The April 3 forum: how the state actually checks its labs, the demand for a true value, and the edibles results slip out

Colorado's Science & Policy Forum met April 3, 2026, with the state lab in the chair (the Division's usual co-chair was away). We transcribed the Division's published recording (about 2 hours) with a speech-recognition model and drafted this summary from that transcript; our editor has reviewed and approved this page, and the recording remains the source of record. This was the promised proficiency-testing deep-dive (the most technical session of the year) and it ended with the year's most contentious exchange about the surveillance program.

Notes from the editor

Two things in this meeting deserve your full attention. The first is an unfairness argument, and it's worth spelling out in plain terms.

Here's how the state checks its labs: it sends every licensed lab a portion of the same carefully prepared cannabis and compares their answers to each other. No lab is expected to hit an exact number: the state grades on a curve, flags only the wild outliers, and accepts as normal that honest labs measuring the same material get somewhat different results. That's good science, it was honestly presented at this meeting, and we credit it.

Now look at how a producer's batch is judged: one sample, one lab, one number. If that single measurement lands more than 15% away from what the label claims, the batch fails, with the product's value, and possibly an enforcement case, riding on that one result. So the manufacturer's argument, stripped to its core: the state's own lab-checking program proves that a single measurement is too noisy to hang a verdict on. And a single measurement is exactly what a producer's batch hangs on. The legal version has teeth, too: if the state ever takes a producer to court over one failed number, the defense gets to quote the state's own program about how much honest labs disagree. And this isn't an abstract unfairness. It's a bill. Product destroyed over a number that was really measurement noise is one of the hidden costs we tally in who actually pays for testing. And shrinking that noise, by giving the system a properly funded reference lab to anchor to, is exactly what the facility argument in that article is about.

Nobody from the state rebutted any of this. Not at this meeting, not at any meeting since. Unanswered is not the same as unanswerable. But a year is a long time to leave the strongest argument in the room standing.

Second: "who made that decision? Because it wasn't the science and policy forum." The surveillance reports had been consolidated into one end-of-program document, and nobody present could name the decision's author. Transparency isn't only about how fast results post. It's about decisions having authors. The chair moved on without an answer, and no later recording we transcribed supplies one. So this stays on our desk: we'll keep watching for the answer, and the question is part of our prep for the September 4 meeting. We are also considering a records request under the Colorado Open Records Act for this decision's paper trail, alongside the aggregate data the statute already requires to be public. We'd like your feedback before we file anything: write to editor@colorado-medical-cannabis.org, or use the September 4 question page.

Since this meeting

Where the threads this meeting opened stand as of August 2026, from the recordings that followed.

Resolved. The edibles results that slipped out under questioning here were presented in full in June. The full numbers (only 47% passing both potency and homogeneity) were considerably harder than April's "a few issues" characterization. The analyte-form revision commissioned here became the live submission form in July. The fall-2026 surveillance timeline set here still held as of the August meeting.

Still in progress. The single-reference-value scoring the statistician disclosed as under internal discussion (the nearest thing to the "North Star" demand the program has offered) has not been committed to in any later recording we transcribed. Method performance requirements reached the forum in August.

Put off or neglected. Blind proficiency testing: wanted by everyone, designed by no one, unchanged since. The ±15%-versus-PT-bands reconciliation drew no rebuttal and no follow-up. And "who made that decision" was never answered.

What was brought up

How Colorado's potency proficiency test actually works [0:05:02]. Three state-lab staff walked through the whole machine. The history: when licensing began in 2014, no relevant proficiency tests existed, only chemical standards that skipped sample prep and matrix effects entirely, so a 2015 expert workgroup built Colorado its own program, beta-tested three rounds, and had an outside panel validate the design. The mechanics: real already-tested cannabis is dried without decarboxylating, ground and homogenized, split into containers with stratified random selection across the batch, then verified homogeneous via ten samples, two preparations each, two injections each (forty data points through trend checks and a nonparametric significance test) before anything ships to labs [0:15:05–0:25:11]. The scoring: because Colorado has few labs, ordinary statistics would let one gross error poison the comparison, so the program uses robust statistics: outlier-resistant means and standard deviations feeding a modified z-score, with anything past three standard deviations flagged [0:30:15]. Contaminant proficiency testing is different: third-party ISO-accredited vendors using hemp as a stand-in matrix, because federal scheduling bars shipping marijuana matrix to providers [0:35:15]. And a structural detail worth remembering: the state lab's own federal registration means the MED must physically transport PT samples. The lab can't legally receive them from anyone but law enforcement [0:15:05].

The true-value demand [0:40:18]. An infused-products manufacturer accepted the program's statistics and then attacked the asymmetry under it: the proficiency system judges labs generously against each other's average, while manufacturers live or die on a single-point result against a hard ±15% rule: "the fact that a lab is on average not variant doesn't matter to me... we are being held by rule to a single point" [0:45:22]. His demand, familiar from earlier meetings but sharpest here: the state lab should function as the true value ("a North Star") that every lab is measured against, because "unless there's an anchor... you couldn't explain to a consumer that 10 milligrams might be 12 milligrams next year because the standard drifted" [0:50:33]. The state's answers were candid: a true value is philosophically unreachable in analytical chemistry; method validation, quality controls, and reference materials are how labs approach it; variability is inherent: "you can test a sample a hundred times and only get the same number a couple of times" [0:55:36]. But the statistician disclosed something concrete: the lab is already internally discussing scoring labs against a single reference value derived from its homogeneity data, instead of against each other's mean [0:45:22].

Blind testing, batted around [1:00:44]. The known flaw of all proficiency testing (everyone performs well when they know they're being watched) drew three workarounds from the room: a rotating all-labs retest of shelf product (degradation should move every lab's number the same direction; a lab matching the original label while everyone else reads lower is the tell) [1:00:44]; blind submissions routed through cooperating licensees' inventory identities [1:15:59]; and paired compliance submissions with a retained split going to competitor labs [1:21:01]. The agencies' response: blind testing would be ideal, and the legal and resource constraints (a fake licensee, the federal receipt bar, who pays) have so far beaten every design they've considered [1:05:55].

The chromium pilot came back [1:26:06]. CDPHE's health-monitoring team ran the analyte worksheet on chromium and returned a verdict: right tool, wrong evidence scale. The worksheet's quality hierarchy is built for clinical medicine. But nobody will ever run a randomized trial of an inhaled toxin, so the scale must credit toxicology reports, occupational data, and toxicogenomic studies instead [1:31:11]. Their proposed process: submitter completes the form, agencies analyze, hard cases go to contracted toxicologists, and conclusions return to the forum for a rulemaking recommendation, with honest open questions about analytes with no evidence (route it to the Institute of Cannabis Research as a study recommendation) and about outcomes short of mandatory testing: surveillance-only testing, or ingredient prohibitions on the vitamin-E-acetate model [1:36:13, 1:41:23].

What was nailed down

• The surveillance program's new timeline: resumption near the end of Q3 2026 (the fall), after the year of HVAC and validation delays [1:46:25].
• The first edibles results, disclosed under questioning: no contaminants detected in the 45-sample edibles phase; "a few issues with either homogeneity or potency labeling accuracy," with the Division "working with those companies" [1:51:33].
• One consolidated report: the Division stated the decision was made to publish the surveillance results as a single report after all phases complete, rather than piecemeal [1:51:33].
• The worksheet revision path: CDPHE and MED to revise the analyte form (shorter, toxicology-appropriate evidence tiers, route-specific, submitter doesn't self-score) for the next meeting [1:46:25].

What was left open

• Who decided one report? The manufacturer's objection stands verbatim in the record: "who made that decision? Because it wasn't the science and policy forum", and the companion argument that a report published a year after sampling describes a market that no longer exists [1:51:33, 1:56:33]. The chair moved the meeting on without an answer.
• Whether enforcement contradicted the program's learning-first framing. The Division's position: samples were never blinded, and acting on failures was always the agreement; the report is not delayed by investigations [1:51:33]. The disagreement about what was agreed is itself now part of the record.
• The true-value question: internally under discussion at the lab, not committed [0:45:22].
• Whether PT bands and the ±15% rule can be reconciled: the manufacturer's legal argument (a prosecution for an out-of-band result is vulnerable if the labs themselves scatter) drew no rebuttal on the record [0:50:33].
• Blind proficiency testing: wanted by everyone, designed by no one [1:05:55].

This summary was drafted by our AI desk from a machine transcript and has been reviewed and approved by our editor (2026-08-13). The Division's published recording is the source of record.