R03 — Legal & Illicit Cannabis Production

Independent deep-research pass for the Cannabis Cultivation Knowledge Engine (CCKE)

ChatGPT research pass • 1 September 2026 • Frozen R03 brief / PROTOCOL v1.3

Research posture. This report separates documented structure, measured magnitudes, and causal claims. “Illicit” is treated as a family of populations rather than one cultivation system. Enforcement visibility is never used as a production measure without an explicit bridge. Major quantitative estimates are reconstructed from observations to final estimate wherever possible; opaque precision is treated as estimate-lore rather than upgraded by repetition.

Executive synthesis

The central finding is an epistemic one: legal cannabis is measured through administrative systems, while illicit cannabis is mostly observed through detection, self-report, remote sensing, residual demand models, and case investigations. Those measurement systems are asymmetric. A rise in eradications, seizures, warrants, or seizure value cannot by itself show that illicit production rose; it can also reflect enforcement intensity, better detection, reporting changes, geographic displacement, or larger targeted cases. The historical record itself demonstrates this problem: California DOJ attributed part of the 2005 increase in eradications to more deployment teams, more aggressive aerial techniques, and garden size, while a 2007 federal cultivation assessment explicitly said precise estimates of un-eradicated plants were not feasible and had to assume an eradication fraction. [S07][S08]

Legalization can demonstrably redirect consumer spending and acquisition toward regulated channels, but legal-market capture is not a production share. Canada provides the strongest demand-side example identified: Hammond et al. estimated that 78% of cannabis expenditures were legal in the 12 months ending September 2022, and the estimated legal expenditures were within two percentage points of government retail sales. That validates the legal-expenditure side of the reconstruction unusually well. It still does not establish that 78% of cannabis cultivated for Canadians was legally produced. [S03]

California illustrates why a single “X% illicit” headline can mislead. ERA Economics, for the Department of Cannabis Control (DCC), modeled 2024 consumption at about 3.8 million pounds and estimated licensed-market capture at 38% under its central consumption scenario, with a 29–54% interval across low/mean/high demand assumptions. DCC’s public-facing summary expressed the complement as roughly 62% “sourced illicit.” The estimate is not a direct census of illicit farms: it combines NSDUH-based consumption modeling, under-reporting adjustment, legal sales/output data, eradication records, geospatial grow-site data, yield assumptions, and an estimated relationship between eradicated plants and plants reaching consumers. The underlying report itself warns that unlicensed production estimates are difficult and uncertain. [S01]

The best recent California evidence on the geography of unlicensed cultivation is stronger than seizure-count inference. A 2026 UC Berkeley Cannabis Research Center DCC-funded mixed-method study combines cultivation mapping, license data, interviews, observation, policy data and modeling. It reports that unlicensed cultivation expanded in the early post-legalization period and later receded toward 2018 levels; nearby licensed farms and permit systems were associated with later declines in unlicensed cultivation, while bans showed no consistent overall suppression and eradication-oriented enforcement appeared to lose deterrent association over time as production adapted and relocated. These are important rung-2 observational findings, not a clean randomized or natural-experiment proof that permits cause decline or enforcement causes displacement. [S02]

Persistence in illicit or informal cultivation is not well explained by one variable. California farmer surveys and ethnography repeatedly identify financial, administrative and psychological compliance burdens, especially for smaller diversified farms; California’s State Auditor later found reviewed local permits took more than 2.5 years on average and that equity programs were often young with few permit recipients. Testing is one measurable component: a 2020 UC Davis model estimated mandatory testing at $136 per pound for an eight-pound batch and 4% rejection rate, about 10% of the then-reported wholesale price, with much larger per-pound burden for small batches. Price, local access, licensing design and market collapse therefore plausibly operate together rather than as a single “taxes caused the illicit market” mechanism. [S04][S05][S06]

“Illicit cannabis” does not have a single harm profile. Public-land trespass cultivation has well-documented site-specific environmental risks: water diversion, rodenticide/insecticide exposure, habitat disturbance, and overlap with sensitive predator habitat. Northern California modeling also estimates greater dry-season water extraction by unpermitted than permitted farms in studied watersheds. But that evidence cannot characterize an unlicensed warehouse, an over-limit household grow, a legacy smallholder on private land, or licensed diversion. The population denominator—what fraction of all illicit production is trespass cultivation—remains poorly established. [S09][S10][S11]

Likewise, legality does not create a safety halo. A 2026 Health Canada laboratory comparison found substantially higher pesticide, microbial, mycotoxin and several metal burdens in 50 law-enforcement-seized illegal flower samples than in 50 regulated samples, but also found shortcomings within regulated products. Separately, a 2025 Colorado study of 277 licensed flower/concentrate products found only 56.7% of flower samples within ±15% of labeled THC, versus 96% of concentrates. The defensible conclusion is asymmetric but not absolutist: regulated systems can materially reduce some contaminant risks while still producing testing, labeling and laboratory-governance failures. [S12][S13]

Criminal organizations, licensed diversion and labor exploitation are documented in specific cases, but case existence is not prevalence. Federal prosecutions in Oklahoma establish large interstate diversion from ostensibly licensed cultivation in particular networks; a 2025 Massachusetts indictment alleged workers were smuggled into grow houses and passports withheld. Yet Maine federal investigators stated in 2024 that, after more than 40 search warrants into a separate illicit-grow phenomenon, they had not found evidence of human trafficking and workers appeared voluntary. These records support “such patterns occur,” not “cartels run most illegal U.S. grows.” [S14][S15][S16]

Enforcement harms are better identified causally than many production-suppression claims. Difference-in-differences work using U.S. arrest data finds cannabis decriminalization associated with large drops in possession arrests and a reduction in Black–White adult arrest disparity, although disparities persist; a 2026 American Economic Journal study likewise finds recreational legalization reduces possession and sales arrests for Black and White populations and narrows but does not eliminate disparities. By contrast, the evidence that eradication measurably reduces aggregate cannabis supply is much weaker and repeatedly confounded by enforcement effort and adaptive relocation. [S17][S18]

International comparison reinforces the category problem. Uruguay’s 2026 national evaluation reports that “classic drug trafficking” as consumers’ usual access route fell from 58.2% in 2014 to 6.7% in 2024, while a roughly 30% gray/informal segment persisted; its regulated system includes pharmacies, membership clubs and registered home cultivation. Germany, as of 2026, legalizes limited personal cultivation and non-commercial cultivation associations rather than a general adult-use commercial retail market. Thailand reversed from broad post-2022 liberalization toward a medical-controlled model from June 2025 onward. “Legal market” therefore names materially different supply institutions across jurisdictions. [S19][S20][S21]

1. Methodology and evidence classes

Methodological spine: define the populations → identify how each becomes observable → reconstruct estimate provenance → grade transformation assumptions → compare only commensurable quantities → evaluate policy causality separately.

Evidence class 1 — documented structure: administrative records, maps, surveys, ethnography, prosecutions, observed regulatory categories.

Evidence class 2 — measured or estimated magnitudes: production, hectares, expenditures, consumer sourcing, contaminant prevalence. Every magnitude is tied to its measurement method and population.

Evidence class 3 — causal policy claims: rung 1 descriptive/before-after; rung 2 controlled observational or quasi-experimental association; rung 3 credible causal design such as defensible difference-in-differences or natural experiment.

Source-selection rule. Primary administrative sources and peer-reviewed studies were preferred. Enforcement and advocacy documents were used for what they directly observe or record, not for population-wide generalizations. Institutional position is noted as context; it is not itself a validity score.

2. Operational taxonomy of the illicit-spectrum

3. Data epistemology: how illicit numbers are made

4. Major quantitative estimates — 11-field provenance records

4.1 California 2024 central licensed capture = 38% (thus ~62% non-licensed sourcing)

4.2 California 2019–24 unlicensed production central estimate ~11.4M lb/year; range ~7.0–16.3M

4.3 Canada 2022 legal expenditure capture = 78%

4.4 Uruguay “classic drug trafficking” usual access 58.2%→6.7%

4.5 Northern California unpermitted cultivation footprint: 5,997 parcels / 406.23 ha vs 1,787 permitted / 243.06 ha in study dataset

5. Legal transition: who enters, who remains outside

5.1 California

California’s transition is not well described as a clean migration from “illegal farms” into licenses. Farmer-level survey research found about one-third of 362 respondents had never applied for a license; non-applicants tended to be smaller and to use cannabis as part of diversified livelihood strategies. Reported barriers were financial, administrative and psychological rather than simply anti-regulatory norms. [S04]

The administrative burden is independently visible in state audit evidence. In six reviewed local jurisdictions, the California State Auditor found reviewed permits took more than 2.5 years on average, fees varied widely, and most equity programs reviewed were new with few equity applicants yet receiving permits. This supports the existence of barriers; it does not by itself quantify how much illicit production they cause. [S06]

Mandatory testing is a measurable legal-market wedge. Valdes-Donoso, Sumner and Goldstein modeled 2019 California testing compliance at $136/lb under their central eight-pound batch / 4% rejection scenario (~10% of reported wholesale price), while one-pound batches had far higher per-pound costs. This is evidence of scale-sensitive compliance cost, not proof that testing “caused” non-licensure. [S05]

The 2026 Berkeley report adds longitudinal spatial context: the presence of nearby licensed farms was associated with subsequent reductions in unlicensed cultivation, permit jurisdictions outperformed bans in later periods on several measures, and unlicensed cultivation became increasingly concentrated among economically tenuous operators. Treat these as rung-2 associations supported by mixed methods, not a single causal coefficient. [S02]

5.2 Canada

Canada shows a clear consumer-channel transition. Health Canada’s survey series reports the “usual source” becoming overwhelmingly legal over time, while Hammond et al.’s demand-side expenditure estimate found 78% legal capture in 2022 with unusually good external validation against official legal retail sales. These measures establish transition in acquisition/spending, not a national legal-production percentage. [S03]

Price and access repeatedly appear as purchase-side barriers. Comparative survey work in Canada and U.S. legal states found price the most commonly reported reason for buying from illegal sources and inconvenience/retail access another important barrier; Canadian legal flower prices fell after legalization but remained above illicit prices in early transition data. These are observational mechanisms, not isolated causal proofs. [S22]

Home cultivation increased modestly after legalization: Canadian ICPS data report 5.8% of consumers home-growing pre-legalization versus 7.9% in 2019 and 8.8% in 2020, with most reported grows within the four-plant limit and lower prevalence in provinces prohibiting home cultivation. This supports a policy effect on home growing itself; it does not demonstrate material commercial diversion. [S23]

5.3 Oregon, Oklahoma, Uruguay, Germany, Thailand

Oregon’s track-and-trace reports are valuable precisely because they measure the licensed system: the Oregon Liquor and Cannabis Commission has repeatedly documented licensed supply exceeding estimated in-state licensed demand. That demonstrates licensed overcapacity/margin pressure, not the size of Oregon’s illicit production or export diversion. [S24]

Oklahoma demonstrates boundary porosity. Federal prosecutions and state inspections document cases in which licensed or purportedly licensed cultivation was used to produce product for interstate illegal distribution, while the state’s license count has contracted sharply. The contraction in licenses is administrative structure; prosecutions establish mechanisms of diversion; neither yields a defensible statewide share of illicit production. [S14]

Uruguay’s supply architecture—pharmacies, membership clubs and registered home cultivation—produced a large shift away from “classic trafficking” in consumer-source surveys, but the government’s 2026 evaluation simultaneously estimates a persistent gray/informal segment near 30%. This is a useful warning against a binary legal/illegal frame. [S19]

Germany’s 2026 Cannabis Act regime permits limited personal cultivation and licensed non-commercial cultivation associations (maximum 500 members) rather than general commercial adult-use retail. Thailand, conversely, tightened cannabis flower into a medical-controlled regime from June 2025 and continued that framework in 2026. Cross-country “legal market capture” comparisons therefore require matching institutional supply categories, not simply coding countries legal/illegal. [S20][S21]

6. Why illicit/informal production persists, and how sectors interact

Enforcement and supply suppression. The most defensible conclusion is that eradication indisputably removes detected plants and can alter where/how some growers operate, but aggregate supply suppression is not well identified. Federal agencies themselves have repeatedly acknowledged that accurate domestic production is unknown; 2009 NDIC reporting noted record eradication alongside high availability and limited supply/price disruption. Berkeley’s recent California work is consistent with adaptation toward remoter sites and diminishing deterrent association. This supports “enforcement changes behavior and removes detected production,” not a quantified national supply elasticity. [S08][S02]

7. Harms — both directions

7.1 Environmental harms: strong evidence, narrow populations

Bauer et al. (2015) modeled surface-water demand from cannabis in four northwestern California watersheds and found that diversion could materially reduce low flows, including scenarios where estimated demand approached or exceeded available streamflow in the driest periods. The study concerns specific watersheds and modeled demand, not all illicit cannabis. [S09]

Dillis et al. (2023) mapped 1,787 permitted and 5,997 unpermitted parcels in Northern California study areas and modeled monthly water extraction. Unpermitted extraction was consistently greater in the dry season in the modeled watersheds. Because unpermitted water use is partly inferred using a model trained on permitted reporting, the result is stronger than rhetoric but remains modeled and regional. [S10]

Thompson et al. (2014) linked toxicant exposure and cannabis-site proximity to fisher survival risk in the Sierra National Forest, while Wengert et al. (2021) used 1,469 known trespass sites to model potential overlap with fisher, marten and spotted-owl habitat. These are compelling trespass/public-land results; neither supplies a denominator for all illicit cultivation. [S11]

7.2 Product safety: neither market gets a halo

Fiering et al. (2026), a Health Canada laboratory study of 50 regulated and 50 law-enforcement-seized illegal flower samples, found pesticide residues in 94% of illegal samples versus trace detections in two regulated samples, higher microbial failures and mycotoxin detections in illegal samples, and higher concentrations of several metals. Because illegal samples were enforcement seizures rather than a probability sample of all illicit cannabis, the direction is informative but the exact percentages should not be generalized to every illicit purchase. [S12]

Licensed-market controls also fail. Giordano et al. (2025) independently analyzed 277 Colorado licensed products purchased from 52 dispensaries: 56.7% of flower and 96.0% of concentrates were within ±15% of labeled THC; observed THC was significantly lower than labeled values for both product classes. This documents labeling/testing governance shortcomings without negating the contaminant-control advantages observed in Canada. [S13]

7.3 Human costs: illicit exploitation and enforcement harms

Specific U.S. cases document large illegal cultivation/distribution networks, licensed-facility diversion, and alleged labor coercion. Those records should be written as case findings (or allegations where indictment-only), never as a national prevalence claim. Contrasting investigations matter: Maine federal authorities publicly said they had not found human-trafficking evidence after dozens of searches in a cluster of illicit grows. [S14][S15][S16]

Criminalization itself has measurable distributional effects. Gunadi and Shi’s difference-in-differences study across 37 states found decriminalization associated with >70% lower adult and >40% lower youth cannabis-possession arrest rates, and ~17% lower Black/White adult arrest disparity; youth disparity did not show the same reduction. Meinhofer, Rubli and Cunningham (2026) similarly find recreational legalization reduces Black and White cannabis possession/sales arrests and narrows but does not eliminate disparities. These are among the strongest causal designs in the R03 evidence base. [S17][S18]

8. Pre-registered myth checks

M1. “Legalization eliminates the illicit market” / “did nothing.” Both formulations fail. Canada and Uruguay show large shifts toward legal/regulated acquisition, while California, Uruguay and others retain substantial unlicensed/gray activity. Magnitude and mechanism are jurisdiction-specific. Supported verdict: both absolutes are false; direction of legal-channel capture is well supported, total elimination is not.

M2. “X% of cannabis in [state] is illicit.” Potentially valid only when the metric is named. California’s ~62% is a modeled 2024 consumption-sourcing complement, not direct production or farm share; its underlying legal-capture interval implies substantial uncertainty. Verdict: frequently category-laundered; trace method before repeating.

M3. “Illicit cannabis is commonly laced with fentanyl.” A Connecticut response system reported one laboratory-confirmed marijuana/fentanyl sample in 2021; later state reporting characterized it as an isolated likely cross-contamination event. Toxicology showing fentanyl + THC in a person cannot identify a laced cannabis source without product testing. Verdict: “commonly laced” unsupported; do not overcorrect to “never occurs.” [S25]

M4. “Legal means tested means safe.” Regulated cannabis shows documented label inaccuracies, recalls/lab-governance issues, and some contaminant failures, even while direct Canada testing found markedly higher contaminant burden in seized illegal samples. Verdict: false as an absolute; regulation changes risk, it does not abolish it.

M5. “Cartels run most illegal U.S. grows.” Organized networks and transnational actors are documented in particular cases and enforcement intelligence. No defensible denominator identified establishes “most.” Verdict: existence supported; prevalence claim unsupported.

M6. “Illicit growers are environmental villains” / “legacy growers are land stewards.” Trespass-site harms are real and well studied; legacy/private-land populations are heterogeneous; permitted farms can also create environmental pressures. Verdict: moral categories collapse distinct populations and are not scientific classifications.

M7. “Eradication and enforcement reduce supply.” Eradication removes detected plants; behavioral displacement/adaptation is documented. Aggregate market supply effect is not well identified, and historical federal reporting notes high availability despite record eradication. Verdict: site-level removal true; broad supply-suppression magnitude unresolved.

M8. “Seizure and eradication totals track production trends.” Historical agencies themselves changed methodology and enforcement intensity; California explicitly attributed increases to deployment/surveillance. Verdict: unsupported without an argued detection/enforcement bridge.

M9. “Home-grow allowances feed the black market.” Legal home-grow provisions increase home cultivation; Canada shows most reported home growers within plant limits. The searched literature did not yield a robust estimate of commercial diversion attributable to legal home-grow allowances. Verdict: plausible/case-specific mechanism, not established general market effect.

9. Candidate claims for editorial review

Illicit cannabis production cannot be inferred directly from seizure or eradication totals because those metrics jointly reflect production, enforcement effort, detection technology, reporting practice and grower adaptation. [S07][S08]

California’s widely repeated ~62% “illicit” 2024 figure is best understood as the complement of a modeled licensed share of consumer consumption under a central demand scenario, not a direct estimate of the share of farms or plants that are illicit. [S01]

The 2024 California market model reports a licensed-consumption share of roughly 29–54% across low/central/high consumption scenarios, demonstrating substantial denominator uncertainty around any single illicit-share headline. [S01]

California’s 2019–24 unlicensed-production estimate (~11.4M lb central; ~7.0–16.3M range) is reconstructable but highly assumption-sensitive because it depends on eradication fractions, geospatial extrapolation, yield assumptions and reporting quality. [S01]

UC Berkeley’s 2026 California mapping study finds unlicensed cultivation expanded after legalization and later receded toward 2018 levels, with important geographic relocation over time. [S02]

In the Berkeley study, nearby licensed cultivation and later permit-program exposure are associated with declines in unlicensed cultivation, while cultivation bans show no consistent overall suppressive association. [S02]

Eradication-oriented enforcement in the Berkeley study is associated with declining deterrent effectiveness over time and adaptation/relocation, a rung-2 finding rather than proof of a statewide causal effect. [S02]

California cannabis farmers who did not enter licensing in survey research were disproportionately small/diversified operators and reported financial, administrative and psychological barriers to compliance. [S04]

California mandatory testing created scale-sensitive compliance costs; a 2020 model estimated ~$136/lb under an eight-pound batch and 4% rejection scenario, about 10% of then-reported wholesale flower price. [S05]

California local permitting can itself be a substantial administrative barrier: a State Auditor sample found reviewed local approvals averaged more than 2.5 years. [S06]

Canada had shifted a large majority of consumer cannabis expenditures to legal channels by 2022: a demand-side study estimated 78% legal expenditure capture and closely matched official legal retail sales. [S03]

Canada’s 78% legal-expenditure estimate cannot be interpreted as 78% legal cultivation/production because it measures spending at acquisition, not farm output. [S03]

Legal home-cultivation provisions increase reported home growing, but available Canadian evidence does not establish that legal home grow is a major source of illicit commercial supply. [S23]

Uruguay’s national evaluation reports a steep decline in “classic trafficking” as consumers’ most common cannabis access route from 58.2% in 2014 to 6.7% in 2024 while a sizable gray/informal market remains. [S19]

Legal cannabis supply models are not institutionally comparable across countries: Germany’s 2026 framework centers limited personal cultivation and non-commercial associations, unlike Canada’s commercial retail system or Uruguay’s state-regulated multi-channel model. [S20][S19]

Trespass/public-land cannabis cultivation has documented environmental harms, including water diversion and toxicant exposure, but those findings cannot be generalized to all unlicensed cultivation. [S09][S11]

In studied Northern California watersheds, modeled dry-season water extraction was greater for unpermitted than permitted cannabis cultivation, with effects spatially concentrated rather than uniform. [S10]

A 2026 Health Canada laboratory comparison found substantially greater pesticide, microbial, mycotoxin and several metal contamination in law-enforcement-seized illegal flower than regulated flower; enforcement-seizure sampling limits prevalence generalization. [S12]

Licensed cannabis does not guarantee accurate potency labels: a 2025 Colorado study found only 56.7% of tested flower products within ±15% of labeled THC, although 96% of concentrates met that threshold. [S13]

Organized criminal cultivation, licensed diversion and alleged labor coercion are documented in specific U.S. cases, but these case records do not establish the prevalence of such networks across all illicit cultivation. [S14][S15][S16]

Cannabis decriminalization/legalization has stronger causal evidence for reducing cannabis arrests than eradication has for suppressing aggregate cannabis supply. [S17][S18][S08]

Decriminalization reduces cannabis-possession arrests for Black and White populations and can narrow adult racial arrest disparities, but available evidence does not show disparities disappear. [S17][S18]

The statement that illicit cannabis is “commonly” fentanyl-laced is unsupported by identified laboratory evidence; isolated contamination has been documented, so “never” is also too strong. [S25]

Where a jurisdiction lacks a reconstructable illicit-production denominator, “not defensibly estimable from available evidence” is methodologically superior to producing a legal/illicit production ratio from seizures or sales residuals alone.

10. Explicit data gaps

A defensible U.S.-national estimate of illicit cannabis production with transparent detection/yield uncertainty was not identified.

A reliable denominator for what fraction of illicit U.S. production occurs on public/trespass land was not identified; ecological trespass findings therefore cannot be population-weighted.

No robust national prevalence estimate was identified for licensed-operator diversion, “burner” licensing, or illicit product entering licensed channels.

No strong estimate was identified for the share of legal home-grown cannabis diverted into commercial illicit markets; much of the literature discusses diversion as a policy concern rather than measures it.

Comparative estimates of compliance-cost components across jurisdictions are incomplete and often industry-sourced; causal attribution of illicit persistence to taxes versus other costs remains weak.

Equity-program evaluation is still immature. Administrative audits document program design and limited early permitting outcomes, but durable business survival, wealth, and transition-from-illicit outcomes are sparsely measured.

Organized-crime and labor-trafficking evidence is case-rich but denominator-poor; “how common” remains much harder to answer than “does it occur.”

Illicit-product contaminant studies often rely on law-enforcement seizures or convenience samples, limiting direct prevalence comparison with probability-sampled consumer purchases.

Americas-wide herb-production monitoring lacks the recurring crop-survey/satellite framework used historically for Moroccan cannabis resin cultivation, leaving cross-national production estimates unusually dependent on indirect data.

The causal effect of eradication intensity on aggregate cannabis availability, price and total output remains unresolved because enforcement effort and grower adaptation are endogenous and poorly measured.

11. Strongest sources and reliability / institutional-position notes

12. Additional source register used in this pass

S14 — U.S. Attorney’s Office, Eastern District of Oklahoma, 2024–2026 prosecutions/indictments concerning interstate marijuana trafficking and licensed-grow diversion; stable DOJ case pages.

S15 — U.S. Attorney’s Office, District of Massachusetts, July 8, 2025 indictment alleging worker smuggling/passport withholding in cannabis grow houses; allegation status preserved.

S16 — U.S. Attorney’s Office, District of Maine, May 3, 2024 statement on illicit marijuana grow investigations reporting no human-trafficking evidence identified after >40 searches.

S20 — German Federal Ministry of Health (BMG). Frequently Asked Questions on the Cannabis Act, current 2026. https://www.bundesgesundheitsministerium.de/en/themen/cannabis/faq-cannabis-act/page

S21 — Royal Thai Government / Public Relations Department and Ministry of Public Health cannabis controlled-herb rules, effective June 2025 with 2026 updates; official government sources.

S22 — Goodman S, Wadsworth E, Hammond D. Reasons for Purchasing Cannabis From Illegal Sources in Legal Markets: Findings among Cannabis Consumers in Canada and U.S. States, 2019–2020. Journal of Studies on Alcohol and Drugs. 2022;83(3):392–401. DOI 10.15288/jsad.2022.83.392; plus Wadsworth et al. 2022 Drug and Alcohol Dependence. DOI 10.1016/j.drugalcdep.2021.109262.

S23 — Wadsworth E et al. Home cultivation across Canadian provinces after cannabis legalization. Addictive Behaviors Reports. 2022;15:100423 (PMCID PMC9006643).

S24 — Oregon Liquor and Cannabis Commission. 2025 Recreational Marijuana Supply and Demand Report, Feb. 2025; METRC-derived licensed-market data.

S25 — Connecticut Overdose Response Strategy / state epidemiological reporting, 2021–2022, on one lab-confirmed cannabis/fentanyl sample later characterized as likely isolated cross-contamination; CDC/APHL interpretive guidance that toxicology alone cannot identify a laced cannabis product.

13. Search-stopping rationale

Searching stopped after convergence across independent evidence streams rather than after a fixed source count. The pass covered: California DCC economic and enforcement material; UC Berkeley Cannabis Research Center mapping/ethnography; California State Auditor licensing/equity audits; Canadian federal surveys and peer-reviewed market-capture studies; Uruguay’s 2026 national evaluation; current official German and Thai regulatory sources; Oregon and Oklahoma administrative/enforcement records; DEA/archived NDIC eradication methodology; peer-reviewed environmental, product-safety, farmer-compliance, home-cultivation and criminal-justice studies; and targeted searches for home-grow diversion, eradication supply effects, equity outcomes, organized crime, labor trafficking and fentanyl-lacing claims. Saturation was reached on the report’s central propositions: (1) legal/illicit measurement asymmetry; (2) consumer-channel capture is measurable but not equivalent to production share; (3) unlicensed populations are heterogeneous; (4) compliance/access/price mechanisms are documented but rarely singly causal; (5) trespass harms and organized-crime cases are population-specific; and (6) seizure/eradication counts lack a stable production denominator.

Areas searched and still substantively empty or denominator-poor are listed in §10. In particular, the absence of robust national U.S. production, licensed-diversion prevalence, home-grow commercial diversion, and trespass-share denominators is treated as a finding rather than filled with industry estimates. Additional searching would be justified if Claude’s cross-examination identifies a specific missing jurisdiction, source lineage, or candidate claim requiring a live two-check; otherwise broader searching is likely to add repetitions rather than change the synthesis.

R03 • ChatGPT independent research pass • 2026-09-01 • For CCKE editorial verification