Measure Before the Step You Cannot Undo

The recommendation that reordered the protocol

A cross-functional group sat down in a weekly review to walk a draft film testing procedure line by line. The document was in good shape. It named what to measure, tied acceptance criteria back to the plant's own quality metric, and closed the loop on whether the pack still worked for the end use rather than only on whether it ran.

The quality leader's recommendation was not about what to measure. It was about when. Before the packaged product goes into thermal processing, hold dedicated racks. Hold them long enough for a marginal seal to fail if it is going to fail. Read the leaker rate at that point, while the product is still recoverable. His reasoning was one sentence long: once it goes through the cook and reaches the back end, there is not a thing you can do with it, and it becomes direct product loss.

Nothing about the test itself changed in that exchange. The test moved one step upstream, and the cost of being wrong changed completely. That is the whole mechanism, and it generalizes well past film and well past protein.

Irreversibility, not defect rate, sets the cost of a failed trial

A process is a sequence of pressure points, and each one does two things to the unit passing through it. It adds cost, and it removes options. Packaging adds material. Thermal processing adds energy and time and converts the pack into a finished good. Packoff adds labor. Palletizing and shipping add freight and a customer expectation.

A weak seal that exists at the first pressure point exists at all of them. The measured rate usually climbs as the product moves, because each subsequent step applies more stress, more handling, and more temperature to the same weakness. So a single physical defect rate is not a single number in the P&L. Its price depends entirely on where you read it.

Read it before the irreversible step and the failure is recoverable product, a line adjustment, and information you can act on. Read it after and the same failure is scrap at full conversion cost, because you have already spent the energy, the labor, and the material on a unit you cannot sell. Read it after the dock and the price is a customer complaint, a credit, freight in both directions, and a conversation about whether your specifications are under control.

That last one changes behavior for years, and it deserves to be named plainly. Nobody wants to learn about a film change from a customer. The cost there is not the product. It is the credibility of every future specification change anyone proposes. A plant burned by one validation escape becomes a plant that quietly stops validating, which becomes a plant that quietly stops changing specifications. Material and packaging decisions are exactly where recoverable margin sits, so an organization that has lost its appetite for a specification trial has lost access to its own best cost lever. The protocol is not only protecting product. It is protecting the plant's willingness to change anything at all.

The threshold at an upstream gate is not the end-of-line target

This is the part most trial procedures get wrong, and the error is structural rather than sloppy.

If a plant's tolerance is a scrap rate measured out the back end of the process, that number is the wrong acceptance criterion for a read taken upstream, because the rate rises between the two points. A gate placed before thermal processing needs a tighter threshold than the end-of-line target, set so that passing it is consistent with hitting the end-of-line number after the remaining steps have added their share.

The quality leader walked exactly that logic in the review, using round numbers he was explicit about having chosen for the sake of the argument rather than measured. If the tolerance out the back end is 1 percent, and a held rack reads 2 or 3 percent before the cook, then the post-cook number will be higher, and the number that reaches the customer higher still. The figures were illustrative. The structure behind them is not. Without some estimate of the amplification between your gate and your dock, an upstream gate will cheerfully pass product that fails where it counts, and you will have built a procedure that manufactures false confidence on a schedule.

Getting that amplification is ordinary work and it does not require precision. For each step between the gate and the dock, you want a rough multiplier out of your own history: of the seals that were marginal at the hold, what share became failures after thermal processing, after packoff, after transit. Knowing whether a step roughly doubles the rate or adds almost nothing is enough, because that is what tells you whether your gate threshold should be half the target or close to equal to it. If nobody in the plant can answer that question for the largest step in the chain, that gap is the first thing worth measuring, ahead of any supplier trial. A gate whose pass line you cannot derive is not a gate.

The corollary is the cheapest discipline in the entire procedure, and it is the rule the group landed on. Do not advance to the next pressure point if you cannot pass the one in front of it. Then escalate exposure in deliberate stages: one rack, then a few racks, then an hour of run, then a full pallet build. Each stage puts more product and more scheduled line time at risk than the last. That is the design intent, not a compromise. The cheap stage is the one that is supposed to fail.

The trigger is the change, not the size of the change

The natural instinct is to scale the procedure to how large the change feels. Full validation for a new structure. A wave-through for the same specification from a different supplier.

That instinct is backwards. The reason a like-for-like swap looks attractive is that nobody expects it to behave differently, and that is precisely the condition under which nobody is watching. Resin lots differ. Converting equipment differs. Slitting, winding, and coefficient of friction differ between two suppliers running the same data sheet. Yield loss out of a specification that reads identical on paper is a familiar way to give back savings that were already booked.

So the trigger is the change itself. A different supplier at the same specification. A different specification from the same supplier. A gauge change. A structure change. One protocol, one trigger, and no judgment call about whether this particular change is the kind that deserves scrutiny.

My own rule goes one step further, and I said it to that group directly: the artifact has to outlive the project. The protocol needed to be theirs, written so any future film decision could run through it with no outside party in the room. A validation procedure that only executes while a consultant is present is not a control. It is a test that happened once.

Sequence the commercial gate ahead of the line-time gate

The same program carried a second ordering decision, and it follows from the identical principle.

Before committing plant product and scheduled line time to physical validation, quantify what each alternative specification is actually worth. Take the current specification and the proposed alternatives to market, price them with the incumbent and with real challengers, and put a number on the prize. Then decide whether a physical trial earns the line time it will consume.

That ordering matters because the resources are not interchangeable. A commercial screen costs a spreadsheet and a handful of supplier conversations. A physical trial costs finished product, scheduled line hours, and quality attention, and quality attention is the scarcest of the three in most mid-market plants. Running the trial first and the pricing second means you sometimes spend the scarce resource to validate a change that was never worth making. Cheapest gate first, in that order, every time.

What a well-run material change reads like

One written protocol, owned by the plant, triggered by any material change including a like-for-like supplier swap. A first quantitative gate placed upstream of the irreversible step, with a hold long enough for the failure mode to express itself rather than a reading taken at the moment of production. A pass threshold at that gate derived from the end-of-line target and the known amplification of the steps in between, written down with its arithmetic. Sample volume that escalates one stage at a time, each advance gated on a written pass of the stage before it. Acceptance criteria tied to the plant's own quality metric and to the end use of the product, not to the supplier's data sheet. A commercial value quantified before any line time is committed.

Find your two steps

Pull the material change procedure your plant runs today. Find the first step that produces a number. Find the first step that cannot be undone. If the second one comes before the first, you are not validating a change. You are discovering it, at the worst available price, from the person least willing to absorb it.