I'm following the discussions about 4/32 devices, and seeing the problems with them and 18.06, it sounds reasonable to me what hannu writes.
We have 479 devices in the ToH (less some outliers) which have either less than 8MB flash or less than 64MB RAM.
Is that recommendation valid for all of the low flash/RAM devices listed (18.06 = approx. 320 devices)? If yes, I could update the dataentries accordingly, in order to steer the users in the right direction.
The advice is questionable and is not 100% true always. E.g. plain wired router use without LuCI could still be quite viable with a 4/32 device. (there are also people arguing that it works ok for them.)
But using LuCI (actually the uhttpd server) increases the memory exhaustion risk, and e.g. opkg package installation has clearly increased the possibility of crashes due to memory spikes, as has been discussed earlier.
I would maybe go with a clear warning in the 4/32 devices that 17.01.X may be more suitable than 18.06.x or newer. That target group of the advice is casual users, who do not compile their own images.
4 MB flash is actually the easier limitation. 32 MB RAM is the hard limiter, as the kernel size and memory consumption has grown.
That's going to be a "challenging" recommendation going forward, as 17.01, as I understand it, will no longer be maintained in 2019.
Perhaps the "432" warning needs to be updated to an 8/128 warning and that the "ideal for OpenWrt" TOH filter be updated to at least 16/128
It's all about your point of view and how you want to phrase (sugarcoat) it.
The majority of 4/32 devices barely boots and even if they do performance is less than stellar except for (more or less) late Atheros based devices which works in some cases.
opkg and LuCI do run unreliably and will cause memory exhaustion, a barebone install will run but you'll have very little memory free which means limited amount of clients and connections. As most devices are quite old and/or aimed for the budget market the SoC is usually quite slow when means you can easily overload the CPU which will also cause unreliable behaviour (reboots, etc).
You can still use such devices in specific applications but I wouldn't recommend such as device in general as you will most likely trigger something mentioned above .
In general I think @jeff is on the right track, I would probably mention that armv7 (or better) is a preferably to performance and overall "support" as MIPS SoCs are slowly being deprecated.
17.01.x for new installations can't be recommended anymore, as maintenance for the 17.01 branch will be dropped with the release of 19.01 (so roughly ~2 months and a final maintenance release left). If the existing devices can't cope with that, the options are either to strip down the images (e.g. opkg and luci are natural choices for starting) or to upgrade the hardware.
Edit: Just to clarify, 17.01.x is still maintained right now - and if you're on 17.01.5, upgrading to 17.01.6 would be fine - but now is the time to plan the upgrade/ migration to 18.06.x, not to freshly install new devices with 17.01.x.
I definitely agree with pointing those not steeped in the knowledge of OpenWrt-capable devices toward ARM devices when available at a reasonable price over MIPS devices.
As vehement as I am in my opinions around insufficient flash and RAM and how much dev and support time I believe they waste, I personally don't see ARM-based devices yet generally available in either the "travel router" (small, often USB-cable powered) or "affordable" (under US$50, for a very rough definition of "affordable") classes.
As soon as there are ARM-based devices from major manufacturers commonly available through most retail channels in the US$20-40 range, for example, my opinion would likely shift to the "don't recommend MIPS" position.