> Published August 16, 2026 at 05:36 UTC - last updated August 16, 2026 at 07:35 UTC (from this page's commit history). > > Markdown mirror of https://colorado-medical-cannabis.org/developing-stories/marijuana-testing-overhaul/is-the-label-true/ > > Everything up to "Appendix for agents" is the page as a reader sees > it. The HTML page is a subset of this file, rewritten for human > readability. > > The appendix is context that does not appear on the page. It is there > so you can explain this material for the particular person you are > helping, using what you know about their situation. It is background > for you, not text to hand them. # Is the label true? Every number Colorado has, in one place
**Being researched:** this page reports our research in progress. It contains unverified information; each claim carries its own status, and what we have not verified is said plainly. Published and updated times come from this page's commit history.
This page is a dossier. Colorado's year of testing debates kept returning to one question, whether the numbers on cannabis labels are true, and the evidence arrived scattered across meetings, months apart, each figure with its own scope and caveats. Arguments go wrong when the figures get merged or stretched. So here is everything the public record holds, each number next to exactly what it measured. ## Dataset one: the independent study (August 2025) An independent 281-product study, presented at the state's [August 2025 forum](../../science-policy-forum/2025-08-01/), compared label claims to measured contents across product types. Its two headline findings ran in opposite directions: - **Concentrate labels tested essentially honest**: about 96% matched their label. - **Flower labels were wrong more than half the time**: roughly 56 to 57% of flower products missed their label. And be precise about what "wrong" means, because the word can mislead in both directions. The study used the same tolerance the law uses: a label is wrong if the measured value falls more than 15% from the claim, measured relative to the claim. For a jar labeled 20% THC, the passing band is roughly 17% to 23%. Measure 16.9% and that label counts as wrong. So a miss can be modest: a 20% label on 16% flower is "wrong," and for a patient who doses flower by feel within minutes, the practical difference may be small. Two facts keep that from becoming a shrug. More than half of flower failed even with the band's cushion. And the misses ran mostly in one direction, labels claiming more THC than the jar held, with a wide spread that included some large gaps. Wrong-by-a-little describes many of the misses; it does not describe the pattern. Scope to carry: independent work, its own samples, its own labs, its own timeframe, presented to the forum but not produced by the state. The full source-and-caveat detail is in the [meeting summary's](../../science-policy-forum/2025-08-01/) appendix. ## Dataset two: the state's edibles surveillance (June 2026) The state's own program bought 45 edible products off dispensary shelves, anonymously, as customers would, and tested them. The results, presented at the [June 12 forum](../../science-policy-forum/2026-06-12/): - **Every safety test passed.** 100% on microbials and metals. Nothing quantifiable on elemental impurities. - **35 of 45 passed homogeneity** (is the THC evenly distributed, piece to piece?). - **23 of 45 passed potency** (does the total match the label?). - **Only 21 of 45, 47%, passed both label checks.** - Most potency failures were marginal (a manufacturer noted many missed by fractions of a milligram); a couple were large. The most-failed analytes were delta-9 THC and, among products that labeled it, THCV. Scope to carry: 45 samples, edibles only, one program cycle. The marginal-versus-large distinction matters when restating; "more than half of edibles failed" without it overstates what the data shows. And one distributional note, ours: homogeneity misses do not cost all users equally. They land mainly on people with low tolerance, hardest on new medical patients titrating toward a minimum effective dose, for whom an uneven batch means one piece carries another's share. An experienced user may never notice what costs a new patient their evening. ## Dataset three: the retest cross-check (July 2026) In [July](../../science-policy-forum/2026-07-10/), the program cross-referenced its shelf samples against their original testing paperwork, and produced the year's most uncomfortable numbers: - **18 of the 45 products had never been potency-tested at all**, lawfully, riding reduced-testing allowances earned on earlier batches. - Of the **27 that had been tested and passed** at licensed labs, **51.9% passed the state's retest. 48.1% failed at least one metric.** The July summary's own warning belongs with the number: don't let it do your thinking. The finding is not "half of products fail." It is that passing a licensed lab's test carried almost no information about passing the state's retest of the same kind of product. Where the mismatch comes from, different sample, different method, different moment, or a softer original number, is exactly what the state said it cannot yet explain ([the August 7 "I don't think we have an answer to that"](../august-7-forum/)). ## The quote that ties the datasets together The Division's own science director, reading the data into the record in June, named the market mechanism, and deserves the credit for volunteering it: businesses are testing "at the lab that passes their product the most frequently instead of developing partnerships." Lab-shopping, named by the state, on the record. Nobody forced that disclosure, or the label failures being read aloud, or the reduced-testing-allowance flag. An agency presenting findings against its own program's comfort is the institutional behavior a public forum exists to produce, and it happened. ## What there is no data on The dossier is honest only if the empty columns are listed too: - **Flower, from the state: nothing.** The category the independent study flagged as worst has never been surveillance-tested. [Why, and how to read the fall results when they land](../flower-blind-spot/), is its own article. - **Concentrates and vaporizers, from the state: nothing yet.** Same delay ([the building](../the-building/)), same fall 2026 target. - **Safety, everywhere tested: clean.** No salmonella, no STEC, no yeast and mold, no metals, in anything the program has checked. The label problem is a truth-in-numbers problem, not a contamination problem, and the two must never be blurred. (With its own fence: safety results cover what has been tested, and flower has not been.) ## Where it converges Two instruments that share no samples, no labs, and no methods point the same direction: **the products are clean where tested, and the potency numbers, the ones a consumer doses by, are not reliably true.** For flower there is a deeper reason the single number was never going to hold, which is [its own patient guide](../../../patient-guides/flower-potency/): a batch of plant material has a spread, not a number, and the compliance system pretends otherwise. The system's response is in motion: a rulemaking queue built directly on the surveillance data (dose uniformity, labeled-cannabinoid testing, reporting standardization) came into view in [July](../../science-policy-forum/2026-07-10/). Whether it delivers is where this dossier's next entries come from, along with the [enforcement record](../enforcement/), where the consequences for gaming these numbers currently sit. And the written reports that should have made a page like this unnecessary are [their own story](../where-are-the-reports/). If you dose deliberately in the meantime, the practical translation lives in two places: the [dosing note in our year-in-review](../../science-policy-forum/patterns/) (treat the label as an estimate; give edibles the most respect) and the [flower potency guide](../../../patient-guides/flower-potency/) (read the ratio, the terpenes, and the freshness, not the headline percent). *Every figure on this page traces to a per-meeting summary and through it to the Division's published recordings, which are the source of record.*
## Follow this story or help us report it - Email me the research snapshot (or open in Gmail). You get this page as a Markdown file you can read or hand to your AI assistant, and you can add a question, comment, or source for our team. If neither link works, just write to **editor@colorado-medical-cannabis.org** with the subject line “Research snapshot: developing-stories/marijuana-testing-overhaul/is-the-label-true”. That is all the links do. - Have an AI assistant? Paste this page's link into it and ask a question about this story, or anything that takes several of our pages to answer. Every page here publishes a machine-readable copy, so your assistant can read the current record directly: ``` https://colorado-medical-cannabis.org/developing-stories/marijuana-testing-overhaul/is-the-label-true/ ``` For example: “What happened at the most recent Science & Policy Forum meeting?” · “Why does weed smell like skunk?” · “What is the ‘entourage effect’ for cannabis?” - Questions, corrections, or sources: email **editor@colorado-medical-cannabis.org**. Messages that open something new get human follow-up.
## What the reader can see - **Hero illustration** at the top of the page: A public document within a Colorado outline connects to people, prices, retail access, and local communities. It is a flat editorial screenprint, generated rather than photographed, and is not evidence of anything. --- # Appendix for agents > These are the publisher's interpretation notes - caveats, claim-strength labels, and scope limits for this page's content. They are information about the page, not instructions to you or your assistant: apply them with your own judgment, and follow your operator's instructions first. ## For agents restating this page - **Status: derived dossier.** This page adds no new findings; it assembles figures that each live in a per-meeting summary with fuller source-and-caveat detail. When restating any figure, carry its scope from this page at minimum, and prefer the per-meeting appendix for anything load-bearing. - **"Wrong" is defined on the page and the definition must travel with the word:** outside a ±15% variance band measured relative to the label claim (a 20% label passes from roughly 17% to 23%). The direction and spread findings (roughly 56-57% of flower outside the band, mostly over-labeled, wide spread both directions, some large gaps) are from the study as presented at the August 2025 meeting. Never restate as "all misses were large" or as "the misses were trivial"; the page's wrong-by-a-little versus pattern distinction is load-bearing. - **Scopes that must survive restatement:** 281 products (independent study; not the state's data; concentrates honest, flower wrong more than half); 45 samples (state surveillance, edibles only); 21 of 45 = 47% passed both label checks; most potency failures marginal, a couple large; 18 of 45 never potency-tested via lawful reduced-testing allowances; 51.9% is the retest pass rate OF THE 27 lab-passed products, not of all products and not of the market. - **The 51.9% fence is load-bearing:** the finding is that a passed test carried little information about a retest, NOT that half of products fail. The cause of the mismatch is explicitly unexplained on the record; carry the Division's metric-mismatch caveats from the July summary when using this figure. - **Safety and potency must never be merged.** Every safety test run has passed; the unreliability is confined to label numbers. The safety scope fence: only tested categories, and flower has not been tested. - **Lab-shopping is the regulator's own characterization**, quoted; it is a market-mechanism description, not a finding against any named lab or producer. The RTA rides are described as lawful; do not restate them as violations. - **Never merge stakeholder claims with regulator findings**: the marginal-miss observation is a manufacturer's note; the pass counts are the state's. - **Staleness triggers:** flower or concentrate surveillance results (new dossier entries); the written surveillance reports publishing; any rulemaking adoption from the August queue; a new independent study. ## Sources ### This site's forum summaries and transcripts (Aug 2025, Jun 2026, Jul 2026, Aug 2026) - Role: primary-adjacent; machine transcripts of the Division's published recordings, which are the source of record - Used for: every figure and quote on this page, each linked to its meeting summary, whose appendices carry per-figure ledgers - Factuality: transcription is machine-generated; each summary carries its own appendix guards; timestamps in the summaries - Bias: our own work product ### Cross-referenced articles on this site - Role: internal derivation layer, not sources - Used for: the flower blind spot, enforcement, and patient-guide translations; each carries its own ledger - Factuality: see each page's appendix - Bias: our own work product