Condition monitoring is not a new idea. In industry it is entirely unremarkable: you instrument a pump, a turbine, a compressor; you learn what healthy looks like; you catch degradation long before it becomes failure. It has worked for decades and nobody argues about it.

It has never reached the boiler on the wall of a house, and the reason is arithmetic rather than physics. The sensing and the analysis have always cost more than the asset. Nobody is going to put a five-thousand-pound monitoring rig on a nine-hundred-pound combi.

The result is that the most safety-critical machine in most homes is also the only one nobody watches. It runs until it stops.

The constraint that shapes everything

The obvious way to instrument a boiler is to go inside it, tap the control board, read the manufacturer's diagnostic bus, integrate with the appliance.

That approach dies on contact with reality. Anything intervening in a gas appliance's operation is gas work: certification, a qualified engineer, a callout. The thing designed to save a callout has just cost one. And it needs either the manufacturer's cooperation or a reverse-engineered protocol that breaks with the next firmware revision.

So the constraint is inverted, deliberately: it observes and does nothing else. It adheres to the outside of the boiler casing. Nothing is opened, nothing is cut, nothing is wired in. It touches neither the gas path nor the heating circuit nor the control board.

That is not a compromise to keep costs down. It is the thing that makes the economics possible at all. Something that only observes needs no certification, no engineer, and no manufacturer's permission, and it is not tied to any manufacturer's protocol, so it does not break when the protocol does.

The part that matters

The natural question is whether anything useful survives that constraint. If you cannot read the fault codes, what is left?

Not a bad noise. By the time a fault is intermittent, a human being has already noticed it. The boiler is announcing its own decline. Most of the value has gone.

The interesting window is much earlier: while the machine is still, to every human sense available, completely fine.

What the device builds instead is an abstract representation of the boiler's behaviour, a learned space with considerably more dimensions than a person has senses for. Degradation appears as movement through that space, long before it surfaces as anything anyone could name. Not a louder pump. Not a hotter case. A drift in the relationship between things, in a direction that has come to mean something.

The signal is there the whole time. It is simply not in a form that human perception has any access to. The abstraction is not a convenience for the model; the abstraction is the evidence.

And the baseline is learned per installation, because a twelve-year-old combi in a cold garage does not behave like a new one in a warm cupboard, and any threshold that assumed otherwise would be useless in both.

It compounds

A model built on one boiler knows one boiler. A model built on thousands begins to know what degradation looks like in general, and can recognise it earlier on a machine it has only just met.

That is the long game, and it is the opposite of a product that is finished the day it ships. The hardware is the same on day one and on day a thousand. What changes is how much it has seen, and therefore how early it can say something.

The scale question

For a householder the payoff is not being caught out: a service booked on a Tuesday in October rather than an emergency callout on Christmas Eve, a replacement planned rather than forced.

For anyone answerable for a portfolio of boilers it is a different proposition entirely. A landlord, a letting agent, a housing association, a facilities manager: hundreds of these machines, no way of knowing which of them are unwell, and a maintenance budget spent reactively, at emergency rates, in the coldest week of the year.

Knowing in October which three of two hundred boilers will fail this winter is worth a great deal more than the devices cost. Same technology, same device, the economics simply improve the more of them you are responsible for.

Where it is

In development. There is no product to buy, no pre-order, and no date I am willing to promise, because I do not have one.

What exists is a principle that is proven everywhere except here, hardware being built and broken, and a question that is genuinely still open: how much of a boiler's condition is legible from the outside of the case? That is an empirical question. It is being answered empirically, and it may yet be answered badly.

The site is at boilerbeacon.co.uk.