The July 10 forum: fixing first-party sampling instead, and the coin-flip result

Colorado's Science & Policy Forum met July 10, 2026, pushed a week for the holiday, lightly attended, and the last meeting before the August 7 session this site covered live. We transcribed the Division's published recording (about 1 hour 41 minutes) 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. Two things happened: the sampling conversation officially pivoted from who collects to how collection is done and enforced, and the Division's deeper cut of the surveillance data produced the year's most uncomfortable numbers.

Notes from the editor

Here's the number this meeting put on the record. The state cross-checked its shelf samples against their original paperwork: of the 27 products that had been tested and passed at licensed labs, only 51.9% passed the state's own retest. Products that had already passed a test, passing the same kind of test again, at basically a coin flip.

Now, don't let that number do your thinking. The finding isn't that half of products fail. It's that passing a licensed lab's test carried almost no information about passing the state's retest. A coin flip means the two results are barely related. And the Division's honest caveats sharpen rather than soften that: if the state and the labs are measuring with different definitions (per serving versus per unit, three different homogeneity calculations), then the number on the label was never well-defined in the first place. The metric mismatch isn't an excuse for the numbers. It is the finding. Rulemaking that nails down definitions is exactly the right response, and it's the one previewed here.

And give the Division its due: it ran this comparison itself, presented it in public knowing how it would read, and had its own staff supply the caveats honestly rather than bury the comparison. An agency that publishes its most uncomfortable number is one whose next numbers you can take seriously.

And here's what a "failure" means for the bag of gummies in your hand. These are label-math failures, not safety failures: every safety test passed. A potency fail means the pieces don't match their number: a "10 milligram" gummy that's really 11 or 12. Most misses were small (fractions of a milligram), a couple were large, and roughly four out of five products ran at or above their label, because manufacturers aim high on purpose (nobody returns a strong gummy). A homogeneity fail is the sneakier one: the bag's average can be right while the doses are uneven: one piece quietly carrying double what its neighbor holds. "One gummy, one dose" stops being true, and that matters most for exactly the people counting milligrams: patients, beginners, anyone splitting doses. And what happened to the packages that failed? Nothing you could see. No recall, no notice, no names: the state "worked with those companies" privately. So the practical takeaway costs you nothing: with any new package, treat the first piece as a test dose, and don't stake anything that matters on the printed number being exact.

The effect-size analysis everyone in the room asked for (does passing at a licensed lab predict anything at all, once you control for the mismatch and the state lab's slightly-higher average) still doesn't exist. Until it does, the honest position for a consumer is the one this site keeps arriving at: the safety tests are holding, and the label number is not yet an instrument you can lean on.

Since this meeting

One meeting has followed this one: August 7, 2026, which we covered in depth.

Resolved. The August meeting happened as promised, and it finally took up the one big 2026 topic the year had kept skipping: whether every licensed lab should have to meet the same quality bar: shared standards for how sensitive and accurate their testing has to be, so two labs measuring the same jar should get close to the same answer (our August 7 article covers it). And the new form that lets anyone formally propose "the state should test for X" is live on the Division's site.

Still in progress. The rule changes this meeting predicted are now being drafted: making each gummy in a package carry the same dose, requiring a test for any cannabinoid a label mentions, and making every lab report results in the same units so numbers can actually be compared. The forum meets next on September 4, 2026; we'll cover the recording when the Division publishes it.

Put off or neglected. Three things were left hanging:

• Which labs? The state tried to work out whether the failed retests traced back to particular labs, and stopped, saying it didn't have enough data ("we really just didn't have a sufficiently large data set"). Nobody has picked it back up.
• The safeguards the labs asked for. Having lost the fight over who collects samples, the lab side proposed backup protections: video-record every sampling event; let labs see how big a batch actually is (today a lab receiving a sample bag has no way to tell whether it truly represents five pounds or fifty); and a standard state form documenting every collection. All three are still just proposals.
• The flavoring trap. Some products get their flavor from terpene blends bought from outside suppliers who keep their recipes secret. If something harmful turns out to be in the blend, the cannabis company eats the recall, for an ingredient it had no way to inspect. The forum agreed this is a real problem, and put it off for another day.

What was brought up

The pivot [0:10:18]. With no consensus for third-party collection, the co-chairs reframed the goal: strengthen the existing self-collection system, meaning the current training rules (Rule 4-225D), the two-collector requirement, and the sample-plan mandate, plus CDPHE's sampling-procedure documents, which the group walked through in detail [0:30:34–0:40:42]. The concrete gaps named: no dedicated sampling procedure for pre-rolls and infused pre-rolls (the existing document is flower-focused and its training video "very slow"), none for vaporizer products, and a bare-bones sample-plan template that operators interpret with "varying degrees of competency" [0:20:23, 0:30:34].

The lab side's fallback demands [0:20:23]. The lab representative who had championed third-party collection offered the alternative regime: high-resolution video surveillance of every sampling event, with batch tags and weights displayed to camera before and after, routine footage requests by the Division as the enforcement backbone; and parent-batch visibility in the inventory system, because today a lab receiving a sample "can't verify that the amount received is the actual minimum required for the total batch size." A department-issued collection form submitted with every test batch rounded out the ask [0:45:46].

The coin-flip result [0:50:46]. The Division's scientists cross- referenced the 45 surveillance samples against their original testing history. Eighteen of 45 had never been potency-tested at all (riding reduced-testing allowances). Of the 27 that had been tested and passed at licensed labs, 51.9% passed the state's surveillance retest on all metrics; 48.1% failed at least one [0:55:47]. A manufacturer said what the room was thinking: "a 50-50 chance... means no relationship. And if there's no relationship between originally passing testing and passing surveillance testing... that's a place we need to look and spend a lot of time" [1:00:51]. The homogeneity cut was stranger still: samples originally submitted for homogeneity failed the state's assessment more often than samples that never were [0:55:47].

The mitigations, honestly stated [1:00:51]. The Division's own staff supplied the caveats before critics could: the state lab evaluated metrics (milligrams per serving and per unit, three separate homogeneity calculations) that the rules don't require licensed labs to use, so part of the failure rate is a measurement-definition mismatch, not necessarily bad testing; the dataset is small; and the right reading is the ratio between groups, not raw counts. The high-bias pattern got a market explanation from a gummy manufacturer: nobody complains about a 105-milligram "100-milligram" package, so manufacturers target high. And indeed roughly 80% of results, at labs and state alike, came in at or above label [1:15:56, 1:26:07].

The rulemaking preview [1:26:07]. The Division sketched what the surveillance findings feed into August's rulemaking season: standardized reporting units (milligrams per serving and per unit), a defined basis for homogeneity's relative standard deviation, clarity that homogeneity batches must also pass potency and that a fail on either reports as both, and, with visible stakeholder support, the principle that any cannabinoid on the label is subject to testing requirements, paired with a proposed mandate that cannabinoids above some threshold must be labeled [1:31:07, 1:36:14].

What was nailed down

• The analyte-proposal form went live on the Division's public-engagement page, the second of the year's two umbrella topics formally closed [0:10:18].
• The pivot itself: sampling work continues as rule refinements and CDPHE procedure expansions (pre-rolls, vapes), with proposed redlines to come back to the group [0:50:46].
• August's agenda: method performance requirements (the third interlab-consistency subtopic, untouched all year) plus continued surveillance discussion [1:36:14]. (That August 7 meeting is covered in depth in our testing story.)
• The full phase-one report was published and linked in the forum's public folder following the June meeting [0:50:46].

What was left open

• What the coin-flip result means. Measurement mismatch, lab variance, sampling behavior, product drift on the shelf. The Division attempted a per-lab comparison and stopped: "we really just didn't have a sufficiently large data set" [1:26:07]. A lab director asked how many licensees the 45 samples even represented; the answer wasn't at hand [1:36:14].
• The effect-size question: testing at a licensed lab should massively predict passing a retest; whether the observed weak relationship survives controls (the state lab's slightly-higher average, the metric mismatch) is the analysis everyone asked for and no one yet has [1:15:56, 1:21:05].
• Video-surveillance sampling rules, parent-batch visibility, and the mandatory collection form: proposed, not adopted [0:20:23, 0:45:46].
• Botanically derived terpene documentation: raised as a good-faith compliance trap (formulation suppliers won't disclose ingredients; licensees carry the recall risk), acknowledged as out of scope for sampling but a future topic [0:40:42].

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.