One bad batch is a fact. A climbing failure rate is a warning.

You know a batch failed the moment you open the vessel.

Sometimes it is mold. Sometimes the smell is wrong: fruity, faintly alcoholic, nothing like what should be coming out of that ferment. Sometimes it is the brine, the wrong color and crystal clear when it should not be. Any one of those and the batch is done. It gets discarded, and it takes with it the ingredients you paid for, the labor that went into it, the facility time, and however long you now spend rearranging everything downstream that assumed those cases were coming.

That is a bad day, and every producer reading this has had one.

Here is the harder problem. That failed batch is a fact. You saw it, you dumped it, you moved on. What you cannot see standing in front of an open vessel is whether it was bad luck or the third instance of something that is going to keep happening.

01 One failure is an incident. Several is a pattern you are paying for.

At Off The Deck Hot Sauce we went through a stretch of experimentation, and experimentation is exactly the condition where this gets hard. You are changing things on purpose. When a batch fails, you have a ready explanation, because you were trying something. So each failure gets logged mentally as a one-off, filed under the change you happened to be making that week.

The details that would connect them are in the records. They are just spread across separate rows, separate runs, separate weeks, and nobody sits down and lines them up. Especially not while you are in motion.

What finally broke it open was not our records at all. I read a water quality report that had hints in it. Then I went back and compared it against our production data, and the picture came together: the failures were not independent. They shared a cause, and the cause was our water.

Our failure rate has been falling ever since.

02 The investigation is where the pattern comes from

The order of operations matters here.

The report did not tell me anything about my batches. My batch records did not tell me anything about water. The finding lived in comparing the two, and I could only do that comparison because there was something on the production side worth comparing to.

That is why the investigation after a failed batch earns its time. As a small producer, every failed batch matters, and every one deserves the same set of questions:

• Was the SOP followed?
• Was anything forgotten or done out of order?
• What was different about this run compared to the last one that worked?
• What conditions were the same as the last failure?

Then write it down. Not in your head, not in a text to yourself. In the batch record, attached to the batch it belongs to, while it is fresh.

Notes are the whole thing. A pattern is only visible if the individual data points survived long enough to be compared. Most small producers lose the pattern for one reason: the detail that would have mattered was never written anywhere it could be found six months later. That is the same discipline behind keeping batch records clean in the first place.

03 The math, once you find it

Once we identified the water pH issue, two things happened.

First, we told other makers it could affect. If your water is doing this to your ferments, it is probably doing it to somebody else's, and that is not information worth sitting on.

Second, we found a cost-effective supplier of purified water and eliminated the variable entirely.

That is easy math. It is not close. But you cannot run that math at all until you know which variable to change, and you cannot know which variable to change until the failures stop looking like separate accidents. A one percent move in what a batch actually costs is nothing next to the batches you stop throwing away.

The fix was cheap. It was cheap the entire time. The expensive part was the months of discarded batches before we could see what they had in common.

04 What a spreadsheet can and cannot do here

I want to be honest about this, because the easy version of this argument is not true.

Could I have found the water problem in a spreadsheet? Yes. The data was there. But it took a lot of time, jumping between files, doing hand calculations, and then re-checking the calculations because I could not fully trust that I had pulled the right rows.

The cost of the spreadsheet is not that the answer is unreachable. It is that the effort required is high enough that you only do it after the problem has already gotten expensive enough to justify a Saturday. Nobody reconstructs a failure-rate analysis on a Tuesday for fun. So the analysis happens late, or it happens never, and the batches keep failing in the meantime.

A number you could calculate but never sit down and calculate is not doing you any good.

05 What to watch, and when it means something

If you start reading your own production this way, keep four things straight.

Watch the rate by product, not just overall. A whole-operation failure rate can look perfectly healthy while one product line quietly gets worse. The number that carries a warning is the one broken out per product, and per year, so you can see direction and not just position.

Direction matters more than the absolute number. Two percent is not inherently good or bad. Two percent trending up on one product across three seasons is a signal. The same two percent flat for two years is just your process.

Small numbers need care. If you have run forty batches of something, one failure moves your rate by two and a half points. That one failure still deserves the full investigation. What it does not deserve is a trend line drawn through it. The rate becomes trustworthy as the batch count grows behind it. Until then, your notes are the evidence and the rate is context.

And the comparison that cracks it open may come from outside your own data entirely. Mine came from a water report. Yours might come from a supplier's CoA and ingredient lot records, a seasonal change in a raw ingredient, or a facility issue that shows up in a completely different document. Keep your production records in a state where you can check an outside hypothesis against them without a two-hour reconstruction.

06 Where this lives in FourFoxes

Production Analytics adds up records your team already logged during normal production. It shows failed-batch rate for every product and year, color-coded so a climbing rate stands out, and sortable so the products that need attention surface first. It shows what is ready to package and what is still in progress, and median production duration from start to packaging.

None of that is separate work. There is no month-end assembly step, because the batch records, CCP checks, and ingredient lots are already in one place. The analysis is built from what the floor captured while the work was happening.

What it will not do is tell you why. That part is still yours: the investigation, the notes, the outside document you think to compare against. The software's job is to make sure the pattern is visible when you go looking, instead of buried across two hundred rows you will never have a free Saturday to reconcile.