In order to validate my choice of the development version of Stockfish to meet with Komodo 10 in Rapidroid, i've played a match between 2016.16.26 and 2016.07.16.
The build of Jun,26th is taken form CEPA3.7 while the other one, most recent one as of today is compiled by Peter Österlund.
Some testers reported than 160626 is stronger and criticized my decision, though there are exceptions like Nathanael Russell's tests which favoured the latest build. I admit that all reported results are valid in the conditions adopted. However, i've choosen to increase the timing a little bit more and also the number of games to play to dig the truth.
Using below conditions (my basic quick test setting) 2016.07.16 had no problem to overcome 2016.06.26:
Processor: RK3188 quad core 1.4 Ghz downclocked to 1.0 Ghz (Throttle free!)
Hash: 256MB (although may not be mandatory, makes no harm)
Openings: TCEC-7 all 306 positions for 612 total number of games
This setup theoritically delivers about 440 Knps x (5+1) sec/move = 2.64 Mn average per move, while the standard Rapidroid setup delivers 800 Knps x (15+1) sec/move = 12 Mn average per move. There's a loss of quality sacrificed for speed here but i guess it's tolerable.
Regarding the results, here's the final screenshot by Chess for Android:
And a quick summary reported by Elostat:
Stockfish 2016-07-16 : 612 (+ 99,=453,- 60), 53.2 %Stockfish 160626 : 612 (+ 60,=453,- 99), 46.8 %
The gap calculated is 22 ELO vs an error margin of 14 ELO. I didn't attempt to calculate LOS but the experience tells the current result is enough to conclude. The latest build is OK.
The games compressed with 7z can be downloaded: HERE