[CQ-Contest] Summarizing the Skimmer Accuracy Thread

[VR2BrettGraham]

N4ZR continued:

> Bret, send me your data in a form I can look at, and we can have a 
> useful discussion.  I can imagine several scenarios for what you 
> describe - harmonics, dueling CQs, keyclicks, and so on.
>
> I'd love to see your data.  Your assertion about 1-2 percent errors is 
> totally out of the ball park. The RBN made 3.7 million spots during 
> ARRL CW - are you really saying there were 37,000-74,000 erroneous 
> spots?  Show me!

I have described the queries, you can download the data yourself & the 
results will be the same.  On the day with the two mechanisms as 
mentioned, was already a tad over 0.3%.

Because I have worked with the data & seen the results as I have worked 
on queries for the other mechanisms, what I said is I believe it could 
be pushing 1% or more in total.  Especially after adding busted calls & 
busted calls on busted frequencies (if I can ever figure out a query for 
that one).  Again, that is an expression of my gut feeling from actually 
working with the data, from a number of days across 2012.

Since it's only certain skimmers sending wrong-freq spots, you can 
imagine all you want Pete, it won't change the fact that the receivers 
have spurious responses & if signals appear there, they will be spotted 
on a frequency or even band that is outright WRONG.  I've seen these 
spurious responses, having used the receivers a lot myself, learned 
what's behind the responses & trained myself to recognize when they 
happen.  There's apparently also one mechanism that is a bug, results in 
the agile version of one of the popular QSD receivers spotting freqs 
with correct kc but wrong Mc digits.

GL overcoming what to me seems like confirmation bias, ex-VR2BG/p.

> On 2/25/2013 6:33 PM, VR2BrettGraham wrote:
>> Here is some more data.  On 2012-09-21, there were 151 RBN spots of 
>> the same call but on another frequency >3 kc away on the same band at 
>> exactly the same time (to the second) as another skimmer spot.  For 
>> this query, I can avoid "dupes", so opening that up to spots of the 
>> same call >3 kc apart on the same band within about 9 seconds of each 
>> other finds 369.
>>
>> That is just one mechanism.  Spots of same call on different band on 
>> the very same second look to have been 140.  That's already 0.3% of 
>> RBN spots that day & that will exclude a LOT of wrong-band spots.  
>> Add busted call spots & who-knows-how-many busted call wrong-band or 
>> wrong-freq spots & then scale all this up by some factor as activity 
>> is much higher on weekends (I'm working with weekday data here, as it 
>> makes it easier to find wrong-band spots as no multi-multis on then) 
>> - I believe RBN could be blowing 1-2% of what it spots, perhaps more.
>