Claim Ledger — R03: Legal & Illicit Cannabis Production

Version: v1.0 — signed off by Danny September 2, 2026 (approval conditional on ChatGPT’s eight editorial calibrations, all applied; see changelog) · Brief: v1.0 FROZEN 2026-09-01 Grading: evidence status × cannabis-specificity × confidence per PROTOCOL v1.3 §3, plus two R03 additions adopted as standing methodology: every quantitative claim carries an observation-class tag — observed (direct administrative/laboratory measurement) / survey / modeled / residual (demand-minus-legal complement) / enforcement (detection-system count) — and every persistence/decline/share wording names its metric. Populations: claims tag the illicit-spectrum population(s) they cover (P1 legacy smallholders · P2 unlicensed warehouses · P3 trespass/public-land · P4 prohibition-state grow houses · P5 licensed diversion/inversion/burner · P6 over-limit home grows · P7 cross-border · P8 traditional regions), per the frozen brief’s boundary discipline. Sources: cited inline by verified tuple; registered as S-136–S-165 in SOURCE_LIBRARY (R03 section). Every citation passed R03’s two-check verification (critique.md, verification/).

A. Epistemology & measurement

R03-C001 — Keystone. Illicit cannabis production is not a single measurable quantity: legalization changes different populations, sourcing channels, and production systems differently, and the indicators used to describe the illicit sector — sales residuals, consumer surveys, cultivation maps, seizures, eradications, prosecutions — observe different phenomena. A defensible claim must name both the population and the measurement system.

Evidence status: established · Specificity: direct · Confidence: high · Populations: all
Sources: convergent finding of both independent runs; instantiated by C005/C006/C021/C023 · Agreement: both
Notes: Wave 1’s closing stone: R02 established there is no global denominator; R03 adds that even within jurisdictions the illicit denominator is reconstructed from heterogeneous visibility systems, not observed.

R03-C002 — Legal and illicit production are measured asymmetrically: legal markets generate production-side administrative data (licensing, track-and-trace, tax, testing, sales), while illicit markets are observed only through detection systems. A change in seizures or eradications can reflect production, enforcement effort, detection, displacement, or reporting — enforcement agencies have said so themselves (“Precise estimates regarding the number of cannabis plants not eradicated… are not feasible,” NDIC 2007, verbatim-verified; CA DOJ attributed 2005 eradication increases partly to deployment and aerial technique).

Evidence status: established · Specificity: direct · Confidence: high · Tag: enforcement
Sources: NDIC Domestic Cannabis Cultivation Assessment 2007/2009 (archived DOJ); CA DOJ CAMP releases 2005–06 · Agreement: both
Notes: Extends R02-C005. Myth M8 verdict lives here.

R03-C003 — Across the national and state jurisdictions examined in R03, no scientifically defensible direct estimate of aggregate illicit cannabis production was identified; circulating totals are transformations of visibility variables, and “not defensibly estimable from available evidence” is frequently the correct answer.

Evidence status: established (as a verified absence) · Specificity: direct · Confidence: high
Sources: both runs’ searched-and-empty declarations; NDIC’s own concessions · Agreement: both
Notes: Extends R02-C002 to the within-jurisdiction level.

Evidence status: established · Specificity: direct · Confidence: high
Sources: Ridgeway & Kilmer 2016 (Drug Alcohol Depend 165:175–180); Davenport & Caulkins 2016 (J Drug Issues 46(4):388–403 — corrected tuple); Butsic & Brenner 2016 (ERL 11:044023); both runs’ method catalogs · Agreement: both

R03-C005 — California’s famous “13.5M lb / 80%+ illicit or exported” figures trace to a single pre-market regulatory forecast — the CDFA CalCannabis SRIA (ERA Economics, Jan 2018, “more than four-fifths” out-of-state) — subsequently repeated without its assumptions; the 15M lb and 90% variants exceed even that source, and the same firm’s 2024 model revised the production estimate downward.

Evidence status: established (provenance) · Specificity: direct · Confidence: high · Tag: modeled (2018 forecast)
Sources: CDFA SRIA 2018 (ERA Economics); ERA/DCC 2024 Market Outlook · Agreement: gemini-trace, editor-completed
Notes: The corpus’s cleanest estimate-lore lineage: ancestor and reconstructable successor share an author. Feeds myth M2 (“X% illicit”).

R03-C006 — The most reconstructable current California estimate identified in R03 is a model, not a census: the state-commissioned ERA/DCC 2024 outlook models ~3.8M lb consumption with 38% licensed capture under the central demand scenario (29–54% across scenarios) — the “62% illicit” headline is that model’s residual complement — and estimates unlicensed production at ~11.4M lb/yr central (7.0–16.3M), built from eradication records (assumed 7.5–18.3% of illicit production), geospatial data, and yield assumptions (0.25–0.5 lb/plant).

Evidence status: probable (as a transparent model) · Specificity: direct · Confidence: high (that this is what the model says) · Tags: residual (capture) + modeled (production)
Sources: ERA Economics for DCC, California Cannabis Market Outlook 2024 (2025), PDF verified verbatim (cdn.cannabis.ca.gov) · Agreement: chatgpt-anchor, editor-verified
Notes: Never quote “62% illicit” without the model framing — the first “X% illicit” figure in the corpus that survives the reconstruction-chain standard, precisely because it stops pretending to be a measurement.

R03-C007 — The DEA/sentencing-era “one pound per plant” yield convention embedded in eradication valuations is roughly an order of magnitude above measured indoor per-plant yields and ignores maturity and sex — a canonical example of denominator laundering in enforcement statistics.

Evidence status: established · Specificity: direct · Confidence: high · Tag: enforcement
Sources: Caulkins, RAND WR-764 (2010); USSC guideline history (1 plant = 100 g, formerly 1 kg at 50+ plants); Toonen et al. yield data · Agreement: gemini-trace, editor-anchored

B. Populations & structure

R03-C008 — “Illicit cannabis” comprises at least eight structurally distinct populations (P1–P8) differing in scale, geography, practices, observability, and market linkage; evidence about one population does not transfer to another, and both independent research passes converged on essentially the same taxonomy.

Evidence status: established · Specificity: direct · Confidence: high
Sources: Bodwitch et al. 2019 (Cal Ag 73:177–184); Polson & Petersen-Rockney 2019 (Cal Ag 73:185–193); Gabriel/Bauer trespass literature; EMCDDA/EUDA 2024; UNODC crop-survey record; prosecution records · Agreement: both
Notes: The population tag on every subsequent claim is this claim in action. Myth M6 verdict.

R03-C009 — Organized-crime involvement in North American cultivation is documented at case level — including licensed-facility diversion at scale (US v. Weng, W.D. Okla., ~56,000 lb interstate; convictions Jan 2024) and alleged labor coercion (D. Mass. indictment, July 2025, passports withheld) — but case existence establishes no prevalence: a parallel Maine investigation reported no human-trafficking evidence after >40 searches, and no defensible denominator exists for “most.”

Evidence status: established (cases) / unknown (prevalence) · Specificity: direct · Confidence: high · Populations: P2, P5 · Tag: observed (judicial records)
Sources: DOJ W.D. Okla. releases (2024); DOJ D. Mass. (2025, allegation status); DOJ D. Maine statement (May 2024) · Agreement: both
Notes: Myth M5 verdict. The claimed $1,400–1,800/lb Weng transaction prices are not in the public record (withdrawn; only a $15/lb transport fee is documented).

R03-C010 — Boundary-crossing mechanisms between legal and illicit sectors — diversion out of licensed grows, inversion of illicit product into track-and-trace, and “burner” licensing — are documented through enforcement actions and audits; their aggregate volumes are unmeasured.

Evidence status: probable (mechanisms) / unknown (magnitudes) · Specificity: direct · Confidence: high · Populations: P5
Sources: DCC enforcement filings; OLCC audit records; W.D. Okla. prosecutions · Agreement: both (Gemini’s vocabulary, ChatGPT’s calibration)

C. The legal transition

A practitioner-record passage is withheld from the public copy.

R03-C012 — The administrative wall is documented, not inferred: reviewed California local permits averaged >2.5 years (State Auditor 2023-116); 62% of California jurisdictions prohibited cannabis retail (DCC, Feb 2022); state cultivation license fees ran $1,205–$77,905/yr by tier (CDFA schedule, verified exactly); equity programs were young with few permit recipients at audit.

Evidence status: established · Specificity: direct · Confidence: high · Tag: observed (administrative)
Sources: CA State Auditor 2023-116 (2024); DCC ban data 2022; CDFA CalCannabis fee schedule · Agreement: both (Gemini’s 62% verified; ChatGPT’s audit anchor)
Notes: Local CUP ($10–30K) and CEQA ($15–50K+) ranges remain practitioner estimates (anecdotal); Gemini’s $330–420/lb smallholder amortization is modeled arithmetic, preserved as inference only.

A practitioner-record passage is withheld from the public copy.

Evidence status: established · Specificity: direct · Confidence: high · Tags: survey (Hammond) + observed (StatCan accounts)
Sources: Hammond et al. 2025 (IJDP 142:104828); Statistics Canada National Cannabis Economic Accounts · Agreement: both (Gemini’s 70–80% deflated to the verified 67–70%)
Notes: Price is the most commonly reported reason for continuing illegal purchases (Goodman et al. 2022, JSAD 83:392–401).

R03-C015 — Canadian home cultivation rose modestly after legalization (5.8% of consumers pre-legalization → 7.9% in 2019 → 8.8% in 2020, lower in prohibiting provinces) and non-compliance was not rare — 42.3% of surveyed home cultivators exceeded the four-plant limit in 2018 (24.5% in 2020) — while evidence that legal home-grow materially supplies commercial illicit markets remains limited.

Evidence status: established (trend, exceedance) / unknown (commercial diversion magnitude) · Specificity: direct · Confidence: high · Populations: P6 · Tag: survey
Sources: Wadsworth et al. 2022 (Addict Behav Rep 15:100423) · Agreement: chatgpt-claim, editor-recalibrated per ChatGPT’s own review
Notes: Myth M9 verdict.

R03-C016 — Regulated-market oversupply is administratively documented: Canadian licensed producers destroyed >1.7 million kg of unsold dried cannabis 2018–mid-2023 (Health Canada compliance data as reported by MJBizDaily), and Oregon’s OLCC repeatedly documented licensed supply exceeding in-state licensed demand with outdoor wholesale prices falling to ~$100–300/lb.

Evidence status: established · Specificity: direct · Confidence: high · Tag: observed (administrative/industry-reported)
Sources: Health Canada data via MJBizDaily 2023; OLCC 2025 Supply & Demand Report · Agreement: both
Notes: Measures the licensed system — not illicit-sector size. Connects to R02-C025/R05-C029 price-collapse canon.

R03-C017 — “Legal market” names materially different institutions across jurisdictions: Uruguay’s pharmacy/club/registered-home-grow system cut “classic drug trafficking” as consumers’ usual access from 58.2% (2014) to 6.7% (2024) while a ~30% gray/informal segment and roughly half-of-users legal reach persisted; Germany (2024–26) built personal cultivation and non-commercial associations, not commercial retail; Thailand reversed toward medical control from June 2025. Cross-country “capture” comparisons require matching institutional categories.

Evidence status: established · Specificity: direct · Confidence: high · Tag: survey (Uruguay) + observed (statutes)
Sources: Uruguay JND/OUD Segunda Evaluación (2026 — endpoints verified); German BMG Cannabis Act materials; Thai MoPH rules; Queirolo et al. 2016/2025 · Agreement: both

D. Persistence, enforcement & causality

R03-C018 — Illicit/informal persistence is multi-mechanism, not mono-causal: documented mechanisms include the tax/price wedge (California statutory layering compounding to ~28–45%+ over untaxed product), retail access gaps (C012’s ban geography), compliance burden (C012–C013), livelihood/cultural continuity (P1 ethnography), and interstate price arbitrage (legal-state wholesale ~$250–450/lb vs. high-priced eastern limited-license markets; prohibition-state illicit prices are not systematically indexed). No study apportions causal shares.

Evidence status: established (mechanisms) / unknown (causal shares) · Specificity: direct · Confidence: high · Tags: mixed; price bands = observed (Benchmarks/CDTFA) with population caveats
Sources: CDTFA/LAO tax structure; Goodman et al. 2022; Cannabis Benchmarks reporting; Polson ethnography line; prosecutions · Agreement: both (Gemini’s wedge quantified and bounded; its prohibition-state band recharacterized)
Notes: The archive’s 2018→2024 “taxes built the grey market” thesis lands here as a documented mechanism at rung 1–2 — supported as contributor, unproven as sole cause.

R03-C019 — The best longitudinal California evidence (UC Berkeley CRC 2026, DCC-funded, mixed-method mapping) finds unlicensed cultivation area expanded early post-legalization then receded toward 2018 levels, with declines associated with nearby licensed farms and permit regimes, no consistent suppressive association for bans, and eradication-oriented enforcement losing deterrent association as growers adapted and relocated — rung-2 observational findings, verified verbatim from the report.

Evidence status: probable · Specificity: direct · Confidence: high · Populations: P1/P2 (mapped outdoor/greenhouse) · Tag: observed (mapped area) + modeled (associations)
Sources: Polson, Reiman, Dillis et al., Berkeley CRC final report (2026), DCC Grant 93284 · Agreement: chatgpt-anchor, editor-verified
Notes: The report itself uses deterrence language; the ledger retains only the observed association. Rung-2, association not coefficient.

R03-C020 — Mapped-area decline and modeled sourcing persistence coexist without contradiction: a jurisdiction can simultaneously show receding unlicensed cultivation area (C019), large modeled unlicensed production (C006), high illicit consumption-sourcing residuals, falling eradication counts, and growing licensed sales — because each metric observes a different phenomenon. Standing rule: every persistence/decline/share statement names its metric.

Evidence status: established (methodological) · Specificity: direct · Confidence: high
Sources: C006 + C019 reconciliation; both editorial reviews adopted · Agreement: both

R03-C021 — Eradication verifiably removes detected plants and changes grower behavior (displacement to remoter sites), but no credible identification of aggregate supply suppression was located: federal reporting has paired record eradication with undiminished availability, and general enforcement-price scholarship finds weak effects — while the strong specific claim “eradication has no supply effect” also lacks direct identification.

Evidence status: established (absence of identification) · Specificity: direct · Confidence: high · Tag: enforcement
Sources: NDIC 2009; Berkeley CRC adaptation findings; Caulkins & Reuter 2010 (Crime & Justice 39:213–271 — corrected tuple; general, not cannabis-specific) · Agreement: both (Gemini’s overreach trimmed)
Notes: Myth M7 verdict.

R03-C022 — Enforcement harms are better causally identified than enforcement benefits: difference-in-differences studies find decriminalization associated with >70% lower adult and >40% lower youth cannabis-possession arrest rates and ~17% narrowing of the Black–White adult arrest disparity, and legalization reduces possession/sales arrests across races while narrowing but not eliminating disparities (the strongest causal designs in the R03 evidence base; ACLU 2020 documents the persisting disparity baseline, institutional position noted).

Evidence status: established · Specificity: direct · Confidence: high · Tag: observed (arrest records, causal designs)
Sources: Gunadi & Shi 2022 (Soc Sci Med 293:114672); Meinhofer, Rubli & Cunningham 2026 (AEJ:EP 18(2):346–374); ACLU 2020 · Agreement: both

E. Harms — both directions

Evidence status: established (P3 harms) / unknown (P3’s fraction) · Specificity: direct · Confidence: high · Populations: P3 only · Tag: observed
Sources: Gabriel et al. 2012 (PLoS ONE e40163) + 2015 (e0140640); Gabriel et al. 2018 (ACE 13(1):2); Thompson et al. 2014 (Cons Lett 7:91–102); Wengert et al. 2021 (PLoS ONE e0256273); Bauer et al. 2015 (e0120016); Cal Ag remediation reporting · Agreement: both
Notes: Population discipline is the claim’s core; myth M6’s factual half.

R03-C024 — In studied Northern California watersheds, modeled dry-season water extraction was greater for unpermitted than permitted cannabis farms (5,997 unpermitted vs 1,787 permitted mapped parcels in the study dataset) — a regional, partly model-inferred result, stronger than rhetoric and weaker than a census.

Evidence status: probable · Specificity: direct · Confidence: medium (exact parcel figures full-text; direction abstract-verified) · Tag: modeled
Sources: Dillis et al. 2023 (Env Res Comm 5:025005) · Agreement: chatgpt-claim, editor-verified with caveat

R03-C025 — Neither market gets a safety halo; in the direct Health Canada comparison, the contamination asymmetry was large: the head-to-head found pesticides in 94% of enforcement-seized illegal flower vs. trace detections in 2 of 50 regulated samples, with higher microbial/mycotoxin burdens in illegal product — while the same study found chromium ~3× higher in legal samples and 48% of legal products >20% off labeled THC; Colorado auditing found only 56.7% of licensed flower within ±15% of label; and potency inflation is demonstrated while “systematic lab shopping” remains reported, not demonstrated.

Evidence status: established · Specificity: direct · Confidence: high · Tag: observed (laboratory)
Sources: Fiering et al. 2026 (J Cannabis Res 8:48 — seizure-sample selection caveat); Giordano et al. 2025 (Sci Rep 15:20822); Schwabe et al. 2023 (PLoS ONE e0282396) · Agreement: both
Notes: Myth M4 verdict. Illicit-sample percentages are enforcement-seizure-selected — direction informative, prevalence not generalizable.

R03-C026 — The myclobutanil hazard is real and correctly bounded: the fungicide thermally decomposes toward hydrogen cyanide on combustion (documented via regulatory/toxicology sources, and the mechanism behind Canada’s 2016–17 recalls), and pesticide-in-smoke transfer is measured — but the corpus cites the chemistry to its real sources, not to the fabricated “Craven et al.”

Evidence status: established · Specificity: direct · Confidence: high · Tag: observed
Sources: Sullivan, Elzinga & Raber 2013 (J Toxicology 2013:378168); Health Canada 2017 advisories · Agreement: gemini-topic, editor re-anchored

R03-C027 — “Illicit cannabis is commonly laced with fentanyl” is unsupported: the one laboratory-confirmed US sample (Connecticut, 2021) was later characterized as likely isolated cross-contamination, and person-level toxicology cannot identify a laced product — while “it never occurs” is also too strong.

Evidence status: established (calibrated absence) · Specificity: direct · Confidence: high · Tag: observed (forensic)
Sources: CT DPH/ORS reporting 2021–22; CDC/APHL interpretive guidance · Agreement: both
Notes: Myth M3 verdict, both directions guarded.

R03-C028 — Labor exploitation in illicit cultivation is a documented case phenomenon, not an established prevalence: indictments allege smuggled workers and passport withholding (D. Mass. 2025), while Maine investigators reported no trafficking evidence across >40 searches of a separate grow cluster.

Evidence status: probable (occurrence) / unknown (prevalence) · Specificity: direct · Confidence: high · Populations: P2/P4 cases · Tag: observed (judicial, allegation-status preserved)
Sources: DOJ D. Mass. 2025; DOJ D. Maine 2024 · Agreement: both

F. Gaps & monitoring

Evidence status: unknown (verified absences) · Specificity: direct · Confidence: high
Sources: convergent gap lists, both runs · Agreement: both

R03-C030 — Institutional crop monitoring exists in cannabis history (UNODC’s Morocco surveys: 134,000 ha, ~3,070 t resin, 2003) but nothing comparable operates for Americas herb production; academic satellite mapping of California outdoor/greenhouse cultivation exists (and underpins C019); indoor cultivation is a major structural blind spot for satellite crop mapping, while outdoor and greenhouse cultivation can be mapped in some settings (small, canopy-obscured, or resolution-limited outdoor sites can still be missed) — the correct claim is “no recurring institutional survey program was identified,” not “invisible from space.”

Evidence status: established · Specificity: direct · Confidence: high · Tag: observed
Sources: UNODC Morocco 2003; Butsic & Brenner 2016; Berkeley CRC 2026 · Agreement: post-correction convergence (Gemini’s CC025 recalibrated)

Myth verdict index (details: critique.md + reports)

M1 legalization eliminates / did nothing — both absolutes fail (C014, C017, C019, C006) · M2 “X% illicit” — category-laundered unless the metric and method travel with it (C005, C006, C020) · M3 fentanyl lacing — “commonly” unsupported, “never” too strong (C027) · M4 legal=tested=safe — false as absolute; asymmetric advantage real (C025) · M5 cartels run most — existence documented, “most” unsupported (C009) · M6 villains/stewards — moral categories collapse populations (C008, C023) · M7 enforcement reduces supply — site-level removal true, aggregate effect unidentified (C021) · M8 seizures track production — unsupported without a detection bridge (C002) · M9 home-grow feeds the black market — cultivation increases documented, commercial diversion unmeasured (C015)

Changelog

v1.0 (2026-09-02) — Danny sign-off after ChatGPT’s full-ledger review. Eight calibrations applied: C003 absence scoped to jurisdictions examined; C006 “best”→“most reconstructable identified”; C011 “not simply a migration”; C014 Hammond-vs-StatCan flagged as methodologically distinct, not a trend; C019 deterrence quote neutralized to observed association; C023 owl figures given sampling denominators; C025 asymmetry localized to the Health Canada comparison; C030 satellite blind spot bounded (indoor major blind spot; outdoor/greenhouse mappable in some settings, with residual misses). No grading changes. S-136–S-165 registered.
v0.1 (2026-09-02) — 30 claims drafted by editor from verified material: ChatGPT run (0 identity failures, 4th clean cycle) + Gemini run as re-anchored through two correction passes. New standing fields debut: observation-class tags and named-metric rule.

Source file in the research repository: research/reports/R03-legal-illicit-production/ledger.md. Private links, some names and practitioner-record passages are removed from the public copy, and broken character encoding from file conversion is repaired; nothing else is changed.