Finished Product Specs: The Numbers That Decide Whether a Batch Ships

Every batch of your sauce leaves the plant because somebody decided it was good enough to ship. The useful question is who decided, and what they compared it to.

If your plant is approving batches by eye, the honest answer is: the product looked, smelled, and tasted right to an experienced person. That person may be very good. But "right" is being measured against their memory of your product, on a busy day, with no written line that says where right stops. When a product drifts slowly, run over run, memory does not catch it, because each batch looks like the one before it.

The fix is a finished product spec written as release criteria: the short list of measured numbers and checks that decide whether a batch ships, is held, or is rejected, with what happens on each outcome written down before anyone needs it. The plant still runs QC and releases product through its own system. You are giving them a standard worth releasing against.

I have written before about what belongs in a production-ready recipe, including a quality spec with targets and ranges. This article picks up where that one stops: turning those numbers into a decision a plant can make the same way on every batch.

Four Documents People Call "the Spec"

Part of the confusion is vocabulary. Brands and plants use "the spec" for four different things:

The formula: what goes in, by weight, with an ingredient spec for each input.
The process notes: how it is made, in what order, at what temperatures and times.
The finished product spec: what comes out, described in numbers and in sensory terms.
The release criteria: which of those finished numbers are gates, how each one is measured, and what happens when one misses.

The first three describe your product. Only the fourth tells someone on a Tuesday afternoon whether the pallet goes on the truck. If you have the first three and not the fourth, the plant is releasing by eye even though a spec exists.

Sort Every Attribute Into a Tier

Not every number deserves to stop a shipment. Before writing ranges, sort each attribute into one of three tiers.

Tier one: safety and process gates

For an acidified product, finished pH is a gate set by the scheduled process. So are the process records the plant keeps: cook and fill temperatures, hold times, and closure checks. These belong to the plant's food safety plan and to your process authority. They are not yours to loosen, and a miss here is never a conversation about whether the batch is "close enough." Water activity sits in this tier for a product that relies on it for shelf stability, and outside it for one that does not.

Your job in tier one is narrower: make sure your finished product spec agrees with the scheduled process, so nobody works from two different numbers. And know that the pH read on fill day is not always the pH the product settles at, which I covered in pH drift after production.

Tier two: product gates

These protect the product your customer bought last time: Brix, viscosity, salt, color, and sensory against the gold standard. A miss here does not make the product unsafe. It makes it a slightly different product. These are the numbers a by-eye release is most likely to let slide, one small step at a time.

Tier three: package and label gates

Net contents, the correct label, a legible lot code, and how the package looks. (Closure and seal checks that the scheduled process relies on stay in tier one.) Fill weight gets checked through the run rather than once, because a single jar tells you little about a run. The plant's own procedures usually cover most of this tier already.

Anything else you measure, record it but do not hold for it. Yield, for example, is worth tracking on every batch and is not a reason to stop a shipment.

Targets, Ranges, and Limits Are Three Different Lines

A spec that only says "Brix 38" leaves the plant to guess what 37.2 means. Each tier-two attribute needs three things:

Target: where the plant aims.
Release range: the band in which the batch ships without discussion.
Hold limit: the point beyond which the batch stops and someone with authority decides what happens next.

Some attributes only need one side. pH has a ceiling that matters for safety, and your flavor sets the floor, if you need one at all. Viscosity and salt usually matter in both directions, though you may care more about thin than thick, or the reverse, depending on how the product is used. Think each one through instead of writing the same plus or minus everywhere.

Where the ranges come from

Three inputs, and you need all three.

What a customer would notice. Make small, deliberate variations on the bench and taste them against the gold standard. The range should stop before the difference becomes noticeable to someone who buys your product every month.
What the plant can actually hold. Collect the real numbers from several approved batches. If your range is tighter than the plant's natural spread, a steady share of good batches goes on hold, people get tired of it, and the spec stops being taken seriously.
What the measurement can tell apart. A range narrower than the instrument's repeatability is a coin toss. If two readings of the same sample differ by more than your range, fix the method first, or the range means nothing.

Sometimes the first two disagree: a customer can taste a difference smaller than the plant's normal spread. That is not a release problem. It is a formula problem, because the product is too sensitive to ordinary variation, and the fix belongs in the formula and the ingredient specs. For how I write a single attribute line in full, with target, range, rejection points, and method, Brix targets for BBQ and condiment manufacturing walks through one.

A Number Without a Method Is an Argument Waiting to Happen

Every release attribute needs its method written next to it, or the plant and you will measure the same jar and get two answers. For each one, specify:

The instrument: pH meter, refractometer, the type of viscometer or consistometer, salt meter or lab method, visual color standard or color instrument.
Sample prep: blended or not, strained or not, oil layer removed or not.
Sample temperature: pH, Brix, and viscosity readings all move with temperature.
Timing: at the line, after cooling, or at a set number of hours after fill.
Which units: start, middle, and end of the run, not three jars from the same minute.

Timing is the easiest one to leave out. Viscosity can keep building for a day or more as starches set up, as I explained in sauce viscosity at scale. pH can creep up as chunks equilibrate. Flavor takes roughly two weeks to find its level. Your release decision is made long before week two, so the release range has to describe what a good batch reads at the time of release, not after it settles. That link is built from data: measure your approved batches at both points for a while, and the right release range shows itself.

Sensory Against the Gold Standard, at Two Speeds

Sensory is the hardest gate to write and the easiest one to leave to memory. Without a reference, "tastes right" moves a little with every batch. Your gold standard is the approved reference version of your product, held at the plant and by you, and every sensory check is a side-by-side against it.

Because the flavor is still moving when the batch ships, I split sensory into two checks:

At release: a quick check at the plant, side by side with the reference, for defects and character. Is anything scorched, flat, sour, or off? Does the heat sit where it should? Is the color in the family? This is pass or hold, not a score.
At week two: a slower comparison on the retained sample, judged against the gold standard for taste, acidity, salt, texture, and color. It cannot stop a batch that has already shipped, but it tells you whether your release checks are catching what matters, and it is where slow drift shows up first.

Write sensory in words both sides can use: the attributes you care about, and what "out" looks like for each. "Garlic present but not raw" is a usable line. "Tastes like ours" is not.

Who Measures What

Quality control is mostly the co-packer's job, and it should be. They have the instruments, the people on the line, and the food safety plan. Your role is to give them a spec worth holding and to check the result from the outside. In practice that means agreeing, in writing, on:

Which release checks the plant runs on every batch, and with what equipment. If they do not own an instrument your spec calls for, decide together whether to buy one, send samples to a lab, or change the method.
What the plant sends you for each batch: usually a short batch report with the release numbers, the lot code, and who signed off.
A sample from every batch, sent to you, so you can test it yourself against the gold standard.
Which tests go to an outside lab, and how often, for anything neither of you measures in house, such as water activity on some products or periodic micro testing if your process authority or a customer asks for it.

This is the same discipline as the go and no-go sheet in a pilot batch protocol, made routine. The pilot proves the product once. Release criteria prove it every batch.

What Happens on a Miss

Decide this before you need it. A miss with no written path gets settled by whoever is in the biggest hurry. The usual outcomes, in order:

Hold. The batch is tagged and stays put. Nothing ships while a question is open.
Confirm. Check the instrument, the sample prep, and the temperature, then measure again. One confirmation is diligence. Re-sampling until something passes is not, so the spec should say how many re-tests are allowed.
Release with sign-off. For a small tier-two miss, the brand can choose to accept the batch, in writing, with the reason noted. This is a quality call, and it should be rare. It is never available for tier one.
Rework. Some products can be adjusted, or blended back into a later batch. Whether that is allowed, and how, is set by the plant's food safety plan and, for acidified products, your process authority. Rework is a process decision, not something to improvise on the floor.
Reject. The batch does not ship as your product.

Whatever the outcome, someone looks for the cause (an ingredient lot, a process step, the measurement, or the formula itself), and the finding goes into the batch file. Who pays for a held or rejected batch is a contract question for you and your co-packer, not something your spec can settle.

Moving a Plant From "Looks Right" to Numbers

If your plant has been releasing by eye, do not land a forty-line spec on them by email. The people on that line know your product, and their judgment has been carrying it. The aim is to give their judgment a reference.

Ask what they already measure and record. Build on their instruments and habits wherever you can.
Run several batches where they record the numbers but release the way they do now. You are learning the real spread, not grading anyone.
Lay those numbers next to your week-two tastings. Set ranges from what good batches actually read.
Agree on the tiers, the methods, and the miss path in one conversation with their QA lead, then put it in the spec.
Review it after a few months. A range that never triggers may be too loose. One that triggers constantly is telling you something about the formula or the process.

A finished product spec for each SKU is also one of the documents the plant will want from you when its own auditor visits, as I covered in what your co-packer's auditor will ask you for. Release criteria make that document do daily work instead of sitting in a binder.

Questions Brands Ask About Release Criteria

Should shelf-life results be part of release?

No. Release is a per-batch decision made in hours or days. Shelf-life testing validates the product and the date you print, over months. A loose release range can quietly shorten your real shelf life, but the two are separate documents with separate jobs.

What if a batch passes every number and still tastes wrong?

Hold it. The numbers are there to catch drift a person cannot see. They do not overrule a person who can taste a problem. Then work out what your spec is missing, because a batch that clears every gate and still tastes wrong means a gate is missing or a range is too wide.

How many attributes should release criteria have?

As few as it takes to protect the product. Every gate costs the plant time on every batch. A short list that gets measured every time beats a long list that gets skipped when the line is busy.

Setting the Ranges Is the Real Work

Listing the attributes is the easy part. The harder part is setting ranges that predict the week-two product from a fill-day reading, finding which ingredient or step drives your spread, and making the formula sturdy enough to sit comfortably inside the range the plant can actually hold. That calibration is specific to your product and your plant, and it is the work I do with brands inside a project, starting from their formula and the plant's real batch numbers.