Guidance is written for intake conditions.

A room average describes nothing a server experiences.

Comfortable halls still contain badly fed racks.

Only inlet sensing finds them.

The measurement point decides the value of the data.

Position before quantity, always.

Three is the common arrangement.

Low, middle and high on the front face.

One at mid height hides both extremes.

The top is where recirculated air arrives first.

The gradient shows whether containment works.

A single point cannot show a gradient.

The temperature rise across the rack.

That reveals airflow problems inlet readings miss.

A starved rack shows a large rise.

Its inlet can still look perfectly correct.

Differential data diagnoses, single points alarm.

Worth it on high density racks at least.

Wired where cable can be run at build.

Wireless suits retrofit into a live hall.

Cable crossing containment is disruptive.

Wireless brings a battery cycle at scale.

Thousands of points means a real programme.

Somebody has to own that cycle.

The one the hardware vendor states.

Industry guidance gives recommended and allowable bands.

Allowable is not a target to run at.

Time spent near limits shortens equipment life.

Record excursions, not just current values.

Agree the band before setting alarms.

Accurate enough to trust a small margin.

Raising supply temperature saves real energy.

That is only safe with credible measurement.

Poor accuracy forces a conservative setpoint.

Check drift and calibration interval.

Cheap sensors cost energy indirectly.

Data arriving where nobody acts.

Alarms configured then permanently ignored.

Sensors displaced during rack work.

Positions undocumented after installation.

Record the position of every point.

Re-check placement after any hall change.