Haven't been by this thread for awhile—and my apologies for that.
AFS: I can answer your questions. The Predictive Thermometer's sensor tube is made from stainless steel. The rear charging contact on the handle is stainless steel. The handle itself is aluminum oxide ceramic. The internal seals are a fairly exotic high temperature silicone. There is a screw in the handle that is nickel-platted copper, which is directly connect to a precision thermistor in the handle—this screw acts as a heat pipe to measure the ambient temperature around the boundary layer of the food. And, yes, it's fine on an induction stove top.
Regarding the predictions, based on the data customers are sending us from our beta-testing program, they seem to generally working pretty well for higher temperature, or relatively quick cooks (pan-roasting). At 30% of the way to the target temperature, the average error across all cooks in our database is just over 11%, but 50% of the way to done the error averages a around 8%, and by the time we're at 90% of the way to done the average error is usually less than 2% and converges nicely to rarely being off by more than a few seconds.
The caveats here are that if you flip the food, change the temperature, move the coals around, etc this changes the speed of cooking. The algorithm does a pretty good job of handling these curve-balls and will just recalculating. That's probably about the best we can hope for since we can't actually anticipate what you're going to do on the fly.
There is a known bug that is tripping up the algorithm when the heating rate is particularly slow, such as sous vide and sometime low-temp smoking or reverse sears. This should get a fix in one of the upcoming firmware updates (probably the mid-February update).
There is also a bug with the instant read filter on some thermometers causing a bit too much oscillating, or sometimes a bit of an undershoot. This seems to be down to some very small variations with the internal sensor position during manufacturing. We're working on retuning this right now using a much bigger sample size of thermometers.
Finally, resting predictions are still a ways off. We've got enough data to have a pretty good idea on how we're going to solve it, but we're going to need a much bigger data set of cooks to do it well. To get to that, we need to finish adding cloud connectivity to our apps, so that we can make it easy for our customers to "donate" their data anonymously to help train the algorithm. I'm hopeful that we'll get there before summertime. But our team is pretty small and there is a lot to do.
A few folks have commented on the lack of formal reviews. I think that this is mostly because we have not set out any samples to reviewers yet (yes, we're getting asked a lot). Reviews have a pretty big impact on a new product like this, and I made the decision that I would prefer to have a few months after scrambling to ship in December to work out the inevitable bugs. So, right now, what we've been focused on is reading as much feedback as our customers give us, using this to prioritize what we're working on improving, and generally stabilizing production and our supply chain.
Feel free to ask questions or make some comments in this thread. I'll try to keep an eye on it. You can also always email hello@combustion.inc. I read nearly all of those emails myself and make sure that feedback or concerns gets responded to.
--Chris Young
