The buffer illusion
Two lines can report the same interruption rate while one is healthy and the other is quietly falling apart. The difference is the stop threshold, a configuration choice that can turn a chronic problem into a clean-looking number. Here's how the illusion works, and the index that breaks it.
- Stop thresholds range from an aggressive 30s to a lenient 5 minutes.
- A lenient threshold lets operators clear faults before any breakdown is coded.
- That masks chronic interruptions and deflates unplanned-downtime reports.
- The Threshold Exposure Index (MTBI ÷ threshold × 100) exposes the illusion.
Measure this on your own line. One sensor per line, live within 48 hours of the kit arriving.
The same number, two very different lines
Put two lines side by side. Both report an interruption rate (interruptions are what standard OEE calls minor stops or micro-stops) in the low teens. One is genuinely stable. The other is stopping every fifteen seconds. Nothing in the headline number tells them apart, and that is exactly the problem. The variable doing the hiding is the stop threshold.
How the buffer masks the mess
A stop threshold is the delay before a pause is logged as downtime. Set it aggressively, at 30 seconds, and almost every stop shows up. Set it leniently, at 5 minutes, and a great deal disappears: operators clear the fault and restart the line before the timer ever trips a breakdown code.
One studied line reported a comfortable 12.86% interruption ratio, apparently optimised. Its real time between stops was 15 seconds. The generous 5-minute buffer let operators absorb a relentless stutter of faults invisibly, so the chronic problem, and the mechanical stress behind it, never reached a single management report.
Worse, the trick works in two directions at once: it inflates the interruption ratio's apparent stability while deflating unplanned downtime, because the breakdowns that would have been coded got resolved under the timer instead.
The index that breaks it
To standardise across lines with wildly different thresholds, we use the Threshold Exposure Index:
- TEI = MTBI ÷ stop threshold × 100
It asks a simple question: how does a line's actual time between stops compare to the buffer it's allowed? The 12.86% line above scored a TEI of just 5.06%, stopping roughly twenty times faster than its threshold, an unmistakable red flag. A line with an aggressive threshold and long runs, by contrast, scores well above 100% and can be trusted. Any low TEI is a warning that the reported numbers are masked.
Clean data, real signal
The payoff of stripping the illusion is sharper analysis everywhere. Across the raw portfolio, MTBI and unplanned downtime correlated at a weak r = 0.41. Filter to the honest, high-TEI lines and the correlation jumps to r = 0.73. The buffer wasn't just hiding losses. It was hiding the relationships that explain them.
Check the TEI before you trust any downtime report. The buffer illusion is one of the core findings of the Interruption Whitepaper, and the reason MTBI should never be read on its own.
What is a stop threshold?
It's the delay a line waits before counting a pause as downtime. Below the threshold, a stop may be treated as an interruption or ignored; above it, the stop is logged with a reason code. Thresholds in the study ranged from 30 seconds to 5 minutes.
How does a threshold create an illusion?
With a lenient 5-minute buffer, operators can clear a fault and restart the line before any breakdown code triggers. The chronic stopping never reaches the downtime report, so the line looks far healthier than it is and its unplanned downtime looks artificially low.
What is the Threshold Exposure Index?
TEI = MTBI ÷ stop threshold × 100. It normalises the time between stops against the configured buffer. A low TEI (well under 100%) means the line is stopping far faster than its threshold, so its reported numbers are masked and untrustworthy.
Why does this corrupt other metrics too?
Because masked breakdown data weakens every downstream analysis. In the study, filtering out low-TEI lines lifted the correlation between MTBI and unplanned downtime from r=0.41 to r=0.73. The noise was hiding the real signal.
See it on your own line.
A 30-minute call with a manufacturing specialist. We look at one of your lines together and show you what the first week of data would say about it.
- Live within 48 hours of your kit arriving
- One sensor, one terminal, no PLC and nothing to integrate
- R4,950 per line per month, ex VAT. 90 days, then a month's notice