Hadn't seen this mentioned but perhaps I missed it, more plumbing hardware than electronic, but perhaps useful for others.
Having tried a variety of plates and heat-sink compounds and tie-wraps to hold these little chips in position on small sections of metal pipe (my rads are fed with fused plastic pipe), I came up with the solution of using pipe clips, the cheap plastic variety from DIY stores and hardware merchants.
The first problem hit was with attaching plated sensor the the boiler clamp nut, it was off by 10c and no way could I get a more accurate reading, presumably angle? Finally found a pipe clip which got hacked in half to fit in the tiny gap in the photo... Some hacking with a Dremel, some soldering to a pin connector, some fiddling around with epoxy and voici....

The gap
The tie wrap in the top of the photo is the original plate mounted version which was "off"..

Having scored on that front with more reliable readings I looked at the roving radiator test kit and realised the same solution might apply so epoxied the chip on the end of some telephone line in a 22mm clamp and it works a treat. Cable and chip fit neatly into the screw-hole, flood with epoxy and done. The pipe it fits is 21.5mm so tighter than a duck's rear end...

As said, a tight fit to a twinwall radiator feed pipe, and a complete swine to get back off even with pliers, but no broken wires or oopsies to contend with, and so much faster to fit.
For those looking to measure metal pipe temperatures, possibly an unusual if fiddly solution, but no need of tie-wraps in the latter case...

Not so much a MySensors build as an example of how even the most basic information can inform changes for the better, in this case space heating.

The system here is fairly basic, an array of DS18B20s, some ultrasonic tank probes and a gas reed sensor, temperature is updated every 5 minutes, the gas updates every 0.05m3...
With winters here down to -20, the first priority last year was insulation, and even though a modern house, the gas bills essentially halved over the year, effectively funding not only the insulation, but replacement axial radiator valves and thermostat heads (Heimeier) to replace the typical arrangement of unknown origin, with spare... But now the MySensors impact..
This autumn's attention turned to the central heating unit, a modern combi unit of good manufacture, installed by a 'certified' heating engineer, but aside what little I knew about condensation boilers and the steep learning curve that followed, I was bemused by the return from the radiator loop almost burning my finger within 10 minutes of the system being fired up. This did not make sense for what I understood of a condensing boilers, which compelled a look inside for the first time, the manual and some googling.

The boiler is a 25kW combi with minimum output 7.6kW, the radiators account for ca 13kW at Delta 60 set for 15c drop (previously set ca 20c drop), settings since day one were 65c and the pump was set at max output of 3, last year's -20 resulted in 13.5m3/day gas consumed, not crazy by historical records, but hmmm.
So now comes tinkering with data from MySensors via Domoticz to inform...
Currently the boiler is set at 55c, the pump is on Low (40 v 84w), but the results are surprising - Slower rising temperature when ON, 42 minutes v 25, but gas use dropped from 0.75 to 0.6m3, but here's the kicker from that longer heating time, not only less energy used per cycle, but longer and thereby fewer cycles per day. Current evaluations are between 15 and 20% savings, so thank you to all the MySensors community and contributors..

Surprised I'm not bald will all the hair pulling and failed attempts to get there but after a year it is finally working...
The task was simple enough, to measure fluid depth in two tanks, one a sewage holding tank, the other a bulk water tank. Just figuring out which method worked with which board was an exercise in frustration, but the JSN-SR04-2.0 turned out the only one of the two I could get to reliably work.

Despite the industrial look, these waterproof light switch enclosures proved ideal, as the batteries in holders fitted beneath the clear plastic. A little hacking out of the switch 'fingers' with a Dremel gives plenty of space in the base area.
Battery holders are held in plastic conduit which are hot glued to the back of the box poking through the front section aperture.

The top case is a 3v3 WhisperNode with onboard RTC and RFM69, and tucked under the spaghetti in the top left back corner is a level converter. The RTC wakens the Node every hour to initiate readings, and despite a few hiccups seems to be reliable. The WhisperNode is allegedly capable of draining the 2 AA cells down to 0.9v, time will tell...

The lower box holds the 5v ProMini, a small latching relay, and the ultrasonic board, the button type ultrasonic (like parking sensor) is mounted on the end of a 3/4" collar to 1/2" galvanised steel tube drilled centrally through the roof of the 1.34m x 1.67m tank area.

Connection between the cases is via Cat5e in 16mm conduit, 2 cores for I2C, 3 cores for relay control and 3.3v power, 2 cores for 5v and ground to the level converter.

Despite worries over echoes in the resulting space, the tank was emptied last night, and the reading is coming in with the occasional glitch at 1620mm depth from the head, which I believe has a cone angle of 45 degrees.
The Node control of a Relay by On/Off pins to Mosfets may seem wasteful, but I had them anyway. The 5ms on/off of this small latching signal relay boots up the ProMini and seems reliable.
Still to incorporate a WDT/Reset on the 5v as it periodically hangs, and a time limiter on the Node to complete the task, but glad it is all all working semi-reliably, even if some of the programming side I still don't understand...
The 5v tests for 2 consecutive identical readings, the 3.3v tests that it fits within acceptable range, and if not calls for a further reading.

Once delivery of a second identical board from China is done, the water tank should be a lot easier. This one is the more important, as ran out of water previously due to corrosion of the level probes in the tank (3 core cable and short bolts) which stopped the borehole pump topping up the tank, and the hidrofor hit it's cut-off electrode. The level monitoring will give early warning to go investigate long before the bulk tank is 'empty'...

Postscript 12/7/18 - Water tank worked flawlessly from first deployment, but led to concerns over dropping battery voltage.. The problem appeared to lie with the Master sinking power through the I2C lines, but a routine to null the SDA/SCL lines prior to sleeping seems to have halted the decay.
One of the unexpected results of the Sewage Tank deployment was detecting a dramatic rise in level during a heavy storm (85mm in 24 hours) initially thought to be ground water intrusion.
Excavation to the incoming pipe connection to the tank found a round pipe in a square hole (not exactly an unexpected building practice here, now packed out and sealed.
Losing 30% of storage capacity would otherwise never have been detected, so it the deployment has already paid for itself handsomely...

@pepson This reminds of that famous quotation "Any ship can be a Minesweeper..."
"Once...."

@neverdie Would this work ?

@sundberg84 Yeah, this brings back memories, particularly of quite a few hangovers.
I used 3 and four wide fireclay pipes for bottle storage, they were used for ducts back in the day before plastics took over, cable went optic fibre. I noticed recently here they are back in stock in stores as....yep...wine storage racks.... but wacko pricing relatively...
The beauty of the fireclay was it's slow temperature and humidity change, and this formed the bulk of the thermal mass in the cellar I made below the floor.
I used a 150mm glazed ceramic drainage pipe dug in under the garden as a loop (rope caulked joints - anybody remember them), from memory down about 1.5m, both avoiding frosts and baking sun, don't think the temperature varied over a degree all year round, the ground acting as a massive heatsink which maintained a constant temperature all year round.
Only when the hatch was opened did the temperature jump, a small fan kicked in when the hatch was closed and ran for 30 minutes, a second contact switched on the lights and shut them off, long before LEDs were so prevalent, old reliable (until you were depending on them) incandescent bulbs.
That was it really.
Biggest problem I found was humidity in the early days, probably the fresh construction...

@tbowmo Hah, you were lucky, I finally upgraded from the monochrome to colour...

@sundberg84 Easy, one is disconnected... At 90 degrees?

Finally the round-tuit UPS got built after the last power cut clobbered the Controller/Gateway system, lesson learned...
Meanwell AD-55A, 7.2Ah Acid gel battery, two usb 5v buck converters, a spare socket for raw volts, and a 15 euro IP66 box from the local shops.. Some drilling and filing to the lid, couple of brackets, spare bolts, banding, soldering and hot glue...
A bit bulky, but disappears in a void under the stairs, two tiny drill holes let the buck converter leds shine threw...
No monitoring as yet, but sailed through a power cut this morning and the Pi didn't skip a beat.. First up is the Pi's RTC then can put the cover back on the Controller...

First of the winter projects done, just the programming to finish and replacement temperature chips to get (broke the legs off both DS18B20s through clumsiness).
With mains/battery backup and RTC, it will record boiler start/stop and run times as well as feed/return temperatures, and as an aside the inevitable power cuts which plague this part of the world.
The MCU plugs into a socketed backplane hot-glued to the back of the box should removal prove necessary, but quite pleased it is sturdy and all fitted into a slim 25mm deep standard (cheap) box.

That's good news.

Moisture would not necessarily have damaged it permanently, it's more a case of humidity can play havoc with tiny electronics and wiring when it condenses.

The advantage of a weatherproof box in my environment is obvious, but it has advantage elsewhere as it's a controlled and sealed environment inside - If closed up dry it stays dry.

@Oumuamua Sorry but I've no idea - I didn't update the MySensors side of the system once everything on the Gateway and Controller were working perfectly 5 years ago - Only Domoticz updates on the Pi Controller have interrupted it, the arduino gateway and disc drive have run 24/7 through multiple power cuts on the UPS.

It does seems unlikely the driver is at fault given so many others updated and none have reported such a bug.

IIRC the new driver has to be installed on ALL devices using the rfm, so I'll update the Gateway in summer when swapping out the UPS battery and drive.

@Oumuamua I strongly suggest your problem is moisture related, been there done that.

My external nodes are all in weatherproof boxes (Gewiss surface mounted modular light switch box with the transparent hinged front cover - I dremel off the switch tangs) with a roasted clay desiccant pack thrown in to cope with any moisture which may creep in on the cables over winter, then the front is latched shut.

No oxidisation is possible so long as the box is weatherproof (sealed), as any internal moisture is absorbed by the desiccant (a reactivated silica gel pack will work) when you close it up.

My gas meter node was like that, running for 4 winters down to -20 and never skipped a beat - It was like new when I changed the batteries, and ultimately took it apart, zero corrosion.

It's an awfully convoluted way of doing things but suspect your error lies in misunderstanding calculations mixing ints and floats and something screwy with your resistor bridge - It is always worth checking the bridge with a multimeter and raw voltage to verify the expectation.
If you have 3v at the top of your voltage divider, you should be getting 0.9593v applied on the ADC pin which will be read as 892 against the 1.1v internal reference.

The easiest way I found was to define a multiplier needed to derive the raw voltage, in this case (((R1+R2)x1.1)/R2) - For your resistor arrangement the max you can read is 3.4404v before exceeding the 1.1v internal reference.
To reverse the ADC reading of 892 to raw voltage is (892x3.4404)/1023 = 2.9984 v.

Perhaps this is a simpler way to do it

float MULTIPLIER= (((1000+470)*1.1)/470);//Resistor bridge values in k and Vref
int sensorValue = analogRead(BATTERY_SENSE_PIN);
float batteryV = (sensorValue * MULTIPLIER)/1023.0; //Note the trailing decimal point on the 1023 for calculations involving floats

@Doubletop Thanks for that.
I haven't range limited the readings on this version (ie ignore bad results) but do check for two consecutive readings, thus far no bogus readings recorded.

In my own case, the 3v node had no spare pins, so had to expand anyway to avoid using to a second dedicated node.

Having read on the 5v returning stable results, went with a second pro-mini at 5v controlling the v3.0 (also enabling further expansion) running off a PSU, talking I2C to the Node.

My principal problem was winter condensation on the transducer head due to the pipe used, aside that the v2.0 ran two years before finally dying.
This time it's on plastic pipe and the v3.0 is socketed for easy swap should it go faulty, fingers crossed for a trouble free winter.

@Doubletop Interesting, as this made me check my own program for the V3.0, which finds a typo - I'd set the trigger at 100 instead of 10, but it never failed to get two consecutive readings.
Can you clarify your experiment on this was at 3v or 5v ?

I've gone back to try ultrasonics again in the underground tank - The transducer is now installed in a plastic pipe which should solve the condensation issues from the previous metal pipe below -10 blinding the transducer.
The 5v requirement of the V3.0 is met by a pro-mini talking over I2C to the 3v node unless it's running on batteries.

One facet of the v3.0 I fell over in setting this up was it's inaccuracy at short range - Measured level 298mm, read level 275mm, from 1m down the error vanished - both devices bought responded identically.
The solution here was to scale the pulseIn for top and bottom for the known volume, and thus far, working reliably.

Having decided to have another go at ultrasonics for level in the underground bulk water tank (this time in plastic tube not steel - Condensation issues below -5c), I ordered up a pair of sensors, which on arrival turned out to be version 3.0.

Looking up the differences, they state 5v so no longer specced as 3v, trigger exceeding 10 microseconds, and the blind spot had increased from 20cm to 23 over V2.0.
Curious whether 5v would restore readings on the original V2.0, was surprised to see it responding accurately again.

Once installed in plastic in the tank, an error was evident but only for near objects.
eg - At 1300mm physical depth to surface the reading near enough correlated, but with the tank full and surface 295mm below the transducer it reads 273, as the level falls to 330 it reads 309 etc., etc..
I bench tested the second set and got the exact same error on short range.

A correction factor will solve my own use case, or perhaps even refit a V2.0, but for any looking to deploy these new V3.0 types it is worth checking short range accuracy.

@Jens-Jensen I've successfully used a 5psi pressure sensor to get raw water tank readings. down to about 1.3m.

@JonAndersen If you find a line of sight either direct or via a repeater site, Yagi or Quad might do the trick....

@Nca78 Agreed.