What does 10 percent accuracy mean? Doesn't that mean it's usually completely wrong?
http://spectrum.ieee.org/geek-life/hands-on/ looks like a great site, I can't wait to build some of these things!
First of all as typical the physicists use a SI unit that joe average USA doesn't use. The physicists like Bq which is basically odds of decay of a single particle. So 1 Bq/liter/second is one decay per second in a one liter jar. Not much. A legacy curie Ci still used by the EPA is something like about how many decays a chunk of radium gives off in a second or the length of the kings foot or some ridiculous. Anyway 1 Bq = 27 pico curies or pCi. So the EPA standard of 4 pCi/L translated into modern terms is 0.148 Bq/L
The dude provides his survey meter in Bq/cubic meter and we're talking Bq/liter so div 1000 because 1000 liters in a cubic meter. So the survey meter is reading 0.159 Bq/L.
I am a bit confused why the dudes test area is reading about 10% over the EPA standard for radon contamination. I guess he needs a source that should just barely be positive. In that way, it makes sense to run the tests in that location.
As a side note I'm guessing his homemade gadget has a volume of a tenth liter. That would imply about 0.0148 Bq/sec activity in his homemade gadget. He claims a detection level of 5.2 Bq/hour (edit: 5.2 hits/hour) so 0.0148 * 60 * 60 = thirty something actual Bq activity (edit: hits) happened in his detector of which he captured 5.2 on average. That means his homemade gadget is about 13% likely to detect any individual decay that happened inside of it. That's actually pretty good!
I can't be bothered to continue with a statistical analysis but the 10% claim sounds about reasonable because of what boils down to quantization noise. So your threshold is about 5 pings per hour. He gathers 20 hours of data thats about 100 pings total. The longer you gather and average data the more sig figs you get to accumulate with a sqrt of the number of samples (way too long story). So he's saying with 100 pings his data reduction indicates with X% certainty the actual number is within 10% of the legacy 4 pCi/L
Someone else can pick up the statistics torch and run with it from here.
The TLDR is given infrequent discrete events you need a lot of them to have a statistically relevant discovery if you rely on averaging them together to gain a sig fig in your results.