The flower blind spot: the worst-labeled product is the least-checked one

Two facts, side by side.

First: the strongest independent evidence Colorado has about label accuracy says flower is the problem category. A 281-product study presented at the state's own Science & Policy Forum in August 2025 found concentrate labels essentially honest and flower labels wrong more than half the time.

Second: Colorado launched an off-the-shelf surveillance testing program in fall 2025 to check, among other things, exactly that. A full year later, flower has never been surveillance-tested. Not one sample.

The product with the worst documented label record is the one product the state's shelf-checking program has never checked. That is the blind spot, and this article is about how it happened, what it costs, and how to read the results when they finally arrive.

How the blind spot happened

Nobody chose it, which is part of what makes it instructive. The program launched on schedule and executed its edibles phase cleanly; the results landed in June. Flower and concentrates were supposed to follow in early 2026. Then reality arrived in layers:

• Methods first. Testing flower off the shelf for pesticides and solvents required method validation work that slipped in November.
• Then the building (its own article). The state reference lab operates out of a roughly 30-year-old facility whose air handlers began failing one by one. In January, one exploded. Metals testing went down for lack of exhaust ventilation.
• Then the repairs found more repairs (February, March), and by April flower and concentrate surveillance had moved officially from the first quarter of 2026 to fall 2026.

The causes deserve to be read sympathetically: method validation is real scientific work, and a building that predates the industry it regulates is a funding decision made long before anyone in the room had their job. Who actually pays for testing, and what a decent facility would cost against the state's cannabis tax receipts, is its own article. But sympathy for the causes doesn't shrink the consequence: the category that most needed checking spent the program's entire first year unchecked.

What the blind spot costs

First, a scoping note that fairness requires: flower is not untested. It may be the most tested product in the state; every harvest batch passes a licensed lab before it can legally be sold. What has never happened is the second look: the same product, bought off a shelf the way a customer buys it, tested again. The blind spot is not testing. It is verification.

For a patient, flower's label problem is more forgiving than it sounds, because inhaled effects begin within a minute or two, so a patient doses from what the first puff actually did rather than from the printed number. Our patient guide on flower potency covers why the single THC percent was always a shaky way to describe a jar of plant material, and what to read instead.

The cost of the blind spot is different: it is a year of not knowing, on the public's behalf. The 281-product study was independent work with its own scope and methods. The state's surveillance program is the instrument that could confirm, refute, or size the problem with official standing, purchase-anonymous sampling, and consequences attached. For flower, that instrument has produced nothing yet. Every policy argument about flower labels, and there were many this year, ran on borrowed data.

How to read the fall results when they land

Flower surveillance is now slated for fall 2026. Before the numbers arrive is the honest time to say how we will read them, so here is the frame we will use, in public, in advance. Everything in this section is labeled analysis: a reading frame, not a prediction.

If flower retests come back at label, the simplest story is that original tests and the state's method agree, and the independent study's window (its samples, its labs, its years) needs its own explanation.

If flower retests come back consistently below label, shelf chemistry is a poor explanation, and that matters, because "it degraded on the shelf" will be the tempting excuse. Two desk-reviewed storage studies say the potency number holds at shelf timescales: a 2024 controlled storage study (MacLaughlin and MacDonald) followed five chemovars for 74 days and found the aroma compounds leaving while total cannabinoids showed no significant loss, with THC actually ticking up as its acid form converts. Long-horizon decay is real but runs on years, not shelf-weeks (a 2025 stored-resin study (Fettoukh et al.) tracked it; resin is not flower, so we take the direction and not the magnitudes). And in dry Colorado there is a thumb on the other side of the scale: flower loses moisture on the shelf, and the state reference lab's posted cannabinoid method contains no moisture-correction step, so a drier sample reads a higher percent, as received. A below-label flower retest runs against shelf physics, not with it.

Which leaves two live explanations for a below-label gap, and they matter differently:

1. The original number described the best buds, not the batch. Flower's chemistry varies within a plant and a batch (documented biology), and whoever selects the test sample chooses where on that spread the number comes from. Note what this explanation does not require: no dishonest lab, no falsified chemistry. An honest lab honestly measuring a selected sample fully accounts for a below-label retest.
2. The methods disagree. Different labs, different procedures, different reporting bases; the proficiency-versus-retest gap the state itself cannot yet explain.

Separating those two takes one specific experiment: send off-the-shelf flower to the same licensed labs that did the original testing. If the labs that produced the high original numbers read the shelf samples low, the original numbers were the problem, because a method cannot disagree with itself; what changed was the sample. If those labs read shelf samples at label while the state lab reads them low, the gap lives between methods. And we will say plainly which way we lean: the selection explanation is our leading hypothesis, because it requires nothing beyond documented incentives and documented biology. But a leading hypothesis is not knowledge. Without the same-lab shelf test, nobody will actually know, and whether the fall program includes that design is one of the things we are asking on September 4.

What we are watching

• The fall start actually starting. The slip history earns the scrutiny: Q1 became fall once already.
• Whether flower results, when they come, are published in writing or presented into the video-only void.
• The September 4 forum agenda, and our paired-sample question.

We will read the results against this frame, in public, whatever they show.

Meeting characterizations are from our machine transcripts of the Division's published recordings, which are the source of record. The two storage studies are desk-reviewed in full; our citation policy and worked reviews are described here.