What we learned from 400 rejected bottles

Before the exercise, if you had asked the yard what kills a bottle, you would have got answers about forklifts. Afterwards, nobody says that.

Crazing was the headline

Twenty-nine percent, and almost all of it concentrated in containers that had held solvents or aromatic-containing blends. Crazing is a network of hairline cracks that forms when a polymer under stress meets a chemistry it does not like. Polyethylene is highly resistant to a great many things and quietly vulnerable to a few.

The practical lesson is that crazing is a chemistry problem masquerading as a wear problem. A container that has held an aromatic solvent for eighteen months is not worn out; it has been chemically attacked, and no amount of gentle handling would have prevented it.

Odour was the surprise

Twenty-three percent of rejections were bottles that came off the line looking perfect, held pressure, showed no film, and smelled of their previous life two days later.

That delay is what makes odour dangerous to grade on. A bottle assessed straight off the steam finish smells of steam. We changed our process on the strength of this data: units now sit for at least twenty-four hours with the cap off before an odour call is made.

Deformation is an outdoor storage problem

Fifteen percent, and when we looked at where those containers came from, the pattern was obvious: units stored outdoors, full or part-full, through at least one winter.

Water expands as it freezes. A bottle that freezes solid takes a set, the walls stop being parallel, and the cage bows outward at the mid-height bars. The container is often still watertight. It is no longer stackable, no longer square in a rack, and no longer a candidate for regulated service because the design type assumed a shape it no longer has.

What this changed about buying

• We ask about solvent history explicitly at intake, because it predicts crazing better than any visual check.
• We ask whether containers were stored outdoors over winter, and we price accordingly.
• We hold before grading, which lowered our post-sale complaint rate more than any other single change.
• We stopped treating forklift damage as the main risk. It is real, it is visible, and it is thirteen percent.

Where the 411 went

Every one of those bottles failed for reuse. None of them was wasted. Three hundred and ninety-four cages underneath them were sound and went to the rebottling bench. The bottles were ground to flake.

That is the number worth ending on: four hundred and eleven bottles reached end of life, and three hundred and ninety-four containers stayed in service anyway. The bottle is the consumable. The cage almost never is.

What we changed as a result

Four process changes came directly out of the exercise, and three of them cost nothing.

1. 01Hold before grading. Twenty-four hours with the cap off before an odour call. This alone removed most of our post-sale complaints.
2. 02Ask about solvent history explicitly at intake, because it predicts crazing better than any visual inspection.
3. 03Ask whether containers were stored outdoors over a winter, and price accordingly rather than discovering it during inspection.
4. 04Move torch inspection indoors. Crazing is invisible in daylight, which is where we used to grade in summer.

What we did not change

We considered tightening the rejection criteria after seeing the crazing figure, and decided against it. The criteria were not the problem — the detection was.

There is a temptation, when a failure category is large, to reject more aggressively across the board. That would have scrapped containers that were perfectly serviceable and would not have caught a single additional crazed bottle, because the crazed ones were passing inspection rather than being judged leniently.

How the numbers compare to the previous year

The crazing and odour figures went up, which sounds like a worse result and is a better one. We were not rejecting more bottles; we were catching more of the ones that should have been rejected all along and were previously reaching customers.

Why we published this

Partly because customers ask what our reject rate is and a number without a breakdown is not an answer. Mostly because the distribution is genuinely counterintuitive, and anybody running a fleet who thinks forklifts are the main risk is inspecting for the wrong thing.

What we would measure next

Three things we do not currently track and probably should.

• Time from retirement to arrival. Our strong suspicion is that most deformation and contamination damage happens after a container leaves service and before it reaches us, and we could prove it by asking one question at intake.
• Cycles completed before failure. We know why bottles fail; we do not know how many fills they achieved first, because that history lives with customers rather than with us.
• Source-level failure rates. Some sellers consistently send stock that passes and some consistently do not, and we know this anecdotally rather than numerically.

The general lesson from the original exercise applies here too: the number you do not record is the number you cannot improve, and the distribution is usually not what anybody in the building would have guessed.