QRSS3 challenge
The challenge

Participants had to "decode" 25 QRSS3 screenshots at SNR levels from -24 to -28dB (@2.5kHz bandwidth). There were 5 different screenshots for each SNR level. The number of characters displayed in each screenshot varied (whatever fitted in a 3 minute time frame). The goal of this challenge was to determine down to what SNR level a QRSS3 signal could be decoded.
27 entries were received. Thanks to all who participated.

The results

The SNR levels -24dB and -25dB seemed rather easy for all participants, with an overall score of resp. 96% and 94%. The real challenge began at -26dB where the score dropped to 86% and then further down to 70% (-27dB) and 46% at -28dB.
When I produced the QRSS3 screenshots I also produced Opera and WSPR audio files at the same SNR levels (based on the peak power, as this is the most relevant with high efficiency class D or E PA's used on LF/MF) and fed these into the appropriate software. The results can be seen here:

In the graph the number of correct decoded characters can be seen, compared to Opera2 and WSPR. Apart from the overall QRSS3 results also the average result of the 10 best entries is shown. No surprise that most of the top 10 results were submitted by well known LF operators.
As expected QRSS3 - with it's human factor - has a gradual drop, in contradiction to the steep drop from 100% copy to no copy for both Opera and WSPR. Due to this it is not easy to compare QRSS with Opera and/or WSPR.
Nevertheless I feel that the QRSS3 treshhold level is at about -25..-26dB, versus -23dB for Opera2 and -29dB for WSPR. So the QRSS3 performance is somewhere inbetween Opera2 and WSPR
Or looking at it another way: Opera8, QRSS10 (or DFCW10) and WSPR should have a more or less equal performance.
Some notes:

The screenshots