Sensorcat
You know, you made a trivial mistake then. The ice is supposed to be moist/moistened and not swim in "cold water". Water has the highest density at 4 degrees Celsius. This causes ice to "dribble to the bottom of the vessel", while ice has a lower density at approximately 0 degrees Celsius, so it "floats". This is one of the reasons why fish survive in lakes in winter. The ice in the thermos must be constantly "drained" of excess water - this is NECESSARY in such an implementation. The sensor is not meant to be "pushed to the very bottom of the vessel". The sensor is to be in the "middle part of the snow block" - pressed.
There is a very nice pdf explaining how to do this on the New Zeland National Laboratory website.
Although you should not think that my own experiments involved a single icecube floating on water, I do know now, thanks to the valuable resources posted by mzzj, that the ice bath has to meet harder requirements with little tolerance (for instance, having small pieces of ice). That's why I prefer to see this as difficult, not as easy, which was the first point I wanted to make in this thread. Murphy is my witness that it is preferable to see tasks as difficult rather than easy, if you want to achieve stable and reliable results. Note that I never claimed that I know how to prepare a good ice bath, so I did not make 'a trivial mistake.'
And this leads to my second point: If you need reliablity, you need a process. A process makes all relevant factors explicit, not implicit, and avoids the human factor. Standards that have to do with assurance of quality or safety therefore involve auditing, with the auditor always asking one question, in endless variations: 'How do you make sure that ...?' And if you reply 'I know that.' - 'It's easy.' - 'Never has been a problem.' you should not get your certificate (provided that the auditor does his job properly), because that's not a process. And having a process is certainly not limited to technology with the highest requirements! Whenever not meeting requirements (tight or loose) has dire consequences, you need a process. Who does not understand this should not think of designing something serious, because if you do anyway, things like this happen:
https://www.eevblog.com/forum/chat/baltimore-francis-scott-key-bridge-collapse/