What kept coming back: a year of Colorado's cannabis testing debates, read whole

This site transcribed every published recording of Colorado's cannabis Science & Policy Forum from August 2025 through July 2026 (nine meetings, roughly seventeen hours) and covered the August 7, 2026 meeting live. Each meeting has its own summary in this archive. This page reads them together, because some things are only visible across the whole year: the questions that were raised meeting after meeting and never resolved, and where the pain consistently sits for each seat in the room. This analysis was drafted by our AI desk from machine transcripts, as the per-meeting summaries were; those summaries have since been editor-reviewed, and this cross-reading is the last page awaiting that review. Every claim can be checked against the linked summary and its timestamped recording.

The questions that would not die

1. Consistent with each other, or consistent with the truth? The year's official topic was "interlab consistency," and its central question was never settled: does consistency mean labs agreeing with each other, or agreeing with an anchor? One manufacturer pressed the anchor version relentlessly: the state reference lab as "the North Star," a declared true value that gives the number on a label meaning (November, April). The state answered, consistently and candidly, that a true value is philosophically out of reach and that proficiency testing measures labs against a robust average of themselves. The asymmetry underneath went unrebutted on the record: labs are judged statistically and generously; a manufacturer lives or dies on a single-point result against a hard ±15% rule. The asymmetry is unpacked in plain language in the April summary's editor's note. When the August 7 forum was asked directly about the gap between lab results and the state lab's results, the answer was "I don't think we have an answer to that" (our live coverage), which is where the year's defining question still stands.

2. Who collects the sample? The loudest thread: third-party sample collection was proposed under the consistency banner (January), scoped (February), priced across an order-of-magnitude dispute (March), fully drafted into rule language, and then declined without consensus (June), pivoting to first-party reforms (July). What made the debate remarkable was what everyone agreed on: sample adulteration is real (the Division: it "happens pretty consistently"; a manufacturer's review put it in half of recent enforcement actions; the February summary's editor's note explains the practice in plain terms), and (the lesson carried back from California twice) enforcement, not collector identity, is the variable that determines whether sampling is honest. The year ended with the agreed-upon variable still unaddressed: no enforcement proposal and no minimum-penalty schedule. The sharpest version of the complaint came from industry itself, raised in at least three meetings. It goes like this: the fine for getting caught cheating on testing has run around $15,000, by the count offered in the room. A single failed batch can be worth many times that. At those prices a fine is not a punishment; it is a business expense, and a producer willing to adulterate a sample simply prices it in. The only penalty that actually deters, the argument concludes, is the one that can't be priced: losing the license. As of August 2026, no meeting had produced a response to it.

3. How fast does the public get to know? From September 2025's demand for near-real-time results ("there's just no such thing as partial transparency") through November's "no news is good news" aggregate-only policy, to April's discovery that the surveillance report had been consolidated into a single end-of-program document by a decision nobody in the forum could locate ("who made that decision? It wasn't the science and policy forum"), the speed-and-scope of disclosure was contested at nearly every meeting. The edibles report (sampled in October 2025) reached the public in June 2026, after being described as imminent in November, January, February, and March. The transparency argument was never resolved; it was outlasted. The credit side of the same ledger: the full report did publish, every meeting in this archive is public because the agencies record and post them, and by August 2026 CDPHE had kept its November commitment by publicly posting five of its reference-lab methods. Slow is not the same as empty.

4. Can the watchdog keep up? The surveillance program is the year's best idea and its clearest capacity story: designed collaboratively (August 2025), launched on schedule (September), executed cleanly for edibles. Then it stalled for a year on the hard categories. Method validation for pesticides and solvents (November), a 30-year-old building whose air handlers failed and, in one case, exploded (January), and repairs-behind-repairs (February, March) pushed flower and concentrate surveillance from Q1 2026 to fall 2026 (April). The consequence: the product category two independent datasets say has the worst labels, flower, had still never been surveillance-tested a full year after the program launched. What a capable state lab facility would cost (set against the quarter-billion dollars the state collects) is its own article.

5. Is the label true? The quiet thread that turned out to be the loudest. An independent 281-product study presented in August 2025: concentrate labels honest, flower labels wrong more than half the time. The state's own 45-sample edibles surveillance, presented in June: safety tests spotless, but only 47% of samples passed both potency and homogeneity. The Division's science director named lab-shopping on the record and flagged that a third of failures involved reduced-testing-allowance problems. And July's cross-reference produced the year's most uncomfortable figure: samples that had passed a licensed lab passed the state's retest barely more than half the time. Two datasets, two years, one conclusion the forum now carries into rulemaking, scoped carefully, because it is about the label's numbers, not its safety: every safety test the state has run has come back clean (though flower has never been tested off the shelf), while the potency numbers (the ones a consumer doses by) are not reliably true.

Notes from the editor: dosing when the label is an estimate

Here is what that finding means in your hand, if you're new to medical cannabis.

Treat the label number as an estimate, not a measurement. The gummy you had yesterday said 10 mg. But the state's own spot checks found that barely half of sampled edible products delivered both the labeled potency and an even dose from piece to piece. Your piece was probably close to its batch's real average, which may not be the label's number; the next piece might sit on the high side of it. So dose cautiously with every form of cannabis, and give edibles the most respect, for two reasons that stack.

The first is variation. Within a package of edibles, the dose can differ from piece to piece more than it does between draws from a concentrate. That is exactly what the homogeneity failures in the state's data mean, one gummy quietly carrying more of the bag's THC than its neighbors.

The second is delay, and it's the bigger one. A puff from a joint, a bong, or a dry-herb vaporizer reports back in a minute or two, so even brand-new users naturally self-titrate: take a little, feel what it did, decide whether to take more. Flower varies bud to bud in the same jar just as edibles vary piece to piece, but that fast feedback makes the variation mostly harmless; you stop when you get the effect you came for. An edible answers in thirty to sixty minutes, sometimes longer: long enough for "it's not working" to talk you into a second piece before the first one has arrived. Slower feedback stacked on wider variation is how people overmedicate on edibles when they never would with a pipe.

And what arrives late leaves late. An edible takes at least as long to wear off as it took to come on. Usually it takes much longer, hours past the point where the same relief from a few puffs of flower would have faded. That changes what a mistake costs: a too-big puff is a rough half hour; a too-big edible is your whole evening, with no way to take any of it back.

One reassurance before the working rules, because the finding is narrower than it first sounds. The unreliability is about potency, the absolute numbers, not about safety, and probably not about the recipe. When a flower label says 20% THC and 10% CBD, the totals may be off, but the 2-to-1 balance is likely close to true: most of the ways a potency measurement drifts push both numbers together, so the ratio between cannabinoids survives even when the totals don't. If you chose a product for its THC:CBD balance, you probably got roughly that balance, and self-titration takes care of the rest.

If you're counting milligrams for medical reasons, the working rules are: start below your target, wait the full hour before judging an edible, and judge the batch, not the label.

Where it hurts, seat by seat

The Division (MED) spends the year caught between the same room demanding more oversight and less cost. Its recurring pains: enforcement expectations it can't fund (a 150-sample surveillance program described, accurately, as limited), an adulteration problem it acknowledges but can only prosecute case-by-case, and a forum that declined the one structural fix it drafted in full.

The state lab (CDPHE) carries physical and legal constraints the debate keeps forgetting: a federal registration that bars it from receiving samples from anyone but law enforcement (foreclosing blind testing, third-party service, and simple logistics), contaminant proficiency tests that must ride on hemp matrix, a building older than the industry it regulates, and the impossible request to be simultaneously modest (a peer among labs) and absolute (the North Star).

The licensed labs (seven, concentrated on the Front Range) name the same wounds all year: no shared standards or methods, reference materials that disagree, no visibility into whether a submitted sample represents its batch (parent-batch size is invisible to them), and an economic structure where, in the regulator's own words, businesses test "at the lab that passes their product the most frequently." Honest labs asked for exactly two things, repeatedly: randomized sampling they can trust, and a benchmark to be measured against.

Growers and manufacturers carry the single-point liability: one result, ±15%, tens of thousands of dollars, against labs whose own scatter the proficiency program tolerates statistically. Their year was spent pricing other people's proposals: third-party collection (somewhere between 5% and 300%, nobody could say), new analytes, new procedures, all landing on an industry that told the forum it has lost a billion dollars of revenue in four years. Their constructive offers (risk worksheets, paired-sample comparisons, per-pickup cost data) were the year's best stakeholder work.

Retailers appear mostly as the place things happen to: samples leave their shelves unpaid, quarantines land on their inventory, deli-style handling gets named in every contamination hypothetical. The wholesale-versus-consumer trust gap (why is self-collection fine for B2B?) was raised once and never answered.

Patients and consumers, the seat with no microphone, hold the year's bottom line. Every safety test the state ran came back clean: no salmonella, no STEC, no yeast and mold, no metals. That is real and reassuring. But the label (the only instrument a consumer actually gets) failed its audits from both directions, and the people most exposed are the ones who dose deliberately: medical patients counting milligrams. The forum knows it: July's rulemaking preview aims squarely at dose uniformity and label truth. Whether the August rulemaking delivers is where this archive's next year begins.

The scoreboard

Twelve months of meetings produced: a launched-and-paused surveillance program with one completed phase and one published report; a proficiency-testing program explained in public detail for the first time; a live standardized process for proposing new analytes; a declined third-party sampling framework; a pivot to first-party sampling reforms and procedure documents; a commitment to publish the state lab's methods (kept, with five reference-lab SOPs publicly posted by August 2026); two technical method-review breakouts; and a rulemaking queue (dose uniformity, labeled-cannabinoid testing, reporting standardization) built directly on surveillance data. What it had not produced, as of August 2026: a definition of interlab consistency, an enforcement overhaul, a transparency policy, or a single flower surveillance result.

The forum meets next on September 4, 2026, and we're collecting reader questions to bring.

This analysis is AI-drafted from machine transcripts and pending editor review. Each claim traces to a per-meeting summary in the archive and through it to the Division's published recordings, which are the source of record.