Acceptable batch deviation (%) in pharmaceutical production of medicine

I am trying to get my batch homogeneity down with 50 batches of chocolate bars. And it seems even if I test the same bar there is still a ~3% difference between left edge, middle and right edge. (middle seems to always be the higher value most likely the highest fat content as it cools down)

3% is amazing and I am more than happy with that. I would consider the HPLC to have a 3% variance from test to test. So I have started working on my batch homogeneity and I am at around 5-6% deviation. Again very happy with this.

I was wondering if there is a pharmaceutical standard that I could go off of so that way I can use it as an example when explaining to a wholesaler or curious consumer. With a quick google search I found some info using Pfizer as an example, saying theyre allowed up to a 20% variation in Xanax production. Which is huge because with the popularity of Xanax and now Pfizer this would be a great example to use as comparison… Hey Pfizer can have up to 20%, ours is 6%.

Any input is appreciated as usual


(Tagging a few people who I think may know or have input)
@MagisterChemist @Akoyeh @Thetetraguy


That’s a rather shocking allowable variance in xanax given its subjective effects.


I cant find the quote again so don’t quote me on it just yet!

edit: maybe this link covers it still reading

After subtracting the reference mean the tolerance limits are similarly adjusted, e.g., 85%-115% of label claim becomes ±15%


Here is the FDA’s guidance on process validation for APIs.

TL;DR: Every process is different and as long as you take all the appropriate steps (a lot of them) to identify and account for sources of process variability and document it all, you’re good as far as the FDA is concerned.

This is how pharmaceutical companies can have entire departments of engineers dedicated to validation and compliance.

I know that’s not super helpful in answering your specific question, but I figured I’d chime in with some context for how the big boys do it.


Still, given that’s it’s rarely prescribed off label, usually only for suppression of acute anxiety, the max daily dose is 4mg, and they can be 2mg pills, that’s wild.

So someone could potentially be 30% above daily recommend dose one day, and 30% under another.

I think you’re doing wonderful at 3%. What is the content of the entire bar?


The bars are 55g and 300mg @Fullmeltalchemist


Another important point to note about process variability is that it’s not only dependent on the % deviation for a single unit, it’s also highly dependent on your variability over entire batches. In the Xanax example, they may be allowed a high degree of variability in individual pills, but as long as the entire batch variability still falls within tight acceptance criteria, they don’t have to scrap a whole batch.

This obviously gets into some statistics work. IDK how into math you are, but my little dive yielded some interesting information.


Bummer this week, for the Xanax takers that got the strong batch last week…


Sadly, I do not have a quick reference set of pharmaceutical standards to pull for you. I had no idea Phizer would be allowed that kind of variance, and quite frankly, it disgusts me.

The first place I would start with is what your particular local regulations allow for variance. What does the local governing body say about variance? How much to they allow? 3% (and 5-6% for batches) is definitely not terrible, but that may be outside of what local regulators allow. Here in Oregon, the powers that be would not be too keen on that much batch variance, which is why I say something.

If you want any help on increasing homogeneity, I’m happy to help with the tips and trick I’ve picked up from working closely with a highly regarded cannabis-chocolate company. Happy to do that in the thread here or via DM to protect any parts of your process you don’t necessarily feel like sharing.

Another person to tag in that seems to be the forum’s new reigning chocolate expert is @Thechocolatier. I have no doubt that they may have the key to increased homogeneity and batch variance (at least for regulations within Canada).


The law is pretty wide on this - what can be allowed. Meaning if something went wrong with a stability test (for things already in commercial sales) a recall would not be needed as long as it was within this range. The range differs for each compound, with the broadest limits mandated by the FDA.

The tightness comes from your Out of Trend processes. So let us say that you are always within this 3% difference - the FDA requires that you continuously monitor for this, and that if all the sudden you have a batch that is greater than 3% (aka Out of Trend) that you would stop what you are doing and do an investigation BEFORE you released it to the public for sale.

I’ve made a lot of pharma products - and a variability of less than 2% is not uncommon, especially with oral solid dosage like a Xanax. That would mean that even if its would me “legally” allowable to go outside of the trend - an investigation would have happened to say what caused the difference, a corrective action would have been taken (perhaps relabeling, to align with actual dosage), and a preventative action would have been put in place to prevent similar issues from happening again.

As a person who works in Quality Assurance and Control - my personal requirements for tolerable limits are as follows, these have worked well for me for almost 20 years and are aligned with industry best practices in food and pharma, and will hopefully be industry best for cannabis as well.

  • For all dosages, I take the mean point of what I have been able to produce, add 25% to both sides, to create my bell curve, and then look at only 3 sigma from the center, assuming the curve is normal. So the range is broader if a higher dosage and tighter if a lower dosage. This is my FAIL limits.
  • Then for my alert limit, I take that same point of what I have been able to produce, without adding any to both sides, and take 3 sigma from the center, assuming the curve is normal. This is where I’m saying my trend should always be based on all the batches I have made before.
  • The for my normal controls, I take that same point of what I have been able to produce, again without adding any to both sides, and take 1 sigma from the center, assuming the curve is normal. This is where I expect my product to be, what my monitoring tools are looking for, and what I plan my consumables, excipients, etc around.

This has worked for me for a long time. Sometimes my process is SOOOOO variable that the range is incredibly wide (say wider than legally allowed…) with those processes, I have to actually change things before I’m willing to release the product to be consumed by someone. I’ve had plenty of bosses that did not like my approach (cause it means rework or failed batches) but the fact is that consumers want a consistent product, they expect a consistent product, and without these kinds of controls in place you are so inconsistent that your brand will suffer.

Hope this helps and I think you are right where you need to be, if you are already tracking things and getting closer and closer with your percentages.


@Cassin :clap: perfect thank you for that info


@Cassin as always, fantastic well written response!


Can you provide the source document which says this?

1 Like

Of course. The FDA put out a guidance about OOS and OOT results in 2006. Here is a link.

Its probably worth looking into their other guidance that has been listed here on Process Validation - it also discusses OOS/OOT items, expectations, and how to go about it.

If you are looking for an even deeper dive - you can always check out - which has loads of pharma specific guidance documents, which in many cases have even been reviewed by regulators (although not specifically endorsed by them). I’ve found ISPE to be a wealth of information.


Thanks for the reply I guess I still have some work to do. I’m in mass / Maine so I will have to research and see what the officials say I’ve never seen this topic discussed before so maybe Cali / Colorados regs would be a good place to start


Haha thats crazy ur getting a complete homogenized chocolate within 3% in any given point in the bar. Thats better than 90% of california edibles, they just go for total potency per edible, they arent splitting the edible and making sure both sides are the same.
As far as im concerned, youve achieved what u need to, more accurate then anyone else ive heard making chocolates.
Maybe i gotta hit u up for a consultation soon :call_me_hand:


10% is permitted in most pharma products - but that’s thorughout the supply chain, product line and lifespan of the product.
You are measuring variation on one tiny part of that and finding unexpected variation - well within a comfort factor.
Similar small variations occur throughout products thus the seemingly large margins permitted for finished products.

Well done.


@anon56994712 I have to agree that you are certainly on track here. The Canadian regulations are below. If you are consistently seeing the 3% difference in the middle if the bar, the way you cool your bars may be a parameter you want to play with. How many grams are your bars? This may also give a clue. Hit the DM if you want to discuss.

If the labelled cannabinoid content exceeds 5 mg, a 15% variance in quantity is permitted; if the labelled quantity is between 2 mg and 5 mg, a 20% variance is permitted; and if the labelled quantity of cannabinoids is less than 2 mg, a 25% variance is permitted. If a cannabis product that is not in discrete units is represented as being able to be divided into discrete units, each represented unit must contain no more than a 25% variance in cannabinoid concentrationz


So basically, you’re killing it!


This is MA, so this means there can be up to a 100% difference from piece to piece?

Here’s a little list I’ve compiled I’ve got all the info I need I’m moving on going to use +/-10% as the goal

FDA allows 15% for pharmaceuticals
Canada 15% for edibles
Colorado 10%
California 10%

Found this helpful calculator too