The Eight Things We Deliberately Do Not Alert On

The failure mode of a monitoring product is not missing a fault. It is crying wolf. A wasted truck roll costs a customer relationship; two cost the product its credibility, and after that the real alerts go unread along with the false ones.

So the harder half of the work is deciding when not to speak. Here are eight conditions that would fire a naive threshold product, on equipment with nothing wrong with it, and what we do instead.

1. A hot day

Head pressure rises with ambient. A fixed high-pressure threshold therefore alarms on the hottest afternoons of the year — precisely when the phones are busiest. Our high-side logic anchors to condenser split and subcooling, never to raw pressure, so a healthy system on a 105-degree day stays silent.

2. An inverter idling below its modulation floor

A variable-speed compressor that cannot turn down any further has to cycle to make less capacity. That is the machine working correctly. Cycle counting alone reads it as short cycling.

3. A heat pump in heating

When the reversing valve swaps the coils, superheat and subcooling are physically inverted — the outdoor coil is now the evaporator. Without a reversing-valve signal we cannot reliably tell heating from cooling, so we suppress rather than guess. A heat pump in heating is monitored for connectivity and nothing else, and we would rather say that than publish a diagnosis we cannot stand behind.

4. A heavy internal load

A full restaurant, a server room, a west-facing afternoon: superheat rises toward the top of its band while subcooling stays flat. A genuine undercharge moves both. That difference is the test.

5. An undersized system

Runtime pinned near 100% with every channel in spec is not a fault signature. It is a sizing decision somebody made years ago. Ranking units by runtime alone would put these at the top of every list forever.

6. Windows open

The system cannot satisfy setpoint and never will, but the refrigerant circuit is fine and the ratios say so. This is a building problem, not an equipment problem, and dispatching a technician for it wastes everyone's afternoon.

7. Startup transients

Superheat sits near zero for the first seconds of every morning start, on every healthy system, forever. Only settled samples are judged.

8. Not enough data

If there are not enough steady runs to be confident, the answer is silence — and the unit reports as still learning rather than as healthy. “We have not seen enough to say” and “we looked and found nothing” are different sentences, and a product that collapses them into one is quietly lying to you.

Why this list is the specification

Every one of these is a real condition that occurs on healthy equipment, and every one is verified silent against a fixture that drives it through the real pipeline. That is the part worth asking any monitoring vendor about — not what it detects, but what it has been tested to ignore.

There is a related rule we hold to: a reading merely outside a textbook band, with no meaningful cost to performance or equipment life, is recorded for the technician and never alerted. Imperfection is not a fault. If the electricity saved would not cover the visit, the correct behaviour is to stay quiet.

This post describes how our own alert engine is built. Where it states a figure from published research the source is linked; where it describes our engineering decisions, that is what it is, and we have labelled it rather than dressing it up as a finding.