Claim Ledger — R04: Environmental Footprint of Cannabis Cultivation

Version: v1.0 — signed off by Danny September 7, 2026 (approval conditional on ChatGPT’s five wording calibrations, all applied, plus two optional polish items, applied; no identity or claim-support failures found in review) · Last updated: September 7, 2026 · Brief: v1.0 FROZEN 2026-09-06 · Protocol: v1.4 Grading: evidence status × cannabis-specificity × confidence per PROTOCOL §3, plus the standing R03 fields (observation-class tag; named-metric rule) and the R04 brief’s additions: every load-bearing claim carries an evidence-type tag — measured / hybrid (measured inputs → modeled outcome) / modeled / observed-administrative / regulatory / absence (searched, none located within the documented search) — and every quantitative claim states its boundary · functional unit · impact category/method · population/system inline (the brief’s eleventh field, “N/A — reason” where no characterization applies). Populations: the R03 spectrum tags are reused where a claim is population-specific (P1 legacy/private-land outdoor · P3 trespass/public-land · P5 licensed commercial) alongside the production system (indoor warehouse · supplemental-light greenhouse · light-dep · outdoor container · outdoor in-ground). Sources: cited by verified tuple; registered as S-166–S-197 in SOURCE_LIBRARY (R04 section); 18 further citations reuse existing entries. Every citation passed R04’s two-check verification (critique.md, verification/). Gemini’s material enters only via editor-verified tuples (verify-D). Inheritance: R05-C001–C010 (energy/carbon spread), C011–C014, C017, C031–C033 are inherited, not restated; R04 claims cite them. R05-C013 was corrected 2026-09-06 (rule 10/13) and R04-C017 carries the corrected range.

A. The map, not the number

R04-C001 — Keystone. There is no defensible single “environmental footprint of cannabis.” Every published figure is specific to a system boundary (operational-only / cradle-to-farm-gate / cradle-to-consumer), a functional unit (kg dried flower, g cannabinoid, m² canopy-year, plant, $), an impact category and characterization method, and a production system and population; figures may be compared only where those four are compatible, and the corpus’s deliverable is the boundary map, not a pooled number.

Evidence status: established · Specificity: general · Confidence: high · Tag: methodological
Sources: R05-C032, R05-C033, R08-C027; the H19 boundary tables of both R04 runs · Agreement: both
Notes: The one attempt in this cycle to violate it (Gemini’s 22.8 kg CO₂e/kg → “300 feet of driving”) produced a number absent from its own source (critique §3.1). Lineage: R02-C002 (no defensible global production figure), R03-C001 (which population, measured how).

R04-C002 — The located cannabis-cultivation LCA literature consists of one modeled US indoor national assessment (S-001/S-131), one Canadian industry model (S-004), one outdoor field-data LCA (S-003, container system, Quebec), a 2026 drying LCA that transfers the S-003 cultivation inventory (S-183), a Washington master’s thesis with outdoor/mixed/indoor producer cases (S-184), and adjacent hemp and prototype studies (S-185, S-186); no LCA of supplemental-light greenhouse, light-deprivation, or in-ground full-sun drug-type cannabis was located within the documented search.

Evidence status: established (state of the literature) · Specificity: direct · Confidence: high · Tag: absence (bounded)
Sources: S-001, S-003, S-004, S-183–S-186; ChatGPT A1 inventory; Gemini master inventory · Agreement: both
Notes: Extends R05-C009 (greenhouse subtypes collapsed). The Bottem thesis means “no mixed-system case exists” is too strong; “no peer-reviewed, system-resolved coverage” is the defensible form. Search cutoff Sept 6, 2026; English; web-indexed databases, not native Scopus/WoS.

Evidence status: established · Specificity: direct · Confidence: high · Tag: methodological (rule 12)
Sources: ChatGPT A3 independence map; Gemini shared-dataset mapping; SOURCE_LIBRARY family notes (S-173, S-178) · Agreement: both
Notes: “Multiple studies find…” must name distinct datasets. Registered families are marked in the library.

B. Energy and carbon — inherited axis, updated and traced

Evidence status: established (provenance) · Specificity: direct · Confidence: high · Tag: modeled (2012)
Boundary: US indoor cultivation electricity vs national electricity denominator, ~2011; LCIA N/A — share of kWh
Sources: S-006, S-188, S-005 (2025 sector scale) · Agreement: both (Gemini’s “$3B” corrected to $6B)
Notes: R05-C033 already bars quoting Mills 2012 as current. Mills 2025’s 595 PJ/yr and ~44 Mt CO₂e/yr (R05-C010) are sector-scale primary energy and emissions, not an electricity share — different metric, different year.

R04-C005 — “One joint equals X miles of driving” has no universal value: the arithmetic depends on flower mass, production system, grid year, cannabis boundary and the vehicle’s boundary. Shown arithmetic on stated inputs: a 1 g indoor-grown joint at 2,283–5,184 kg CO₂e/kg (R05-C001) embodies 2.3–5.2 kg CO₂e ≈ 5.6–12.8 miles at EPA’s 0.404 kg CO₂e/mile; a 1 g outdoor-container joint at 61.8–110.7 kg CO₂e/kg (S-003) embodies 0.062–0.111 kg ≈ 0.15–0.27 miles. The EUDA’s published illustration (S-189) is an agency recalculation from older indoor assumptions, not measurement.

Evidence status: probable (as arithmetic on modeled inputs) · Specificity: direct · Confidence: medium · Tag: modeled inference
Boundary: cradle-to-farm-gate cannabis (S-001 US grids 2021; S-003 Quebec 2020–22) vs tailpipe-only vehicle factor; GWP100 (TRACI 2.1 / ReCiPe midpoint H) — mixed methods, illustrative only
Sources: S-001, S-003, S-189, EPA equivalency factor · Agreement: both (Gemini’s outdoor figure recomputed after correction)
Notes: Any downstream use must carry both boundaries and the grid year. Never quote as “a joint = X miles.”

R04-C006 — Publications since January 2025 (S-183 drying LCA; S-184 Washington thesis; S-185 Czech hemp LCA; S-186 Thai condensate prototype; S-021 cross-crop synthesis) broaden the cases and process questions available but do not justify replacing or pooling the inherited R05 carbon range: none supplies a representative new construction/equipment inventory, a greenhouse-subtype factor, or a drug-type field footprint.

Evidence status: established · Specificity: direct · Confidence: high · Tag: absence (bounded)
Sources: S-021, S-183–S-186; R05-C001–C010 · Agreement: chatgpt (Gemini did not review the 2025–26 cluster)

R04-C007 — Drying is a distinct GWP endpoint from cultivation: in the only located drying LCA (S-183, IMPACT 2002+, Saskatchewan-adapted inventory), conventional controlled-environment drying contributed 11.31 kg CO₂e per kg dried cannabis, and cold-plasma or microwave-infrared pretreatment reduced the drying-stage figure to 5.68 and 3.25 kg CO₂e (≈50% and ≈72% drying-stage reductions); the cultivation-inclusive reduction is far smaller (reported ≈6% in full text, not re-extracted by the verifier).

Evidence status: probable · Specificity: direct · Confidence: medium · Tag: hybrid (measured drying × transferred cultivation inventory)
Boundary: cultivation-through-drying, 1 kg dried material, packaging/retail excluded; IMPACT 2002+ (SimaPro 9); co-product treatment incompletely inspected
Sources: S-183 · Agreement: chatgpt-only
Notes: A drying-only weighted-damage reduction must never be substituted for a whole-production GWP reduction (ChatGPT K21). Boundary-orphaned for external comparison.

R04-C008 — Cross-crop and cross-product comparisons are defensible only on aligned boundaries and units: the per-growing-surface-year framing of the cross-CEA synthesis (S-021) supports energy comparisons among controlled-environment crops, but per-kg comparisons are distorted by dry cannabis flower (≈10–12% moisture) versus fresh produce (≈95% water), and no adequately aligned per-dose or per-dollar comparison of cannabis with other consumer goods was verified; food-LCA frames (S-197) are not boundary-matched.

Evidence status: established · Specificity: adjacent · Confidence: high · Tag: methodological + absence
Sources: S-021, S-197 · Agreement: both
Notes: “Cannabis is the most energy-intensive crop” is a per-area CEA statement at best (R05 sphere); it says nothing about total environmental harm or agricultural energy share.

C. Water

R04-C009 — The “22.7 L (6 gal) per plant per day” figure is a 2010 Humboldt County draft-ordinance assumption adopted in Bauer et al. 2015 (S-008; source range 3.8–56.8 L/plant/day), not a measurement, and the widely repeated “cannabis uses twice the water of wine grapes” descends from the same per-plant assumption (22.7 vs ≈12.64 L/day in S-008’s introduction) and from Carah et al. 2015’s area-season extrapolation (S-169) — a chain of assumptions, not a paired crop experiment.

Evidence status: established (provenance) · Specificity: direct · Confidence: high · Tag: modeled/assumed
Boundary: assumed daily plant demand; L plant⁻¹ day⁻¹ and L km⁻² season⁻¹; water application; LCIA N/A — direct volume assumption
Sources: S-008, S-169; R05-C014 · Agreement: both
Notes: Restates R05-C014 with the comparator chain. Carah’s ≈430 vs 271 million L km⁻² season⁻¹ figures are ChatGPT’s extraction, not re-extracted (PDF); direction verified.

R04-C010 — Direct measurement of cannabis irrigation exists at experimental scale: in an outdoor foil-tunnel trial of three CBD-rich chemotype-III genotypes (Germany, 2024), subsurface drip irrigation used 188.19 L m⁻² over the season versus 231.23 L m⁻² for surface drip (−18.6%), reduced weed dry biomass by 93.2%, and increased inflorescence yield by 5% and CBD concentration by 9%.

Evidence status: probable · Specificity: adjacent (CBD-type, tunnel) · Confidence: medium · Tag: measured
Boundary: applied irrigation only, one site-season; L m⁻² canopy season⁻¹; water application; LCIA N/A — direct measurement, no characterization; buried-infrastructure lifecycle not assessed
Sources: S-180 (PMC full text) · Agreement: both (Gemini via correction round)
Notes: A within-trial ordering (SDI < DI), not a universal farm-saving percentage (ChatGPT K07).

R04-C011 — The only located field evapotranspiration measurement for cannabis is an eddy-covariance study of floral hemp at low density (2,000 plants/ha; KwaZulu-Natal, Dec 2022–Apr 2023) reporting high water use and low water productivity; field ET includes soil evaporation and interrow vegetation and is neither plant-only transpiration nor blue-water withdrawal, so it cannot validate any per-plant irrigation figure.

Evidence status: probable · Specificity: adjacent (hemp, field) · Confidence: medium · Tag: measured
Boundary: field-area ET over the study period (ChatGPT extracted 377 mm; not re-extracted); mm; water use; LCIA N/A
Sources: S-181 · Agreement: chatgpt-only

R04-C012 — Indoor deficit-irrigation and leaf-level water-use-efficiency findings (S-182, high-CBD plants, Germany) show that moderate drought can maintain inflorescence yield, but leaf-level efficiency is not facility water productivity and no full-cycle facility saving can be derived from them.

Evidence status: probable · Specificity: direct · Confidence: medium · Tag: measured (abstract-level)
Sources: S-182 · Agreement: chatgpt-only

R04-C013 — Among 901 cultivators enrolled in the North Coast cannabis water program (2017 reports), wells were the most frequently reported source (58% of sites), then surface diversions (22%) and springs (16%); wells were reported by 68% of compliant versus 33% of non-compliant sites; these are frequencies of sites, not shares of water volume, and enrollment is a selected population.

Evidence status: established · Specificity: direct · Confidence: high · Tag: observed-administrative
Boundary: reported source per enrolled site, Northern California, 2017; % of sites; LCIA N/A — no impact characterization
Sources: S-167 · Agreement: both (Gemini’s “428 cultivators” corrected)
Notes: Frequent well use is not evidence that wells have no streamflow effect (ChatGPT K03); groundwater-lag is unquantified in this literature.

R04-C014 — Supplemental stored water shifts the seasonal pattern of extraction away from the pattern of demand, but enrolled farms generally lacked enough storage to forbear fully from growing-season surface diversion; California’s Cannabis Cultivation Policy has required forbearance from surface diversion during the dry season (April through October) since 2019, so compliance requires storage, reduced demand, or alternative sources.

Evidence status: established · Specificity: direct · Confidence: high · Tag: modeled (reports → storage/extraction model) + regulatory
Boundary: monthly extraction vs application, enrolled North Coast farms; regulatory calendar; LCIA N/A
Sources: S-009, S-193 · Agreement: both
Notes: Storage sufficiency decreased with cultivation area; storage construction/evaporation effects unassessed. Exact forbearance dates per policy Attachment A (not re-extracted).

R04-C015 — Modeled dry-season water extraction was consistently greater for unpermitted than permitted cannabis farms in Northern California; in scenario modeling, universal permitting with dry-season diversion limits would cut dry-season extraction by ≥50% in 34 of 115 study watersheds, and modest off-stream storage expansion by all farms would do so in 96 of 115.

Evidence status: probable · Specificity: direct · Confidence: medium · Tag: modeled (imagery + reports → prediction)
Boundary: watershed dry-season extraction, permitted vs unpermitted, 2017–20 training data; volume and % reduction; LCIA N/A
Sources: S-160 (Dillis 2023); R03-C024 · Agreement: both
Notes: Transfers permitted-farm parameters to unpermitted farms; population-specific to P1/P5 private-land outdoor.

R04-C016 — Keystone (water). In 14 upper watersheds across 10 Northern California counties (2022 demand vs 1992–2022 flows), cannabis was the top water user in only 71 of 4,367 catchment-months (1.6%), versus residential use (34.6%) and non-cannabis agriculture (29.8%); cannabis demand was more seasonal, more interannually variable, and more evenly spread; modeled potential streamflow pressure from residential and other agriculture appeared at annual scales, while cannabis-associated pressure appeared only at monthly scale, chiefly in dry years — so cannabis is a spatially and temporally concentrated modeled pressure, additive to larger users, not a uniformly dominant one.

Evidence status: probable · Specificity: direct · Confidence: high (as graded) · Tag: hybrid (mapping + inherited demand models vs historical flows)
Boundary: catchment-month demand relative to natural flow; % of catchment-months; potential streamflow pressure; hydrologic/statistical method, no LCIA; valley groundwater basins excluded; pumping lag omitted
Sources: S-166 (full text) · Agreement: both (Gemini via correction round, CC07)
Notes: First scale-contextualized cannabis water finding in the corpus. Demand is modeled, not metered; not evidence of measured aquatic mortality. Replaces both “devastating watersheds” and “negligible” rhetoric (brief M3/M5 sphere).

R04-C017 — In Bauer et al.’s four northwestern California study watersheds, modeled cannabis water demand (mapped plants × the 22.7 L assumption) ranged from 2–23% of annual seven-day low flow in the least-impacted watershed (Upper Redwood Creek) to 34–165% (Redwood Creek South) and 36–173% (Salmon Creek), and exceeded 100% of a near-zero minimum in Outlet Creek — demand exceeded low-season streamflow in three of four watersheds.

Evidence status: established · Specificity: direct · Confidence: high · Tag: modeled
Boundary: watershed demand vs gauged/modeled seven-day low flow, imagery 2011–12, private-land outdoor/greenhouse mix; %; streamflow pressure; LCIA N/A
Sources: S-008; R05-C013 (corrected 2026-09-06, R05…/verification/R05-C013-correction-2026-09-06.md) · Agreement: both
Notes: Carries the corrected R05 range. The demand parameter is the ordinance assumption in C009.

R04-C018 — In sealed indoor facilities, about 70% of applied irrigation water was transpired and recovered as HVAC condensate in the one metered field study (two Sacramento facilities), a recovery opportunity estimated at 17.8 million gal/yr statewide for California; recoverable is not reused, and no fully verified, representative commercial indoor water balance (external intake, RO reject, condensate reuse, nutrient discharge, vapor export) was located within the documented search.

Evidence status: probable (recovery); absence (balance) · Specificity: direct · Confidence: medium · Tag: measured (n=2) + absence
Boundary: operational facility water flows; % of applied water; LCIA N/A — direct measurement; treatment, re-mineralization and RO reject unaccounted
Sources: S-007; S-186 (prototype only); R05-C007 · Agreement: both (Gemini’s 80% upper bound withdrawn — second cycle of the same inflation)
Notes: Danny’s archive taught “drain the dehumidifier bin every day” (archive-audit §2) — the practitioner counterpart. The “6.1 L m⁻² day⁻¹” indoor figure Gemini attributed to the study is not in it.

D. Land, habitat, wildlife

R04-C019 — Systematic mapping of Humboldt County (2016) digitized 4,428 grow sites in 60 watersheds: 68% were >500 m from developed roads (a remoteness/fragmentation proxy, not a measured fragmentation outcome), 22% on steep slopes (erosion risk), 5% within 100 m of threatened-fish habitat, and an estimated 297,954 plants implied ≈700,000 m³ of water demand under the standard per-plant assumption; grows were clustered, implying disproportionate impacts in ecologically important locales.

Evidence status: established · Specificity: direct · Confidence: high · Tag: observed (mapping) + modeled (water)
Boundary: siting/land-cover proxies, Humboldt, Google Earth imagery (date not re-verified); counts and %; LCIA N/A; not erosion or mortality measurements
Sources: S-160 (Butsic & Brenner 2016) · Agreement: both (Gemini’s “68 watersheds” corrected — 60)

R04-C020 — Comparing cannabis agriculture with timber harvest in northern California (2000–2013), timber had the larger total landscape impact but cannabis produced far greater forest loss and fragmentation per unit area converted; secondary coverage reports ≈1.5× the forest loss and ≈2.5× the fragmentation per unit area, multipliers not verified against the paywalled full text.

Evidence status: probable · Specificity: direct · Confidence: medium · Tag: observed (land-cover analysis)
Boundary: landscape change per unit area, mixed legal status, 2000–13; fragmentation metrics; LCIA N/A
Sources: S-170 (abstract; multipliers secondary) · Agreement: both
Notes: Fragmentation intensity and total converted area answer different scale questions (ChatGPT K10); Gemini’s “0.16 ha mean clearing” is unverified.

Evidence status: established (absence) · Specificity: direct · Confidence: high · Tag: absence
Sources: R05-C031, R02-C022; DCC license structure (S-013); provenance search (verify-B) · Agreement: both (after Gemini’s withdrawal)
Notes: Local mapping (C019) establishes local extent, not a national share. The absence is the finding.

R04-C022 — Post-legalization spatial analysis of California (2018–24) finds licensed-farm presence associated with lower unlicensed cultivation and no suppression from local bans; these are observational associations and do not establish a causal environmental benefit of licensing, because enforcement, markets, bans and site conditions change together.

Evidence status: probable (association) · Specificity: direct · Confidence: medium · Tag: modeled (mapping → GAMs)
Sources: S-171; S-138 (R03) · Agreement: chatgpt-only
Notes: R03-C019 lineage; causal rung 1–2.

Evidence status: established · Specificity: direct · Confidence: high · Tag: measured (assays/necropsies) + modeled (habitat overlap)
Boundary: exposure/diagnosis among recovered animals, California/S. Oregon, 2007–2021; % of tested; ecotoxicity; LCIA N/A; attribution to cannabis per site, not per animal
Sources: S-158, S-159; R03-C023 · Agreement: both
Notes: Inherited from R03 with the two-paper fisher split (79%/85% never pooled). Franklin 2018 (S-174) is a single-carcass case note, not a prevalence source.

R04-C024 — Source attribution for wildlife rodenticide exposure is unresolved outside trespass sites: 62% of 127 California barred owls and hybrids tested positive for anticoagulant rodenticides with no clear link to illegal cannabis cultivation and a significant association with the wildland–urban interface, leaving dwellings versus cultivation as competing sources.

Evidence status: established · Specificity: direct · Confidence: high · Tag: measured + spatial association
Sources: S-172 · Agreement: chatgpt-only
Notes: The counterweight to C023: trespass-site evidence cannot characterize exposure across forests generally.

Evidence status: probable · Specificity: direct · Confidence: low · Tag: measured (cameras) + modeled
Sources: S-173 · Agreement: chatgpt-only
Notes: Fencing-specific effects and cannabis-specific erosion magnitudes remain unverified in the located literature.

E. Chemical inputs, nutrients, air

R04-C026 — Environmental pesticide fate has been measured only at trespass sites: passive samplers confirmed diazinon and carbofuran in headwater streams immediately downstream of eradicated public-land complexes in four National Forests (diazinon transported up to 186 m; carbofuran detected >490 days after last estimated application), with no detections in higher-order catchments below private-land cultivation gradients; a 2025 USGS study found residues in soil at inactive trespass sites and none in companion stream samples. No environmental-fate monitoring of licensed farms was located; product-residue testing is not fate monitoring.

Evidence status: established (P3); absence (P5) · Specificity: direct · Confidence: high · Tag: measured
Boundary: POCIS time-weighted soluble fractions; soil/sediment/water residues; detection, not injury; LCIA N/A
Sources: S-020, S-019 · Agreement: both (Gemini’s “11% of stations” withdrawn — not in the source)
Notes: R05-C017 lineage. Detection ≠ measured ecological injury; non-detection is bounded by method and sampling.

R04-C027 — No pesticide is registered for cannabis in the United States; California permits products whose active ingredients are exempt from residue tolerances and either exempt from registration or usable without conflicting with a registered label (DPR ENF 24-19), publishes a list of registered products reviewed for similar crops (25(b) products are excluded from the list), and warns that permitted products can still fail residue testing on rate and timing — so “zero synthetic pesticides permitted” and “legal means clean” are both misstatements of the instrument.

Evidence status: established · Specificity: direct · Confidence: high · Tag: regulatory
Sources: S-192; S-017 (illicit-indoor residues as the contrast) · Agreement: both (Gemini’s wording corrected twice)
Notes: Danny’s 2016–2019 input lists (neem, azadirachtin, karanja, Bt, silicate, DE, predators) fall in the admitted class; the archive’s Prop 215-era testimony (“ornamental pesticides, PGRs”) is practitioner evidence, anecdotal/direct, population-tagged (archive-audit).

R04-C028 — The only measured nutrient-loss study located is a controlled-environment trial of high-CBD hemp: tripling root-zone phosphorus (25 → 75 mg P/L) increased cumulative leachate P 12-fold with no effect on flower yield or cannabinoid concentration, while phosphorus-use efficiency rose from 31% to 80% as input fell — avoiding excess P was a measured input reduction with no observed yield or cannabinoid penalty in this trial; no field study of nitrate, phosphate or potassium leaching from commercial cannabis was located.

Evidence status: probable · Specificity: adjacent (high-CBD, chamber) · Confidence: medium · Tag: measured
Boundary: leachate P per growing area over eight reproductive weeks; g P m⁻²; nutrient discharge; LCIA N/A — physical loss, not characterized eutrophication; two trays per treatment; industry funding disclosed
Sources: S-175 · Agreement: both (Gemini’s data-gap register agrees on the field-study absence)
Notes: R01-C008/C020 (N asynchrony) supply a mechanism for loss; a timing mismatch is not a quantified emission until discharge or a reconstructable fate model connects it to an endpoint.

R04-C029 — In the outdoor field-data LCA (S-003, Quebec 2020–22), the higher-fertilizer treatment (H–) had the lower footprint per kg of dry flower (61.8 vs 110.7 kg CO₂e/kg for L+) because it needed about half as many plants, and hence half the potting media, per kilogram; the low-N/high-K treatment raised THC concentration ≈30% via nitrogen deficiency, and switching the functional unit to 100 g THC narrowed but did not reverse the ordering (60.7 vs 91.7 kg CO₂e per 100 g THC) — so “less fertilizer = lower footprint” is not established at the per-product level.

Evidence status: probable · Specificity: direct · Confidence: medium · Tag: hybrid (measured trial inputs/yields → modeled LCA)
Boundary: cradle-to-farm-gate incl. drying; 1 kg dry flower and 100 g THC; GWP (ReCiPe midpoint hierarchist v1.13, OpenLCA 2.0); outdoor container, single site; allocation to flower per paper (supplement not audited)
Sources: S-003 (PMC full text) · Agreement: both (Gemini’s “22.8 kg/kg” and kg N/ha rates were not in the paper and were withdrawn)
Notes: R05-C012’s “N-heavy fertilization reduced cannabinoid content” stands as chemistry; the footprint conclusion depends on the unit.

R04-C030 — Potting media contributed 65–75% of GWP in both treatments of the outdoor container LCA; a modeled on-site steam-sterilization reuse scenario cut the H– footprint from 60.7 to 52.7 kg CO₂e per 100 g THC (media emissions 41.5 → 4.2 kg CO₂e per functional unit; transport −85%), with diesel steaming becoming the largest remaining term; perlite production dominated fossil depletion (40–45%). Media reuse was a substantial modeled carbon lever in this outdoor-container system; as a separate cross-system observation, media are a minor indoor term (R08-C027), and no bed or in-ground system was assessed.

Evidence status: probable · Specificity: direct · Confidence: medium · Tag: hybrid/modeled scenario
Boundary: as C029; scenario assumes 26 L diesel per m³ media; GWP and fossil-depletion categories separately — never read a fossil/metal-depletion share as GWP (the R08-C027 audit lesson)
Sources: S-003; R05-C012; R08-C027; S-110 (peat reuse repeatability, adjacent) · Agreement: both
Notes: Media’s share of GWP is not the reduction achievable by reuse (ChatGPT K20). Danny’s 2016 “pots over in-ground” preference is the practice this claim grades (archive-audit).

R04-C031 — Cannabis terpene emissions are measured and their regional ozone effect is modeled as sub-ppb in the studied Colorado scenarios: leaf-enclosure emission capacities of 4.9–8.7 µgC g⁻¹ h⁻¹ across four strains (β-myrcene and eucalyptol dominant); a Colorado inventory of 66–657 t/yr of terpenes; +362 t/yr in Denver County modeled to raise hourly ozone by up to 0.34 ppb (morning) and 0.67 ppb (night), with sensitivities of 1 ppb per 1,000 t/yr daytime and +0.30 ppb MDA8 per 1,000 t/yr; measured facility exhaust of 2.13–11.12 lb VOC per ton harvested with no isoprene; in-room concentrations of 110–5,500 µg m⁻³. Regional air-quality significance is location- and inventory-dependent and not established as large; odor nuisance is a separate, real question.

Evidence status: established (measurement); probable (regional modeling) · Specificity: direct · Confidence: high · Tag: measured + modeled
Boundary: leaf enclosure / facility exhaust / room air / regional CAMx model; µgC g⁻¹ h⁻¹, t yr⁻¹, ppb; ozone-formation potential; inventories “highly uncertain”
Sources: S-177, S-176, S-178, S-179 · Agreement: both (Gemini’s 0.1–0.3 ppb and 0.5–2.5 µgC figures corrected to the sources)
Notes: Room generation rates (S-178, 2023) are not outdoor release rates; the two Urso papers are one evidence family.

R04-C032 — No separately verified net-impact study of CO₂ enrichment in cannabis cultivation was located; enrichment gas is an input in the inherited LCA frameworks, plant uptake is temporary and not sequestration, and the fraction vented is unquantified in the literature — no generic credit or debit for “recovered” or “vented” CO₂ is assigned.

Evidence status: unknown · Specificity: direct · Confidence: low · Tag: absence
Sources: ChatGPT §E; Gemini mechanism (venting figure withdrawn) · Agreement: both
Notes: S-004’s “industrial symbiosis” (co-location with CO₂ emitters) is a proposal, not a measured effect.

F. Waste and materials

R04-C033 — No representative mass-flow inventory of cannabis growing media, agricultural plastics, lamps, plant waste, or farm-gate packaging was located within the documented search; Canadian packaging figures in the legislative review are administrative activity indicators, not measured tonnage or lifecycle impact; no transparent retail-packaging LCA was located, so “packaging is the industry’s biggest environmental impact” is unsupported — and that does not establish that packaging is negligible.

Evidence status: established (absence) · Specificity: direct · Confidence: high · Tag: absence
Sources: S-194; ChatGPT §F; Gemini data-gap register (its “<1–3% of GWP” packaging figure was not in S-004 and was withdrawn) · Agreement: both
Notes: Danny’s 2024 “hydroponics leads to a lot of waste (rockwool)” is an assertion the literature also cannot quantify (archive-audit). Lamp replacement (“every 3 cycles,” Luna 2020) is a real, unmeasured stream.

R04-C034 — California’s cannabis waste rule (4 CCR §17223) expressly permits on-premises composting, organic-waste collection, delivery to permitted compost/digestion/landfill facilities, and reintroduction to agricultural land by tilling or no-till; the “no more than 50% cannabis by volume” rule governs how waste is rendered unrecognizable, not where it goes — so claims that state rules “force cannabis biomass into landfills” are false for California. Separately, Health Canada’s March 2025 streamlining removed certain cultivation-waste witness/recordkeeping and trimmings-reporting obligations, creating a discontinuity in reported waste trends.

Evidence status: established · Specificity: direct · Confidence: high · Tag: regulatory
Sources: S-191, S-194 · Agreement: conflict resolved (Gemini’s landfill-mandate reading withdrawn after the text was quoted)
Notes: Rule existence ≠ measured diversion outcome; no California diversion data were located.

G. Mitigation and governance

R04-C035 — Mitigation levers rank only within an impact category and only on a shared baseline: for carbon, across several non-equivalent models, shifting production away from energy-intensive indoor systems toward outdoor production consistently produces large modeled GHG reductions — the reported effect sizes are not pooled because their baselines and boundaries differ (examples, individually: a −76% US sector scenario, R05-C010; ~90% per kg in a Canadian comparative model, R05-C003; “50 times less” per kg in the outdoor field-data LCA’s abstract, S-003) — and, in the one outdoor-container system modeled, media reuse (C030); LED/HVAC/renewable procurement are inherited from R05 without new effect sizes; for water quantity, subsurface drip (−18.6% in one trial, C010) and, for timing, storage/forbearance (modeled, C014–C015); for chemical loss, avoiding excess P (measured, C028); for habitat and waste, no comparable intervention effect sizes exist. Cross-category tradeoffs are real (outdoor shift → diversion timing, siting, media; LED → winter heating, R05-C005), and no overall ranking is defensible without an explicit, sourced weighting.

Evidence status: established (as a ranking rule and inventory) · Specificity: direct · Confidence: high · Tag: methodological
Sources: R05-C003, C004, C005, C010; S-003; S-180; S-175; S-009/S-160 · Agreement: both (Gemini’s cross-category ranking table withdrawn in favor of per-category)
Notes: Adoption, certification or installed equipment alone never establishes benefit (brief; ChatGPT G1).

R04-C036 — Regulatory instruments establish requirements, not measured environmental effects: California’s DCC-2025-01-R (approved April 28, 2026; effective July 1, 2026) deleted the renewal electricity-reporting and carbon-offset-purchase requirements and the pest-management-plan submission; Massachusetts’ lighting power/efficacy standards are not facility carbon caps; Illinois’ statute sets equipment compliance paths; California’s Water Board policy sets diversion timing and source-specific conditions; and no causal evaluation of any instrument’s environmental outcome was located.

Evidence status: established · Specificity: direct · Confidence: high · Tag: regulatory
Sources: S-190, S-195, S-193 · Agreement: chatgpt-only (Gemini’s regulatory section predates the July 2026 change)
Notes: Descriptions of California electricity reporting and offsets are historical after July 1, 2026 (ChatGPT K23).

R04-C037 — Certification and marketing schemes (Sun+Earth, with a CCOF inspection partnership announced April 2026; Clean Green; Envirocann; California’s OCal) establish practice criteria, inspection and traceability, and “carbon-neutral” labels require an inventory scope, period, residual emissions and credit evidence — none verified establishes a numerical footprint improvement per unit product, and providers have adoption incentives that are context, not disqualification.

Evidence status: established (program description) · Specificity: direct · Confidence: high · Tag: observed (program records)
Sources: S-196 · Agreement: both
Notes: Practice certification ≠ measured benefit (ChatGPT K25; brief practice–effect separation).

Public-claim verdict index (brief M1–M10; details: critique.md, both reports)

M1 “1% of US electricity” — historical modeled claim (2012), unsupported as a current fact (C004) · M2 joint = X miles — no universal equivalence; boundary-dependent shown arithmetic only (C005) · M3 twice the water of grapes — assumption chain, not a crop experiment (C009); the scale question is answered by C016 · M4 sun-grown sustainable by definition — unsupported as a label; media, nitrogen, timing, siting (C029, C030, C017, C019) · M5 illegal grows are the problem / legal grows are clean — both blanket forms unsupported; P3 evidence is population-specific (C023, C024, C026, C027) · M6 cannabis/hemp is carbon-negative — not established for drug-type flower; product- and boundary-dependent (S-187; C032) · M7 terpenes are a regional air-quality problem — small modeled regional effect, real nuisance, location-dependent (C031) · M8 packaging is the biggest impact — unsupported within the documented evidence; not shown negligible (C033) · M9 LEDs solve indoor carbon — unsupported as a complete solution (R05-C004/C005; C006) · M10 organic/living-soil has a lower footprint — not established by label; unit-dependent (C029; S-045; R01-C033).

Data gaps (searched and none located, September 6, 2026 cutoff)

Field nutrient-leaching studies in commercial cannabis · in-ground full-sun and greenhouse-subtype LCAs · representative commercial indoor water balance · farm-gate packaging, plastics, lamp and plant-waste mass flows · transparent retail-packaging LCA · national/global occupied-area denominator · causal evaluations of regulatory instruments and certifications · CO₂-enrichment net-impact study · licensed-farm pesticide-fate monitoring · home-scale (6-plant) footprint of any kind (R21 handoff).

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

Practitioner-record section withheld from the public copy.

Changelog

v1.0 (2026-09-07) — ChatGPT full-ledger review: conditional approval, no grading changes. Five calibrations applied: C016 “impair streamflow” → “modeled potential streamflow pressure”; C023 “best-documented harms” bounded to “within the literature located for R04, the strongest direct evidence of wildlife toxicological harm”; C028 “no-cost mitigation” → “no observed yield or cannabinoid penalty in this trial”; C030 “large carbon lever for outdoor container systems” localized to “in this outdoor-container system,” indoor contrast made a separate observation; C035 outdoor-shift effect sizes un-pooled and cited individually (the claim had violated its own comparison rule). Optional polish applied: C031 “small” → “sub-ppb in the studied Colorado scenarios”; C019 road distance labeled a remoteness/fragmentation proxy. Evidence-type tag and inline boundary line adopted as standing PROTOCOL fields (§3; ChatGPT and Danny concurring). Danny’s sign-off stamped 2026-09-07.
v0.1 (2026-09-06) — 37 claims drafted by editor from verified material: ChatGPT run (0 identity failures, fifth clean cycle; 0 claim-support failures) + Gemini run as re-anchored through one correction pass (verify-D); three gates closed before drafting (archive audit; Gemini corrections; R05-C013 independent second check). New standing fields debut: evidence-type tag and inline boundary/unit/category line. S-166–S-197 registered.

Source file in the research repository: research/reports/R04-environmental-footprint/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.