OEE at 78% and the line still can't ship. Look at the stops under a minute.

Every year somebody publishes OEE benchmarks and every year a plant manager gets handed a slide that says their line should be at 85%. The latest round is out — Godlan's 2025 vertical breakdown has numbers for plastics, metals, food, and the usual suspects. It's a fine reference. It's a terrible target.

The problem with OEE as a reported KPI is that the number gets managed before it gets understood. The problem with OEE as a diagnostic is that most plants aren't measuring the thing that's actually dragging it down.

The stops that don't get logged

If your availability number is coming from downtime reasons entered by operators, you are missing most of what's happening. Operators log the 20-minute stop. They do not log the 40-second stop that happened eleven times this shift because the infeed hiccuped. That's not laziness. That's physics — nobody can key a reason code faster than the line can recover.

ReliaMag's piece on the hidden factory makes the right argument: micro-stoppages are a measurement problem before they're a maintenance problem. If a stop is shorter than your logging threshold, it rolls into performance loss, or it disappears entirely into a historian gap nobody looks at. Either way, your availability bucket looks healthy and your performance bucket looks like a mystery.

In steel, the shape of this is specific. iFactory's writeup on sub-10-minute stoppages in steel plants walks through where they cluster — cobbles at the finishing stands, cast-end transitions, coil handoffs. None of these are rare. All of them are short enough that the shift report calls the line "running."

The bottleneck is usually upstream of where it looks

Here's the pattern we see most often. A downstream station reports 85% availability and gets flagged as the constraint because it's the slowest reported step. You go stand next to it for an hour. It's starved half the time. The real constraint is two stations upstream, dropping out for 30 to 90 seconds at a time, 15 to 40 times a shift. The downstream OEE report is a symptom, not a diagnosis.

On one extrusion line we worked, moving the micro-stop category out of "performance loss" and into its own tracked bucket added roughly 6% to effective run time in the first month. We didn't fix anything mechanical. We just stopped hiding the losses in the wrong column.

What to actually do

Three things, in order:

1. Instrument machine state at sub-minute resolution from the PLC or the historian, not from operator entry. If your tag dictionary doesn't distinguish "running," "idle," "starved," and "blocked," fix that first. 2. Reconcile. Match operator-logged downtime to state transitions. The delta is your hidden factory. 3. Rank micro-stop causes by cumulative minutes, not frequency. The annoying one is rarely the expensive one.

Then you can talk about whether 78% is good. Probably it isn't. But at least you'll know why.