Generated 2026-10-02. All numbers below are computed from the source files; nothing was imputed.

Downloads: summary by strain (CSV) and full aggregates (JSON). Guides with the original test rows provide a downloadable per-test CSV. Guides with only the published aggregates provide a labelled JSON summary and omit per-test histograms.

1. Source

Dataset: Cannlytics, cannabis_results — https://huggingface.co/datasets/cannlytics/cannabis_results
License: Creative Commons Attribution 4.0 International (CC BY 4.0) — https://creativecommons.org/licenses/by/4.0/
Repository version used: lastModified 2026-02-02T19:10:03Z, commit 8724df97f3c40b3f844157606dd91e97509a8cb9.
Files (per-file last-commit date from the Hugging Face API):

State	File	Bytes	Last commit	Rows read	Name fields in file	Flower tests used
CA	`data/ca/ca-results-latest.csv`	1,585,774,619	2026-02-02	85,487	strain_name, product_name, product	410
FL	`data/fl/fl-results-latest.csv`	305,684,201	2026-02-02	16,309	strain_name, product_name	43
NV	`data/nv/nv-results-latest.csv`	98,519,804	2024-08-16	153,064	product_name	4,520
OR	`data/or/or-results-latest.csv`	38,932,645	2024-08-16	196,900	— (none)	0
MA	`data/ma/ma-results-latest.csv`	35,460,727	2024-08-16	75,164	product_name	350
MD	`data/md/md-results-latest.csv`	32,885,944	2024-08-16	105,013	strain_name	4,394
HI	`data/hi/hi-results-latest.csv`	27,109,389	2024-08-16	13,485	strain_name, product_name	257
MI	`data/mi/mi-results-latest.csv`	13,171,269	2024-08-16	89,956	product_name	3,707
UT	`data/ut/ut-results-latest.csv`	9,639,780	2024-08-16	1,230	strain_name, product_name	22
RI	`data/ri/ri-results-latest.csv`	7,318,634	2024-08-16	25,832	— (none)	0
NY	`data/ny/ny-results-latest.csv`	2,374,548	2024-08-16	330	strain_name, product_name	10
CO	`data/co/co-results-latest.xlsx`	13,405,514	2024-08-16	25,798	product_name	138
CT	`data/ct/ct-results-latest.xlsx`	74,288,499	2024-08-16	19,963	product_name	18
WA	`data/wa/wa-results-latest.xlsx`	56,124,010	2024-08-16	202,812	strain_name, product_name, inventory_name, product_description	3,924

Files deliberately not used: data/all/all-results-latest.csv (2.5 GB; the per-state files were used instead); the .xlsx twins of the CSV files; and the newer Connecticut COA exports data/ct/ct-coa-data-*.xlsx (last commit 2026-02-02), which were not used in this release. The CT figures here come only from ct-results-latest.xlsx (sheet Values, 19,963 records = the full file).

The dataset README notes that the data are a non-random subset of all lab results and that data about individuals may not be used commercially. The public CSV produced here contains no producer, licensee or personal fields.

How to cite (use exactly)

Source: Cannlytics, cannabis_results (CC BY 4.0), analysis by Official Cannabis.

CC BY 4.0 also asks for a link to the license and an indication that changes were made; on the web page, link “CC BY 4.0” to the license URL above and note that the data were filtered, de-duplicated and aggregated.

2. Pipeline

Streaming. Every CSV was streamed over HTTPS straight into a CSV parser (the 1.59 GB CA file was never written to disk). The three xlsx-only states (CO, CT, WA) were downloaded temporarily, read with openpyxl and deleted. A deliberately broad regex pre-filter kept every row whose name fields mention any strain stem (e.g. blue\W*dream, chem, diesel, g\W?13), so strict matching could be redone offline.
Name field. strain_name if present, else product_name (WA: strain_name, checked against the product name / inventory name; UT: strain parsed from the “Strain: … Lab Tracking” text when needed).
Strict matching (section 3), product-type filter (section 4), value extraction (section 5), de-duplication (section 6), plausibility filter (0 < total THC ≤ 40 %), then statistics (section 7).

3. Strain matching rules

Normalisation. Lower-case; accents and curly quotes folded; possessives collapsed (lamb's → lambs); content in parentheses/brackets set aside; compound packaging words joined (pre-roll → preroll, pop corn → popcorn, fresh frozen → frozen, sugar shake → shake); tokens split on any punctuation. Aliases may be written with or without spaces/hyphens (BlueDream, G-13, C-99, Chem-Dawg all match).

STRICT match = after normalisation the name consists of exactly one alias (contiguous tokens) plus only non-name tokens, before or after it. Non-name tokens are: weights and fractions (3.5g, 1/8, oz, lb, eighth, dime …), numbers and alphanumeric lot/batch/test codes (#2, T18.8, TLMB1206202201), packaging and form words for flower (flower, bud(s), smalls, popcorn, bulk, lot, jar, pouch, sample, test, QA …), quality grades (A/B, AAA, tier, top shelf, premium), cultivation words (indoor, outdoor, greenhouse, light-dep, organic …), lineage words (sativa, indica, hybrid, S/I/H, fem), and 1–2-letter codes (except name-like abbreviations such as OG, NL, SD, NY, LA, SF, UK).

Excluded (counted as near-matches): any extra strain-name word before or after the alias (Santa Cruz Blue Dream, Super Sour Diesel, Blue Dream OG, Platinum GSC, Gelato Cake, Royal Wedding Cake); any cross marker (x, cross, bx, f1/f2, s1); names containing two different target strains; glued variants (Blue Dreamsicle); CBD variants (CBD Blue Dream, Lambsbread (CBD)); a cross or another strain inside parentheses. In rows whose source product type is flower, concentrate/edible words are treated as name words (so Blueberry Crumble — a different cultivar — is excluded rather than mistaken for a concentrate).

Brand / producer prefixes stripped only where obvious: (a) a first segment before a separator (-, |, :) that recurs in front of ≥3 different product names in the same state file (e.g. Cloud Cover - GSC - Bulk, Virtue - G13 - Lab Sample); (b) leading words equal to the leading words of the row’s producer/brand field (≥2 words, or one ≥4-letter word that is not a common strain word); (c) a short curated list of brand/descriptor names verified in the data (none is a strain word): travis x james, true humboldt, true mom pop, true mom & pop, almora farm, oz farms, joy ridge, circle s farms, summit buds, cloud cover, franklin fields, charlie walter, luman chichester, med to au transfer, proficiency test, for packaged flower, farmer and the felon, time machine, trb, tcs, summa, sst, cannasol, curaleaf, virtue, kushman, wcc, wwp, led, hps. Examples: TRB = Tahoe-Reno Botanicals (producer field), SUMMA = Euphoria Wellness, Kushman = one NV producer’s line used in front of many different strains. (d) breeder-of-record prefixes, accepted only in front of the cultivar that breeder released: DJ Short + Blueberry, Ken's + Granddaddy Purple, Josh D + OG Kush.

Alias table: each strain’s common names plus obvious spelling and spacing variants found in the data (sour deisel, jack herrer, grand daddy purple, train wreck, girl scout cookie, northern light, lamb breath …); glued/hyphenated forms (BlueDream, G-13) are matched automatically:

Strain	Group	Aliases matched
Blue Dream	network	blue dream
Sour Diesel	network	sour diesel, sour d, sour deez, sour deisel, sour diesal
Green Crack	network	green crack, green crush, green cush
Northern Lights	network	northern lights, northern light
Purple Urkle	network	purple urkle, purple urkel
Blue Cheese	network	blue cheese
Cinderella 99	network	cinderella 99, c99, cindy 99
Hindu Kush	network	hindu kush
Chemdawg	network	chemdawg, chemdog, chem dawg, chem dog, chem 91, chem 4, chem d
Grape Ape	network	grape ape
G13 Haze	network	g13 haze
G13	network	g13
Lemon Kush	network	lemon kush
Agent Orange	network	agent orange
Lamb’s Bread	network	lambs bread, lambsbread, lambs breath, lamb bread, lamb breath
Lime OG	network	lime og
Critical Jack	network	critical jack
Pineapple Express	comparison	pineapple express
OG Kush	comparison	og kush
Jack Herer	comparison	jack herer, jack herrer
Durban Poison	comparison	durban poison
Granddaddy Purple	comparison	granddaddy purple, grandaddy purple, grand daddy purple, granddaddy purp, grandaddy purp, grand daddy purp, gdp, granddaddy purps, grandaddy purps, grand daddy purps
Super Lemon Haze	comparison	super lemon haze
Girl Scout Cookies	comparison	girl scout cookies, girl scout cookie, gsc
Gelato	comparison	gelato
Wedding Cake	comparison	wedding cake
Blueberry	comparison	blueberry
Trainwreck	comparison	trainwreck, train wreck
Strawberry Cough	comparison	strawberry cough

Chemdawg sub-buckets. All strict Chemdawg matches form the main bucket; sub-buckets are plain (no designator), 91 (Chemdawg 91 / Chem 91 / Chem ’91), 4 (Chemdog #4 / Chem 4), d (Chem D / Chemdog D) and other_number (pheno-hunt/lot numbers such as Chem Dawg #173). Chem 4 and Chem D are accepted as the standard short forms of Chemdog #4 / Chemdog D; they are reported separately in the aggregates JSON (chemdawg.sub_buckets) so they can be dropped. G13 Haze and G13 are separate strains (G13 Haze never counts as G13).

strain_name vs product name. Where a row has both, the product name is checked: if it names a different target strain, a cross/variant of the strain, or another cultivar, the row is excluded as a conflict.

Washington. In the WA file the joined strain_name frequently contradicts the product / inventory name (e.g. strain “Jack Herer” on product “Flower Lot (A-bud) - A-bud - Nitro Cookies - (g)”). Among WA flower rows involving a target strain: 4,329 have both fields agreeing; 1,018 conflict; 422 have a product name for a target strain but a different strain_name; and 7,118 carry only a generic product name (“Flower Lot”). Because the join is unreliable, a WA test is used only when strain_name and the product/inventory name independently name the same strain; the generic-name rows are excluded and counted as wa_uncorroborated_flower.

4. Flower-only filter

Each row’s source product type is mapped to: flower, shake/trim, pre-roll, infused pre-roll, concentrate, edible/topical, wet/whole plant, hemp, other, unknown. Only “flower” enters the statistics. Flower types by state: CA Flower; Flower, Inhalable; Flower, Medical Inhalable; Flower, Product Inhalable; Plant (Flower - Cured) · CO Flower, Inhaled Product · CT Raw material (flower), Usable Marijuana, Flower… · FL Flower, Flower Inhalable · HI Flower Lot · MA flower · MD Buds · MI Flower · NV Marijuana Flowers/Buds, Small/Popcorn Buds, …-Packaged · NY Flower · UT Flower, Packaged Flower, Bulk Flower · WA Flower Lot, Flower Unlotted, Usable Cannabis/Marijuana (Leaf: from the [strain - type - size] bracket of the inventory name). Shake/trim, hemp and “Flower, Non-Inhalable” are excluded. A flower-typed row is re-classified (and excluded) when its name says pre-roll/joint/blunt/cone, infused, shake/trim/grind, fresh-frozen/whole-plant, or proficiency test. UT: most rows have no type; rows whose UDAF tracking number starts with “F” are treated as flower (all explicitly typed UT flower rows carry an F number) and flagged in flags.

5. Values and units

Total THC (%): the reported total if 0 < x ≤ 40. Otherwise, if absent, Δ9-THC + 0.877 × THCA (recorded per strain in thc_value_method). A reported total of 0 alongside positive THCA is treated as missing.
Units: values are % w/w. mg/g is converted (÷10) only with evidence: (i) the reported total is 10× the %-components (CA: Excelbis, SQRD, Decano), or (ii) all values sit on one mg/g scale with THCA between 100 and 400 (impossible as %; NV CANN Labs, a few HI/MD/WA rows). Other values above 40 are not “fixed” — they fail plausibility.
Plausibility: 0 < total THC ≤ 40 %; other tests are dropped and counted (flower_thc_le_0, flower_thc_gt_40).
Total CBD (%): reported total; if absent, or if it contradicts the lab’s own CBD/CBDA results (FL TerpLife Labs), CBD + 0.877 × CBDA. ND/<LOQ count as 0; empty/NT/NR as missing. MI has no CBD field.
Terpenes (% w/w): flat columns (NV, CT, NY, UT), COA results JSON (CA, CO, FL, MA; mg_g/10 or value with its unit), total-only (MD terpenes). Limonene = d-limonene/limonene; humulene = α-humulene; ocimene = reported ocimene, else the sum of α/β/cis/trans isomers. ND → 0. A test “has terpene data” only if at least one terpene (or total) is > 0 (all-zero NV rows are untested panels). Values above Cannlytics’ schema limits (5 % per terpene, 15 % total) are dropped. WA, MI and HI report no terpenes.
Dates: date_tested (NV fallback test_performed_date; WA CCRS test_date; WA Leaf created_at). Dates outside 2013-01-01 … 2026-02-02 are set to null.

6. De-duplication

The CA file contains the same certificate parsed several times (different sample_hash, same lot and values); Metrc states repeat one lab result on every child package. Within each strain, tests are dropped when they share (1) a sample_id, results_hash or WA lab_result_id, or (2) the same state, lab, producer, test date, total THC, total CBD, THCA and Δ9-THC (rounded to 0.001). Rule (2) is not applied to WA CCRS rows, whose potency is reported in whole or tenth percent and which have a unique lab_result_id per test.

7. Statistics

n_tests = de-duplicated, strictly matched, flower, plausible tests. Percentiles use linear interpolation. share_ge_25pct and share_ge_1pct are fractions (0–1). Per-state medians are given for states with n ≥ 10. Top-3 terpenes rank individual terpenes by median (ties by mean); n is given for each and a note flags fewer than 10 tests. max_single_lab_share = largest share of a strain’s tests (with a lab recorded) from one lab; lab IDs are state-specific (WA/MD license IDs, MI/MA anonymised in the source, HI none).

8. Exclusions and quality checks (per strain)

Near-matches and conflicts are counted over candidate rows of all product types before de-duplication.

Strain	Tests used	Near-matches excl. (all types)	…of which flower	Name conflicts	WA uncorroborated flower	Other product types	concentrate	shake/trim	pre-rolls	Duplicates removed	Flower, no THC value	THC ≤0 or >40
Blue Dream	2,118	780	311	319	537	1,347	946	179	143	267	87	1
Sour Diesel	1,071	2,304	1,449	228	405	642	320	170	84	65	44	11
Green Crack	1,236	156	56	347	815	352	216	91	20	123	25	0
Northern Lights	406	251	92	153	155	437	306	94	13	10	18	1
Purple Urkle	35	12	9	28	44	7	6	0	0	2	0	0
Blue Cheese	288	50	38	59	98	70	37	17	0	2	3	6
Cinderella 99	598	150	57	37	21	179	6	172	0	6	14	1
Hindu Kush	17	217	93	105	103	19	16	0	3	1	0	1
Chemdawg	668	1,244	790	105	222	428	141	192	57	91	23	0
Grape Ape	202	74	16	277	231	147	94	40	7	5	0	0
G13 Haze	14	2	1	4	1	0	0	0	0	0	0	0
G13	97	117	57	37	6	51	30	11	0	4	1	0
Lemon Kush	29	251	151	27	48	62	60	1	0	1	1	0
Agent Orange	140	9	3	29	167	85	16	66	1	3	7	1
Lamb’s Bread	117	174	90	32	38	19	11	0	8	4	9	0
Lime OG	29	35	10	0	0	39	33	2	0	1	0	0
Critical Jack	8	14	12	7	1	5	0	0	0	0	0	0
Pineapple Express	461	379	133	248	300	418	309	23	42	15	41	0
OG Kush	379	1,000	707	172	250	483	235	63	61	24	50	0
Jack Herer	910	297	102	257	132	568	344	120	25	18	48	0
Durban Poison	528	175	12	124	157	407	269	59	41	22	33	1
Granddaddy Purple	512	163	36	196	286	368	292	23	22	19	22	0
Super Lemon Haze	1,543	275	30	321	375	532	269	203	14	64	40	1
Girl Scout Cookies	567	818	515	141	232	306	186	89	6	32	14	1
Gelato	1,845	7,792	5,666	346	636	670	322	156	102	84	39	0
Wedding Cake	3,085	1,132	841	603	624	1,190	561	420	136	95	129	0
Blueberry	411	6,005	3,083	304	738	299	77	44	15	21	26	1
Trainwreck	248	772	440	249	121	219	131	33	38	48	14	0
Strawberry Cough	231	223	14	176	375	271	203	11	17	55	3	0
Total	17,793	24,871	14,814	4,931	7,118	9,620				1,082	691	26

The most frequent excluded near-match names per strain are listed in the aggregates JSON (exclusions.near_match_top_names).

9. Concentration of the evidence

Producer shares are computed only over tests that record a producer (MD and MI record none).

Strain	Tests	Labs	Max share one lab	Max share one producer (tests with producer)	Largest state share
Blue Dream	2,118	49	0.167	0.316 (1250)	0.29 (WA)
Sour Diesel	1,071	45	0.253	0.166 (385)	0.56 (MD)
Green Crack	1,236	37	0.235	0.256 (951)	0.61 (WA)
Northern Lights	406	24	0.433	0.214 (56)	0.76 (MD)
Purple Urkle	35	10	0.265	0.519 (27)	0.66 (WA)
Blue Cheese	288	12	0.91	0.556 (9)	0.93 (MD)
Cinderella 99	598	15	0.263	0.684 (57)	0.87 (MD)
Hindu Kush	17	7	0.353	0.375 (16)	0.82 (WA)
Chemdawg	668	28	0.308	0.331 (450)	0.50 (NV)
Grape Ape	202	23	0.441	0.451 (162)	0.47 (NV)
G13 Haze	14	4	0.571	0.714 (14)	0.86 (WA)
G13	97	10	0.299	0.726 (73)	0.55 (NV)
Lemon Kush	29	7	0.5	0.412 (17)	0.41 (MI)
Agent Orange	140	12	0.443	0.444 (18)	0.84 (MD)
Lamb’s Bread	117	6	0.761	0.846 (26)	0.78 (MD)
Lime OG	29	9	0.552	0.833 (6)	0.79 (MI)
Critical Jack	8	2	0.5	0.25 (4)	0.50 (MA)
Pineapple Express	461	33	0.233	0.185 (265)	0.42 (MI)
OG Kush	379	32	0.231	0.282 (291)	0.32 (WA)
Jack Herer	910	26	0.293	0.209 (508)	0.44 (NV)
Durban Poison	528	35	0.237	0.153 (314)	0.42 (NV)
Granddaddy Purple	512	30	0.336	0.34 (247)	0.43 (MD)
Super Lemon Haze	1,543	35	0.189	0.288 (1176)	0.44 (WA)
Girl Scout Cookies	567	36	0.165	0.43 (251)	0.41 (MD)
Gelato	1,845	47	0.162	0.091 (692)	0.37 (MI)
Wedding Cake	3,085	49	0.181	0.188 (1802)	0.43 (NV)
Blueberry	411	35	0.399	0.144 (194)	0.43 (MD)
Trainwreck	248	24	0.348	0.38 (205)	0.37 (NV)
Strawberry Cough	231	17	0.498	0.572 (194)	0.52 (WA)

10. Caveats

Lab shopping / THC inflation. Producers can choose labs, and labs differ systematically; reported THC is known to be inflated at some labs. Several strains rest heavily on one lab (table above) — e.g. MD Blue Cheese comes almost entirely from MD lab 2501, whose median total THC across all matched strains (23.2 %, n≈1,200) is above the other Maryland labs (20.3–21.5 %); the same lab reported implausibly high Δ9-THC for some 2018 flower (tests over 40 % were dropped; others remain).
Sampling bias by state and period. The data are whatever each state released or Cannlytics could parse: MD, NV, MI and WA dominate; CA is mainly 2023–2024 COAs from a few labs; MA is the MCR Labs COA subset (mostly 2014–2019); HI has no THC values for tests dated 2020-12 to 2023-08. Year-to-year changes mix real trends with changing state mix.
Names are not genetics. Matching is by label text; the same name from different growers can be different genetics, and pheno-hunt series (e.g. CA “Strawberry Cough #110–#139”, “Trainwreck #297–#322” from one producer) are included as the named strain.
Small samples. Critical Jack (n = 8), G13 Haze (14), Hindu Kush (17), Lemon Kush (29), Lime OG (29) and Purple Urkle (35) are too small for robust distribution statistics; terpene top-3 lists for strains with < 10 terpene tests are anecdotal.
Agent Orange (median 12.8 %) is 118 of 140 tests from Maryland, where all four labs report low values — this reflects that market’s product, not necessarily the cultivar everywhere.
Terpene coverage is limited to NV, CA, MA, FL, CT, CO, NY and UT (MD: total only; WA, MI, HI: none), so terpene figures describe a different subset of tests than the THC figures.
Parsing artefacts in the source were found and handled rather than silently used: WA CCRS flat potency columns (used the COA results instead), FL TerpLife total CBD, UT total THC = 0.0, CA duplicate COAs.

11. Per-state notes

CA — Name: product_name (strain_name populated in <2% of rows). Flower types: Flower; Flower, Inhalable; Flower, Medical Inhalable; Flower, Product Inhalable; Plant (Flower - Cured). Hemp flower excluded. Reported total THC used; a few labs (Excelbis, SQRD, Decano) report totals in mg/g, converted only when the % components confirm. Terpenes from COA results. Many COAs are contaminant-only (no potency) -> no THC value.
CO — Name: product_name. Flower type: “Flower, Inhaled Product” (pre-rolls, blunts and fresh-frozen samples in that type are excluded by name). Many SC Labs COAs are contaminant-only (no potency).
CT — Name: product_name. Flower types: Raw material (flower), Usable Marijuana, Flower variants. Several product registrations share one lab test -> collapsed by value de-duplication.
FL — Name: strain_name (product_name used only to check for contradictions). Flower types: Flower, Flower Inhalable. TerpLife Labs total_cbd column contradicts its own CBD/CBDA results -> CBD recomputed from components.
HI — Name: strain_name. Flower type: Flower Lot. No total THC column -> delta-9 THC + 0.877 x THCA. THC/THCA are blank in the source for tests dated 2020-12-31 to 2023-08-18 -> those tests have no THC value and are excluded. No lab field. No terpenes.
MA — Name: product_name, present only for the MCR Labs COA subset (state public-records rows carry no product name and cannot be matched). Total THC computed from delta-9 THC + 0.877 x THCA. Mostly 2014-2019 tests.
MD — Name: strain_name only. Flower type: Buds. No total THC column -> computed. Only total terpenes reported. Lab given as a lab ID code; no producer field.
MI — Name: product_name. Flower type: Flower. Total THC only (no CBD, no terpenes). Labs anonymised in the source (Lab 1, Lab 2, …).
NV — Name: product_name. Flower types: Marijuana Flowers/Buds, Small/Popcorn Buds, Marijuana Flowers/Buds-Packaged. Terpenes reported (no ocimene column). Some rows have no test date.
NY — Very small file (330 rows). Type from standard_product_type.
UT — Product type missing for most rows: rows with a UDAF tracking number starting with “F” are treated as flower (every explicitly typed flower row carries an F number). Flat total_thc is a 0.0 parsing artefact for these rows -> THC computed from the COA results.
WA — Two systems: Leaf Data Systems (2018-2019 rows; type from the inventory-name bracket, date = record created_at) and CCRS (2021-2023; type = inventory_type, date = test_date, potency from the results JSON because the flat columns disagree with it in ~50% of rows). The joined strain_name contradicts the product/inventory name in a large share of rows, so a WA test is used only when strain_name and the product/inventory name independently name the same strain. No terpenes.
OR — No product or strain name field in the file -> cannot be matched (0 tests).
RI — No product or strain name field in the file (only a Metrc tag) -> cannot be matched (0 tests).

12. Reproducibility

The pipeline is scripted end to end (streaming pre-filter, matching and classification, de-duplication and aggregation) and is rerun when Cannlytics publishes new data. Every figure on our strain guides is computed from these outputs at build time; none is typed in by hand.

How we compute strain lab statistics