (C048–C060)

Findings are reported as the period reported them. Each trial’s primary outcome is stated where the abstract states it, because a secondary result presented as the finding is a known distortion in this literature (C051).

Spasticity in MS was among the most extensively trialled indications, and its trials repeatedly split between what patients reported and what clinicians measured (C048). CAMS (2003; 630 treated) found no effect of oral THC or cannabis extract on its primary clinician-rated Ashworth score (p=0.40), while patient-reported spasticity improved (C048). The nabiximols trials reported mixed primary results; one ITT primary was not significant (C048). An enriched-enrolment design, which randomized only early responders, reported a significant difference (P=0.0002), and that effect applies to responders (C048). Smoked cannabis reduced Ashworth scores in a 37-patient crossover (C048). Whether blinding held in these trials is partly unresolved: CAMS reported “a degree of unmasking”, and its correct-guess rate is [HELD] for full text (C048, C066).

Pain produced the largest and most mixed body of trials:

• Central and MS-related pain: small dronabinol and nabiximols trials reported benefit. The largest dronabinol trial (240 patients) found no difference on its primary outcome; it completed in 2009 and was published in 2017 with a conclusion that led on safety (C049, C067).
• Smoked or vaporised cannabis in neuropathic pain: short crossover and parallel trials reported analgesia over days to weeks. One trial found a 34% vs 17% median pain reduction. Another found a 0.7-point difference at 9.4% THC. A pooled individual-patient analysis estimated short-term relief for about one patient in five or six (C050). These trials were small, used research cannabis of fixed potency, and over-represented experienced users (C050).
• Nabiximols in cancer and peripheral neuropathic pain: four trials gave mixed results. One reported benefit for THC:CBD, but not for THC alone, on its primary pain outcome; a second reported benefit in allodynic peripheral neuropathic pain; a third, in diabetic neuropathy, found no difference from placebo; and a fourth missed its primary 30%-responder endpoint (P=.59), although secondary continuous-responder analyses favoured the low and medium doses (C051).
• Nabilone in other chronic pain: nabilone was inferior to dihydrocodeine on the primary outcome, with more side effects (C052). It improved the primary pain score in a 40-patient fibromyalgia trial, with more side effects per person (C052). A 2016 Cochrane review found no convincing evidence for nabilone in fibromyalgia and noted low tolerability (C052).

Chemotherapy-induced nausea and vomiting. The period mostly re-synthesised trials run before 2000 (C053). The one new comparison with a modern 5-HT3 antiemetic reported similar response rates for dronabinol and ondansetron in delayed nausea (54% vs 58%), with no added benefit from combining them; the trial was small and not designed to show equivalence (C053). Applicability to current antiemetic regimens is limited (C053).

Appetite and cachexia. In cancer, controlled trials showed no advantage over standard care or placebo. Megestrol gave better appetite palliation than dronabinol (469 patients), and extract, THC and placebo did not differ (C054). In HIV:

• a laboratory study found dronabinol and smoked cannabis both increased food intake in experienced users;
• a 21-day trial found no adverse effect on viral load or immune markers;
• a 2013 Cochrane review judged efficacy evidence lacking (C054).

Paediatric treatment-resistant epilepsy became a cannabinoid target through a documented sequence (C055):

1. A 2014 Cochrane review found four small CBD trials and no reliable conclusions.
2. A parent survey and a retrospective series reported uncontrolled change.
3. An open-label expanded-access programme (patients aged 1–30) reported a 36.5% median reduction in motor seizures, with adverse events in 79% and serious adverse events in 30%.
4. The first randomized trial (Dravet syndrome, 120 patients, sponsor-funded) reported monthly convulsive seizures falling from a median 12.4 to 5.9 on CBD vs 14.9 to 14.1 on placebo (adjusted difference −22.8 percentage points; P=0.01), with more adverse events.

Open-label studies documented interactions with clobazam and other antiepileptics (C055). The FDA approval that followed falls in 2018 and belongs to P3 (C055, C063).

Other indications had single or pilot trials, each reported by its measured outcome rather than as efficacy for the condition (C056–C058):

• CBD vs amisulpride in acute schizophrenia: similar improvement, with a better side-effect profile for CBD (C056).
• THC in Tourette syndrome: small trials; a Cochrane review found not enough evidence (C056).
• A single-dose CBD experiment in 24 treatment-naïve patients with social anxiety, using a simulated public-speaking challenge; an experimental challenge, not a treatment trial (C056).
• Nabilone for PTSD nightmares: a small pilot (C056).
• Dementia: no evidence of benefit in a Cochrane review, and no effect of low-dose THC on the primary neuropsychiatric outcome in a randomized trial (C056).
• Crohn’s disease: the smoked-cannabis trial missed its primary outcome, remission (5 of 11 vs 1 of 10, P=.43), while reporting clinical response, and a low-dose CBD trial found no benefit (C057).
• Glaucoma: a single 5 mg THC dose lowered intraocular pressure for about two hours in six patients, while CBD did not (C058).

Cancer and disease modification. The only human anticancer study identified was a phase I safety pilot in nine patients with recurrent glioblastoma. It drew no efficacy conclusion; other human reports were case reports (C059). The one large human disease-modification trial, CUPID (498 patients with progressive MS, three years of oral THC), found no effect on progression (HR 0.92, 95% CI 0.68–1.23) (C060). Neither researcher identified CUPID (C060).

5. Products and regulatory decisions (C061–C065)

The period’s studies used distinct products that cannot be swapped for one another (C061):

• oromucosal THC:CBD extract spray (nabiximols);
• oral cannabis extract, dronabinol, nabilone and purified CBD;
• CBD-enriched oils of variable composition;
• smoked or vaporised research cannabis of fixed potency.

Findings for one do not transfer to another, and none is “cannabis” in general (C061).

Regulatory approvals were specific to a product and an indication, and approval is not efficacy evidence (C062–C063):

• Canada: Sativex received conditional approvals for MS neuropathic pain (2005-04-15) and advanced cancer pain (2007-08-01) (C062). Retrospective: both conditional indications were removed in 2019 after confirmatory studies did not confirm benefit (C062).
• UK and Spain: marketing authorisation on 16 June 2010 (UK, MS spasticity) and 20 July 2010 (Spain) (C062).
• US: an oral dronabinol solution was approved in 2016 and placed in Schedule II in 2017. The 2006 FDA action on nabilone was a labeling supplement, not an approval. No plant-derived cannabinoid medicine was FDA-approved within the window (C063).

Drugs aimed at the endocannabinoid system reached patients with adverse outcomes (C064):

• The CB1 antagonist rimonabant was authorised in the EU in 2006 for weight loss and withdrawn in January 2009 after psychiatric-risk review.
• One FAAH inhibitor was stopped for futility in osteoarthritis pain.
• Another, BIA 10-2474, caused an acute neurological syndrome in three of four reported participants in a 2016 phase I trial, one of whom became brain dead; the French regulatory inquiry reported one death and four other participants with severe neurological disorders (convergence, verifier read pending); off-target enzyme inhibition was later demonstrated.

Patents of the period are records of claimed inventions, not evidence that the uses work (C065). They included:

• a US government patent on cannabinoids as antioxidants and neuroprotectants (granted 2003);
• company patents on CBD for partial seizures (2015) and for FIRES (2016);
• a university-hospital patent on CBD for Type II diabetes (2011).

Both researchers mischaracterised at least one patent’s scope (C065).

6. How good the evidence was, and what went unreported (C066–C067)

The controlled evidence was numerous but mostly small and short. A 2015 review found 79 trials with 6,462 participants, only four at low risk of bias (C066). Crossover and enriched-enrolment designs were common (C066). Blinding was a known concern for psychoactive interventions. An analysis of three nabiximols trials found no evidence of widespread unblinding (C066), while CAMS reported some unmasking (C048).

The registry cross-check is a disclosure control, not an estimate of publication bias (C067). The verifier worked down from ClinicalTrials.gov in stages:

For the other 11, no publication was identified in the documented searches (C067). They include a 103-patient dronabinol trial in HAART-associated symptoms and three nabilone trials by one sponsor (C067). One matched trial, a 72-patient nabiximols trial in children with cerebral palsy completed in 2017, was published only in 2020 (C067). The WHO and EU registries were not searched (C067). (The v1.0 table started from a looser filter, and its “32 patient trials” included 12 non-therapeutic studies; corrected 2026-10-13.)

Reporting problems also appear in published trials (C067):

• a trial registered only after it ended;
• a mismatch between a trial’s registered and published primary outcome, [HELD] for registry history;
• a negative primary published eight years after completion.

7. The conditions under which the research was done (C068–C069)

In the United States, research cannabis came from a single federally contracted source throughout the window (C068). A court challenge to the refusal to register another manufacturer failed in 2013 (C068). The DEA announced in August 2016 that it would accept further applications, with no new registrant within the window (C068). Period commentators described Schedule I as an obstacle to research, and NASEM’s 2017 report recommended addressing regulatory and supply barriers (C068). A state-funded California centre nonetheless ran controlled trials (C068). The documented effects are limited to the cases named; no estimate of research forgone is supportable (C068).

National programmes produced some published data (C069):

• Dutch pharmacy records showed 5,540 patients receiving prescription cannabis in 2003–2010.
• A Canadian survey reported substitution for prescription drugs.
• A one-year Canadian cohort found more non-serious, but not more serious, adverse events among cannabis users than controls.

The legal bases and start dates of the programmes are [HELD] for primary legal texts (C069).

8. Harms research (C070–C077)

Harms were investigated through observational cohorts, systematic reviews, experimental studies and therapeutic-trial safety data (C070–C077). Findings on longer-term outcomes — psychosis, cognition, lung function, pregnancy — are associations that did not establish causation (C070, C071, C074, C075). Withdrawal, acute driving impairment and adverse events during therapeutic use were studied more directly, and each is read within its own design (C072, C073, C077):

• Psychosis. Cohort and meta-analytic evidence associated heavier cannabis use with later psychotic outcomes, with a dose–response gradient (C070). In 50,087 Swedish conscripts, use on more than 50 occasions was associated with schizophrenia (adjusted OR 6.7) (C070). A 2007 review reported a pooled OR of 1.41 for any psychotic outcome among ever-users (C070). NASEM found substantial evidence of a statistical association (C070).
• Cognition. In the Dunedin cohort, persistent users showed neuropsychological decline concentrated among those who started in adolescence. A re-analysis argued that confounding could reproduce the result (C071). The “8 IQ points” figure that circulates for this study is in the full text, but it refers only to the 23 persistent users who started in adolescence; the most persistent users overall lost 5–6 points (C071).
• Withdrawal and dependence. A withdrawal syndrome was characterised prospectively, with onset in 1–3 days, a peak at 2–6 days and most effects lasting 4–14 days (C072). Dependence was treated in randomized trials, and a 2014 Cochrane review found incomplete evidence for any medication (C072).
• Driving. Acute use was associated with motor-vehicle collisions (OR 1.92 across nine studies), and simulator lane-weaving at moderate blood THC resembled that at legal alcohol limits (C073).
• Lungs and heart. Quitting smoked cannabis resolved chronic-bronchitis symptoms (C074). Low cumulative exposure was not associated with lower lung function in the CARDIA cohort (C074). Smoking cannabis was a rare trigger of myocardial infarction (C074). NASEM found substantial evidence of worse respiratory symptoms with long-term cannabis smoking, alongside moderate evidence of no association with lung cancer (C074).
• Pregnancy. Prenatal exposure was associated with maternal anaemia and low birth weight in a 2016 meta-analysis, with tobacco confounding unresolved (C075).
• Acute toxicity. Cannabinoid hyperemesis was described as a syndrome in 2004 (C076). NASEM found moderate evidence of more paediatric overdose injuries in states where cannabis was legal, and insufficient evidence on deaths from overdose (C076).

Harms inside therapeutic use were reviewed in 2008 (C077). The review covered 31 studies (23 randomized trials, 8 observational). Within the 23 trials, 96.6% of the 4,779 reported adverse events were non-serious; non-serious events were more frequent than in controls (RR 1.86), serious events did not differ significantly, and long-term risks were poorly characterised (C077). A 2015 review of a different, larger set of trials (Whiting) reported higher odds of serious adverse events (OR 1.41, 1.04–1.92) and of withdrawal for adverse events (OR 2.94). The corrected abstract confirms an increased risk of short-term adverse events “including serious AEs”; the odds ratios come from the review’s Table 3, verified on 2026-10-14 (C077). Neither researcher identified this review (C077).

9. What the period’s syntheses concluded (C078–C082)

These are the bodies’ own conclusions, attributed and dated. They are not this programme’s findings.

• Whiting et al. (JAMA, 2015): moderate-quality evidence for chronic pain and spasticity; low-quality evidence for chemotherapy nausea, weight gain in HIV, sleep disorders and Tourette syndrome; an increased risk of short-term adverse events (C078). The journal later corrected the review’s spasticity estimate. In the corrected article, the clinician-rated Ashworth result is not significant (−0.12, 95% CI −0.24 to 0.01; 5 trials), while a significant patient-rated result is reported in the full text (not yet verified). This matches the split in C048 (C078).
• NASEM (January 2017, reviewing literature through August 2016): conclusive or substantial evidence for cannabis in adult chronic pain, and for oral cannabinoids in chemotherapy-induced nausea and vomiting and in patient-reported MS spasticity (C079). Moderate evidence for short-term sleep outcomes; limited evidence for several other indications; limited evidence of no effect in dementia, glaucoma and depression; insufficient evidence for cancers, cachexia, irritable bowel syndrome and epilepsy (C079). The epilepsy rating predates the 2017 Dravet trial (C079).
• American Academy of Neurology (2014): rated oral cannabis extract effective, and nabiximols and THC probably effective, for patient-centred measures of MS spasticity (C080). The record now carries a later “retired” status, separate from those 2014 conclusions (C080).
• Cochrane reviews of the period: mostly found evidence insufficient or absent, across epilepsy, dementia, Tourette syndrome, HIV/AIDS, fibromyalgia, cannabis in schizophrenia and treatment of dependence. The exception was chemotherapy nausea, where cannabis-based medications “may be useful” for refractory cases (C081).
• Professional bodies, including US obstetric and paediatric societies, issued statements in 2015 (C082). Their recommendation content is [HELD], unchecked against the statements (C082).

10. Opioids, and what was not found (C083–C084)

In 2014, an ecological analysis found that states with medical-cannabis laws had 24.8% lower mean annual opioid-analgesic overdose mortality over 1999–2010 (C083). That is an association, and the authors did not present it as causal (C083). Patient surveys reported substituting cannabis for opioids (C083). An uncontrolled pharmacokinetic study in 21 patients on sustained-release morphine or oxycodone found no statistically significant change in plasma opioid levels after vaporised cannabis, while pain decreased (C083). No controlled trial of cannabis as an opioid substitute was identified (C083). Retrospective (2019): extending the analysis through 2017 reversed the association, and the authors judged it likely spurious (C083).

None of the following was identified in the documented searches; that does not mean none exists (C084):

• a controlled human trial of cannabinoid anticancer efficacy;
• any disease-modification benefit;
• a controlled opioid-substitution trial;
• a controlled trial of whole-plant cannabis or THC for PTSD (a separate nabilone pilot exists);
• long-term controlled trials of smoked or vaporised cannabis;
• published results for several completed registered trials.

11. The eight public claims (C085)

These labels come from the probe appendix and are not §M9 dispositions.

12. What the record could support by the end of 2017

The three group headings below are the brief’s descriptive groups. They are not a certainty scale, and the groups are not ranked against each other. Formal certainty ratings belong only to Whiting 2015 and NASEM 2017 (C078, C079).

Replicated or controlled human evidence identified

• Several randomized trials:
  • patient-reported relief of MS spasticity, with clinician-rated results inconsistent (C048);
  • short-term relief of neuropathic pain with smoked or vaporised cannabis in small trials, with mixed nabiximols results (C050, C051);
  • antiemesis in chemotherapy nausea, mainly from pre-2000 trials (C053);
  • cancer appetite and cachexia — two large controlled trials, neither showing an advantage: megestrol gave better appetite palliation than dronabinol (469 patients), and cannabis extract, THC and placebo did not differ (243 patients) (C054).
• A single randomized trial: CBD for convulsive seizures in Dravet syndrome — one sponsor-funded trial of 120 patients, with more adverse events on CBD (C055).
• Harms: psychosis association, withdrawal and dependence, crash risk, excess non-serious adverse events in trials, CBD–antiepileptic interactions, and paediatric overdose injuries (C070, C072, C073, C077, C055, C076).

Preliminary human evidence, or non-human evidence only

• Single or small trials, or uncontrolled reports:
  • HIV appetite — a laboratory study and a 21-day safety trial; anorexia nervosa — one small trial (C054);
  • acute schizophrenia (CBD vs amisulpride), Tourette syndrome (small THC trials), PTSD nightmares (a small nabilone crossover pilot), and dementia, where no benefit was shown (C056);
  • Crohn’s disease — a 21-patient smoked-cannabis trial missed its primary remission outcome while reporting clinical response; a later low-dose CBD trial found no benefit (C057);
  • glaucoma — a transient fall in intraocular pressure after a single THC dose in six patients (C058);
  • fibromyalgia — small trials reporting specific pain or sleep outcomes, with tolerability concerns and a negative Cochrane assessment (C052);
  • anticancer use — one phase I safety study and case reports (C059).
• Experimental challenges: constituent interaction in healthy volunteers (C047); single-dose CBD under a simulated public-speaking test in social anxiety (C056).
• Non-human only: endocannabinoid biology and CBD mechanisms (C044, C045).

Claims appearing in period sources for which this search did not identify traceable support

• tumour shrinkage or cure in people (C059);
• disease modification — tested once and negative (C060);
• endocannabinoid deficiency as an established disease mechanism (C046);
• a causal reduction of opioid deaths by medical-cannabis laws (C083);
• risk-free CBD (C085).

By the end of 2017, controlled cannabinoid trials were numerous, although most remained small and short (C066). In MS spasticity trials, patient-reported and clinician-rated outcomes did not consistently agree (C048). The one large disease-modification trial identified in the window, CUPID, found no effect on MS progression (C060). A new randomized result for purified CBD in Dravet syndrome appeared in 2017, the window’s final year: one sponsor-funded trial of 120 patients, with more adverse events in the CBD group (C055).

13. How this evidence was found, and what may be missing

This is not a systematic review.

• The researchers’ searches: PubMed/MEDLINE, Cochrane, ClinicalTrials.gov, regulator and government sites, and publisher pages, with search logs filed.
• The verifier’s checks:
  • re-ran the canonical count;
  • resolved every identifier (but did not, until the cross-audit of 2026-10-13, check PubMed’s links to published corrections; see below);
  • checked integrity flags (none found);
  • read abstracts for every clinical figure;
  • opened official records: NASEM, Health Canada, MHRA, EMA, FDA, patents, the Federal Register and the court record;
  • ran a completeness probe that found the keystones in §1 and fifteen of sixteen period Cochrane reviews uncited by either researcher;
  • ran the registry cross-check in §6.

Most clinical figures are abstract-level; full text was read for very few. The probe was targeted, not exhaustive. Non-English literature, conference abstracts, WHO and EU trial registries, and non-US regulators other than those named are under-represented.

Held at ledger v1.0, carried into this synthesis and the cross-audit:

• Updated 2026-10-13 after the cross-audit: CAMS correct-guess rate and capsule composition; Novotna’s adjusted difference (C048);
• the CBD–clobazam mechanism and FDA orphan-designation dates (C055);
• a trial’s registry history (C057);
• the Australian Sativex record (C062);
• regulatory-cohort counts for the 2016 phase I trial (C064);
• 19 unscreened registry records and the unsearched WHO/EU registries (C067);
• programme legal bases (C069);
• published errata for five sources, whose content is still to be checked against the ledgered figures (C044, C055, C064, C083);
• Whiting 2015’s patient-rated spasticity estimate (C078);
• professional-body recommendation content (C082).

Each will be resolved, or remain open, by logged correction.

14. Handed to P3 (2018 → cutoff), and what stays with P2

Research threads for P3:

• the 2018 Lennox-Gastaut CBD trials and the 2018 FDA approval of purified CBD;
• the 2019 removal of Sativex’s Canadian conditional indications;
• the 2019 reversal of the opioid-mortality association;
• post-2017 publication outcomes for trials completed by 2017 whose publication status remains unresolved in the P2 ledger (C067). Finishing the original P2 matching, screening the 19 errored lookups and the documented registry searches remain P2 verification tasks (§13);
• whether US research-supply registrations followed the 2016 policy;
• post-2017 Cochrane and national reassessments that revised the period’s ratings;
• longer-term trials of inhaled cannabis.

Unfinished P2 verification stays with P2 and is resolved in its audit or carried as explicitly unresolved P2 findings, not transferred: the held list in §13.

Ledger coverage: all 45 P2 claims (C041–C085) cited. Verifier-found components (C041, C054, C060, C067, C077 and the verifier-found sources in C048–C057, C070, C081) and all held components are mandatory targets for the §M10a cross-audit. Plain-English derivatives follow this text and may not simplify its uncertainty.