I know that a singular market focused opinion is a data point and not something we can base a conclusion on thus I can't solely take your points onboard. I know as a contractor, looking in from the outside has a lower degree of assumptions than looking out from the inside.
I disagree on a couple of your points.
Firstly, I can't possibly agree that anything with the complexity of a Nest device is a good thing. The thing is an IoT device that actually presents multiple attack surfaces and doesn't necessarily have sound functionality. I cite this as my source:
https://www.cpsc.gov/Recalls/2014/Nest-Labs-Recalls-to-Repair-Nest-Protect-Smoke-CO-Alarms/The failings are:
1. Nest Wave was turned on.
2. This disabled the smoke alarm if there was any activity near it.
3. It had to be cloud connected and pull a software update to disable the feature.
Alas the last issue due to the sheer number of unaccounted for devices, ended up in a full recall of 440,000 devices. This "seemed like a good idea at the time" is how most of the IoT market operates. The inevitable reply I'm going to see which I will mention here was "that was an old version" and "it doesn't work like that now". That's not an excuse for fucking it up to start with.
And then there's the problem of what happens if your smoke alarm is silently bricked or broken by an update? Google, Nest's parent company, hosed 25 handsets we had via automatic updates. And what happens when Google eventually decide to can it. It's no longer smart. It's an attack surface and a risk. IoT devices are already posing a significant risk to security.
Secondly, price does not necessarily reflect fit for purpose. A £3 device may be considerably better than a £100 device. That's what your aforementioned EN 14604 standard is about. Both the £3 and the £100 device should pass functional testing to conform to that standard.
What is important for smoke detection really based on my trite evaluation?
Well the risk appears to be defined by two classes:
1. It didn't go off.
2. It packed in.
Thus evaluating that into characteristics to operate smoke alarms:
1. Diversity. One class of device from one vendor is clearly untrustworthy. That is obvious by the sensor disparity between optical and ionisation sensors and the functionality that is provided. Deploy multiple optical and ionisation detectors from different vendors (I have 2 of each in my house for ref now).
2. Regular testing. Weekly basic "is the device still working". Quarterly "does this thing actually still detect smoke". This builds a failure model for the devices and pre-empts failures.
3. The MTBF appears to be pretty shit despite advertisements otherwise. Replace it every year. That pre-empts failures. This is cost effective.
That's slightly more sound engineering thinking than throwing all your eggs in one IoT basket.