Industries / Food, farming and forestry / Agriculture and farming / Cannabis and hemp
Tracking from seed to sale
Keeping a record of every licensed plant, batch and package from planting to sale, so a regulator can check that nothing leaves the legal market and nothing enters it from outside.
6
Milestones achieved, from 2011 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2011
A tag on every plant
Franwell, a Florida supply-chain technology company, built Colorado's tracking software for medical marijuana in 2011. The system became Metrc: every plant and package in a licensed facility carries a plastic radio-frequency ID (RFID) tag, a chip that a reader can scan without line of sight, with a unique 24-digit number, and each move or change of the product is recorded against it. Colorado's Marijuana Enforcement Division now requires every licensed business to use Metrc as its "primary inventory tracking system of record," and Metrc's site listed 29 US states, territories and the District of Columbia as clients in November 2026 (in January 2024 MJBizMagazine counted 23 state contracts for Metrc and nine for its nearest rival, BioTrack).
Credit
Franwell and Metrc; Colorado's Marijuana Enforcement Division
2017
A national buyer registry
Uruguay regulates the whole chain under its Law 19.172, and adults must register to buy. The register of pharmacy buyers opened on 2 May 2017 and pharmacies began selling 5-gram packs on 19 July 2017. By 31 December 2025, 113,548 people were registered: 83,567 to buy in pharmacies, 10,392 home growers and 19,589 members of 557 clubs. Pharmacies had recorded 3.9 million sales to registered buyers, by the count of Uruguay's Drug Observatory (Observatorio Uruguayo de Drogas) from the regulator's registers.
Credit
Uruguay's Institute for the Regulation and Control of Cannabis (IRCCA); Observatorio Uruguayo de Drogas
2018
A vendor switch fails
Washington tracked cannabis with BioTrackTHC from 2013 to 2017, then moved to MJ Freeway's Leaf Data Systems (Franwell had first been named the top bidder, in May 2017). In the week Leaf went live in February 2018, an intruder exploited a flaw and likely downloaded a copy of the tracking database, the state's Liquor and Cannabis Board told licensees. Washington's State Auditor found in 2024 that Leaf "ultimately failed due to project management and vendor issues"; in 2021 the board replaced it with a stopgap, the Cannabis Central Reporting System (CCRS), which takes weekly uploads, gives enforcement officers no real-time data, and lacks a single product ID that would speed recalls. The auditor reported that a full replacement is not expected until 2031.
LessonA tracking system fails or succeeds with the project behind it, and Washington has run on a stopgap since 2021.
newcannabisventures.comsao.wa.govlcb.wa.govcontent.govdelivery.comnewcannabisventures.com
Credit
Washington State Liquor and Cannabis Board; MJ Freeway (later Akerna); Washington State Auditor's Office
2019
National data, published monthly
Every Canadian licensed producer reports to Health Canada under the Cannabis Tracking System Order, which Health Canada says came into force on 17 October 2019. Health Canada publishes the totals as open data: sales, packaged and unpackaged inventory, licensed area and destruction, by product type and month. Its latest tables show 45.9 million packaged units in inventory in June 2026, and 11.0% of the unpackaged dried cannabis produced in the second quarter of 2026 destroyed.
Credit
Health Canada
2024
Clubs keep their own books
Germany's Consumer Cannabis Act (Konsumcannabisgesetz) of 27 March 2024 lets licensed non-profit growing clubs supply their members, and it uses no central tag system. Instead, section 26 makes each club record, as it goes, where its seeds and cuttings came from, how much it grew and destroyed, and each handout: the member's name and birth year, the grams, the average THC content (THC is the compound that causes a high) and the date. Clubs keep the records for five years, hand them over electronically on request, and send anonymized yearly totals to the authorities by 31 January for the law's evaluation.
Credit
Germany's federal legislature
2026
Data is not oversight
In HNHPC Inc. v. Department of Cannabis Control, Judge Lee Gabriel of the Orange County Superior Court found that California's CCTT–Metrc system "does not currently flag irregular transactions for investigation," as section 26067 of the state's Business and Professions Code requires; analysts reviewed the data by hand with no set definition of an irregular transaction. The final judgment, entered on 4 August 2026, gave the Department of Cannabis Control six months to set objective criteria for an irregularity.
LessonCollecting tracking data catches nothing until someone defines what to look for.
Credit
HNHPC Inc., parent of the retailer Catalyst, which brought the suit
Watch for
California's irregularity criteria
The August 2026 judgment requires California's Department of Cannabis Control to set objective criteria for flagging irregular transactions in its CCTT–Metrc data within six months, without replacing Metrc.
Achieved whenthe department publishes the criteria and reports how many transactions they flagged and how many were investigated.
Claimed by
Orange County Superior Court judgment in HNHPC Inc. v. Department of Cannabis Control, as summarized by the plaintiff's lawyer, August 2026
A QR code per unit
Metrc says its Retail ID gives each finished product unit a QR code tied to its tracking record and lab results, saving about 60 seconds of receiving labor per unit and $25,000 or more a year per retailer, and allowing faster recalls.
Achieved whena regulator or an independent study measures receiving time and recall speed with and without unit-level codes.
Claimed by
Metrc, on its Retail ID page as read in November 2026
Open
01
An open diversion rulebook
A public set of objective rules that flag possibly irregular activity in seed-to-sale records (diversion means legal product leaking to the illegal market). Examples: a harvest heavier than its plant count makes plausible, packages sold before a lab result was recorded, repeated weight "adjustments," or sudden transfers between licensees that never traded before. Write each rule as code with a plain-language reason, and run the set over a public traceability dataset.
The briefCredit · unclaimed02
Score flags against enforcement
A labeled test set of licensees whose products were recalled or who were penalized for diversion or inventory violations, taken from public notices and matched to their license numbers, so that any flagging method (a state's, a vendor's or the rulebook above) can be scored on whether it would have raised an alarm before the regulator acted.
The briefCredit · unclaimed
Open 01 · the brief
An open diversion rulebook
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of objective rules that flag possibly irregular activity in seed-to-sale records (diversion means legal product leaking to the illegal market). Examples: a harvest heavier than its plant count makes plausible, packages sold before a lab result was recorded, repeated weight "adjustments," or sudden transfers between licensees that never traded before. Write each rule as code with a plain-language reason, and run the set over a public traceability dataset.
Why it is still open
California's court found that its system "does not automatically identify potentially irregular activity using objective criteria," and Washington's State Auditor found in 2024 that the regulator's automated alerts "do not fully address" its 2018 call for data-driven enforcement. Searches of the web, GitHub and the OpenAlex index of research papers found no published, open rule set.
Built with what exists
Washington's licensee-level traceability files released under the state's public-records law (Cannlytics' open Cannabis Data Science lessons show how to load the Leaf-era files); Cannlytics' open-source Python package, which includes a Metrc API client; Python with pandas or DuckDB.
First step
Load one month of Washington's released traceability data with Cannlytics' example code, and write the first three rules from California's statute and the Washington audit.
Done looks like
At least ten rules, each with a cited reason, run over one full year of one state's data, with the share of licensees each rule flags published, and the same rules rerun on a second year to show the flag rates are stable.
Open 02 · the brief
Score flags against enforcement
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A labeled test set of licensees whose products were recalled or who were penalized for diversion or inventory violations, taken from public notices and matched to their license numbers, so that any flagging method (a state's, a vendor's or the rulebook above) can be scored on whether it would have raised an alarm before the regulator acted.
Why it is still open
Regulators review tracking data by hand or with partial alerts (California court, Washington auditor), and no public set of enforcement cases linked to tracking records was found. The largest open record of US cannabis recalls, an independent analyst's dataset updated monthly on GitHub, lists recalls by state but names "parse out batches" and a full database back end among its unfinished tasks.
Built with what exists
the US Cannabis Recalls by State dataset; Cannlytics' cannabis_licenses dataset on Hugging Face for license numbers; state enforcement notices and violation records; an AI model to pull license numbers and violation types out of PDF notices, every row checked by hand (the Note "Seven more flagged" covers an AI model reviewing drug-making records for problems).
First step
Download the recall dataset, take one state's recalls, and match each to a license number in the cannabis_licenses dataset.
Done looks like
At least 100 recall or enforcement cases linked to license numbers, released with code that reports, for any rule set, the share of cases flagged at least 90 days before the action and the share of all licensees flagged (the false-alarm load).