The air conditions around the sensor are clearly significant. I'm not sure that "a bit of air flow" is the best recipe for getting the most accurate output.
Alas I do not have a Stephenson Screen to put them in.
Instead I gathered the four sensors indoors, close together but not touching, so the airflow over each should be similar.
The 4 ESP8266s are all powered by a single 4 port USB power supply, though the USB cables are not matched.
I paid no attention to the orientation. It's not improbable that a board with the sensor facing down might register slightly warmer than one facing up.
The gross error in one of them is clearly related to a high resistance solder joint.
changing your ESP8266 code so that the BME280 will be in deep sleep and only will be waked up for a measurement
The ESPs run Tasmota. My Node-red control flow can turn on deep sleep and I do intend to compare the results with that. But I suspect that the BME, wired to VCC and GND will still be powered up when the ESP is asleep. I'll have to use a GPIO pin to power the sensor.
Excellent find! I have been soldering microelectronics for years, even changing simple SMD devices with a soldering iron, and still get the odd bad joint. I now check with an eyepiece to see the joints are bright. It was far easier, if a little more toxic, with the free flowing lead based solders.
My ESP8266 all run on Tasmota as well. Found it easy to install and configure. Also good if you might have comms problems for MQTT/NR flow errors as you can just log straight in to the sensor unit and see what it reports.
Good luck with new devices should you decide to replace.
Would you be so kind and share the results. I would appreciate this.
I am using BME280s as well and while they mostly work well, I did not find a good solution to put them in some sort of housing without them measuring 1-3 Ā°C more but inconsistent (meaning I cannot simply subtract a fixed value).
I'm not ignoring your request for more info around deep sleep, but it is proving difficult.
I can tell if a device is sleeping by the Last Will & Testament MQTT, though there is a lag reporting offline (equal to the Keep-alive time - 30 seconds?)
By this measure, ESP #3 seems to wake, report and sleep again over just a few seconds
But #1,2 & 4 wake up, report more than once (I have experimentally set teleperiod to 60s) and stay awake for well over a minute.