Interpreting the output

[dgtlrift]

I understand the tests run for FIPS, but what is considered an acceptable level of failures? I've seen some StackOverflow comments (which I can't find now) suggesting that below 1% is acceptable. I've also seen where FIPS 140-2 Change Notice 2 of 2002 December has indicated that PRNGs can game the test and TRNGs can show false issues. If there are other better tests, great, but using this as a qualitative test - how do we interpret the output of the aggregated failures or the individual tests error count.

For example, in testing on CygWin /dev/random there's an output of

rngtest: bits received from input: 60000000032
rngtest: FIPS 140-2 successes: 2997621
rngtest: FIPS 140-2 failures: 2379
rngtest: FIPS 140-2(2001-10-10) Monobit: 327
rngtest: FIPS 140-2(2001-10-10) Poker: 306
rngtest: FIPS 140-2(2001-10-10) Runs: 865
rngtest: FIPS 140-2(2001-10-10) Long run: 896
rngtest: FIPS 140-2(2001-10-10) Continuous run: 0
rngtest: input channel speed: (min=763.864; avg=66458.326; max=85663.377)Kibits/s
rngtest: FIPS tests speed: (min=587.866; avg=126697.729; max=164128.151)Kibits/s
rngtest: Program run time: 1345118200 microseconds

which if I calculate out Sigma values:

                 Control Level | 4.32 Sigma
                         Yield | 99.9984%
                   Defects (%) | 0.0016%
                          DPMO | 15.860
Rolling Throughput Yield (RTY) | 99.99%
              Defect Units (%) | 0.01%
                           DPM | 79.30

which are not very acceptable for manufacturing, but I'm not sure if that's the correct way to qualify.

The ask is what is an acceptable score?