Flower potency: why the THC number is shaky, and what to read instead

You are at the counter looking at two jars. One says 24% THC, the other says 19%, and the 24% costs more. The single number feels like the measure of what you are buying.

This guide is about why, for flower specifically, that number is the least trustworthy thing printed on the label, and what a medical patient should read instead. Not because anyone necessarily lied (though the incentive exists and we will name it), but because of something more basic: a batch of flower does not have one THC number. It has a range. Printing a single percent on a jar of plant material is a polite fiction that biology never agreed to.

A plant is not a solution

A concentrate, prepared correctly, can honestly carry one potency number. Homogenization is a real manufacturing step: the batch is stirred until every gram is the same as every other gram, and only that step makes the number true. Do it well and testing any corner of the batch tests the batch. Skip it or rush it and a concentrate can lie the same way an edible can. That is why, in the 281-product independent study presented at the state's own forum, concentrate labels tested honest while flower labels were wrong more than half the time. Concentrates are the product where one number can be true.

Flower is the opposite case, and this is documented plant biology, not opinion. Cannabinoids are not evenly spread through a plant:

• Top to bottom. Flowers at the top of the plant run measurably stronger than flowers at the bottom of the same plant. A 2018 study that sampled inflorescences (the botanical word for the flower clusters, which in cannabis means the buds) by their position along the stem (Namdar et al.) found cannabinoid and terpene content highest in the apical (topmost) flowers and declining down the stem, with light penetration the leading explanation. Later work has measured bottom flowers at a fraction of the top flowers' THC on the same plant. Real grows fight that gradient, with side lighting, reflective surfaces, trellising, and canopy management, and the mitigation is real but partial: the pruning and plant-architecture research below exists precisely because substantial variation survives good practice.
• Plant to plant. Commercial growers know this so well that there is a research literature on pruning and plant architecture aimed specifically at reducing the spread (Danziger and Bernstein), and on how packing plants closer together increases yield while making cannabinoid levels less uniform (Frontiers in Plant Science, 2022). Uniformity is something a grow has to fight for, and the fight is never fully won.

A harvest batch mixes tops and bottoms across many plants. The jar you are holding got some of each. So when the label says 20%, the honest reading is "somewhere in a band around whatever the batch average really is, and the batch average is itself somewhere in a band around 20, maybe."

Then the measurement wobbles too

The plant's natural spread sits underneath a second layer of wobble: the test itself. Colorado's own data made this vivid this year. When the state re-tested products that had already passed a licensed lab, the samples passed the state's check barely more than half the time. Same product class, different lab, different answer, roughly a coin flip. Lab-to-lab disagreement on cannabis potency is comparable in size to the legal tolerance a label gets (plus or minus 15%), which is how the same jar can be legal at one lab and a violation at another. The year-in-review of the state's testing debates unpacks that asymmetry in full, and every label-accuracy number Colorado has, each with its scope, is collected in one place.

And the coin flip is probably not a fair coin. Two forces push the producer's number high specifically: the test sample is chosen by the producer (the next section), and the regulator's own description of the market is that businesses test at the lab that passes their product most frequently. Sample selection and lab-shopping both flatter the first number. The state's number has no one to flatter: it buys the way you do, off the shelf and unannounced, so its sample is the product as a customer actually receives it. Which gives you a practical rule: when the label and an off-the-shelf check disagree, the off-the-shelf number is likely the better estimate of what is in your jar. Claim strength: analysis. The direction of the bias follows from the incentives and the regulator's own lab-shopping description; no published data splits the retest gap by direction yet.

And somebody chooses the sample

Here is where the plant's biology and human incentive meet. The test sample for a flower batch is a handful of buds. Someone picks them. If the picker knows the tops test stronger, and the tops do test stronger, then the difference between an honest number and an inflated one is nothing more than which buds go in the envelope. No chemistry is falsified; the plant did all the work.

This is analysis, not an accusation against anyone. But it is not hypothetical either: sample integrity was the loudest argument in Colorado cannabis policy this year, the regulator calls sample adulteration a common violation, and the enforcement record on what happens to the caught is its own story. The system asks a distribution to produce one number and then lets interested parties choose where on the distribution to sample. You should read flower labels with that in mind.

The moisture wrinkle

One more wobble, small but real. A THC percent is a ratio: milligrams of THC per gram of flower. The flower's weight includes its water, and water content changes on the shelf. Flower that dries out between harvest and testing, or between one test and another, reads higher in percent terms with no change at all in what reaches you.

We were unable to find a moisture-correction step anywhere in the state reference laboratory's posted cannabinoid method; results describe the sample as received, at whatever dryness it arrived. Claim strength: the published method document; an absence, not a finding; licensed labs' own methods may handle moisture differently. The practical point for a patient is just this: a percent measured on a different day is a slightly different percent, even when everything was done honestly.

What to read instead

So the headline number is a rough estimate wearing a lab coat. What actually helps?

The THC-to-CBD ratio. Most of the ways a potency measurement drifts (instrument calibration, extraction efficiency, moisture, which buds got sampled) push all the cannabinoid numbers together. The totals move; the balance between them mostly survives. A jar labeled 20% THC and 10% CBD may not deliver 20 and 10, but it very likely delivers something close to 2-to-1. If you chose a product for its balance, you probably got roughly that balance. Claim strength: analytical reasoning; "likely" is doing real work in that sentence.

And that ratio is just the pair that reliably makes the label. The minor cannabinoids (CBG, CBN, THCV, CBC and their acid forms) are part of the same medicine, the science on them is young but real (our entourage guide covers what is and isn't documented), and on most labels they are simply invisible. This site advocates for more complete cannabinoid labeling. The case doesn't depend on settling the pharmacology first: you cannot study what is not measured, and a patient cannot journal against numbers they are never shown. Until fuller panels are standard, a producer who prints one is telling you something real about their confidence, the same way a printed terpene total does.

The terpenes, and their freshness. The idea that cannabis compounds work as a group is called the entourage effect (sometimes the ensemble effect), and we wrote a whole guide on what the science does and does not show. The short version for the counter: THC potency is mostly genetics, but how a flower smells, tastes, and feels is mostly the grow and the handling. In one study of genetically identical clones, the indoor batch contained no myrcene at all while its outdoor twin had it; same DNA, different jar. Terpenes also fade with age in a way THC does not. So the packaged-on date and your own nose at the counter tell you things the potency number cannot. If a terpene total is printed, that is a producer telling you something real; loud, fresh aroma is the entourage announcing itself. Our aroma map is the vocabulary for it.

Your own feedback loop. Flower's saving grace is speed: inhaled THC reaches the bloodstream through the lungs, so the effects of a puff begin within a minute or two. That means you feel what one puff actually did before you decide on the next one. Whatever the label says, the first puff tells you what this jar really is, and every decision after that runs on evidence, not on the label. That fast loop is why flower's label problem, real as it is, costs patients less than the edibles version of the same problem. Take the label as an opening estimate; let the first session set the real number.

A journal beats a label. Dose, method, product, and what it did for the thing you dose for. Three weeks of notes will predict your next jar better than any number the industry prints. Our pain field notes show what that looks like in practice over five years.

The counter summary

Buying flower by THC percent is like buying a band's album because the loudest song peaks at 104 decibels. The number is technically a measurement of something. It just is not a measurement of what you came for. The ratio tells you the shape of the medicine, the terpenes and freshness tell you the quality of this particular jar, and your first session tells you the truth. The percent mostly tells you what the jar costs.

This guide is information, not medical advice, and nothing in it is a recommendation to use any product. If you dose against a clinician's plan, the plan wins.