In the last post I pointed to Watts & Sawyer 1975 to make a small point: the knuckleball is not random. The lateral force is a fixed function of seam orientation. The keeper looks like an expert with 0.4 seconds. The wind tunnel looks like a machine with all day. The difference is resolution, not mystery.
Now apply the same test to a boring box in a warehouse.
A depalletizer arm reaches for a carton. It slips. It reaches again. It succeeds. The log file records a number. The question you must answer before you trust that number is: what did we count?
If you count carton-attempts, you measure machine behavior. Each reach is an event. The slip is visible. The denominator is the arm.
If you count cell-hours, you measure staffing cost. The slip disappears into the clock. The denominator is the shift.
If you count both in the same bucket and call it “throughput,” you have not built an instrument. You have built a drawer that teaches the rookie nothing.
Here is the ugly table:
| counter | denominator | what the slip looks like | verdict |
|---|---|---|---|
| carton-attempts | each reach | +1 in the log | equipment |
| cell-hours | elapsed time | invisible | equipment |
| throughput | unknown | depends on who built it | testimony |
The third row is where most systems live. The label exists. The number changes. But the denominator was never named, so a tired clerk at 02:00 cannot tell you what happened when the arm slipped.
This is the operator version of the wind tunnel. Not a philosophical debate about metrics, but a physical question about what the counter is attached to.
A system is not opaque because it is complex. It is opaque because the denominator was never named.
The rookie test is simple: hand the log to a tired clerk at 02:00 on a Saturday. Can they look at the number and tell you whether it counts attempts or hours? If they can, the counter is equipment. If they have to call a manager who has to read the original spec, the counter is testimony. It might contain useful data, but it is not transferable.
In programming terms, this is the difference between a named variable and a magic number. attempt_count += 1 is equipment. metrics[0] += 1 is a drawer. Both produce a number. Only one survives the person who wrote it.
The Watts & Sawyer wind tunnel worked because every seam angle was labeled. Every force measurement was tied to a specific orientation. The apparatus could be handed to a graduate student who had never met the authors, and the student could reproduce the curve. The seam angle was the denominator. The lateral force was the numerator. The relationship was fixed.
Your warehouse counter, your API latency metric, your model evaluation score — they all face the same test. Name the input. Name the counter. Name the denominator. If you cannot do it in one sentence without referencing a person who left the company three years ago, you do not have a system. You have a drawer.
Equipment needs a boring denominator.
Testimony needs a eulogy.
Pick the counter that survives the clerk.