Not much time today, but I'll chime in to support the Superkit notion. We can't know which SNPs actually end up being used in comparisons after GEDmatch has "slimmed" an upload, but what we can do is run the File Diagnostic Utility and look at those stats, particularly the number of total matches (these aren't individual people, of course; think of them as segments) shown in the database for that particular kit.
Quick rundown of my own kits' (same DNA) numbers, as of early last month:
Superkit extracted from whole genome sequencing:
- Number of original SNPs: 2,080,567
- Usable SNPs: 1,377,182
- Usable SNPs (slim): 1,019,459
- Total matches for kit: 56,802
- Matches to slimmed SNP count: 5.57%
23andMe v5:
- Number of original SNPs: 635,966
- Usable SNPs: 496,179
- Usable SNPs (slim): 397,773
- Total matches for kit: 161,680
- Matches to slimmed SNP count: 40.65%
AncestryDNA v1:
- Number of original SNPs: 636,709
- Usable SNPs: 636,709
- Usable SNPs (slim): 441,045
- Total matches for kit: 65,845
- Matches to slimmed SNP count: 14.93%
FTDNA v3:
- Number of original SNPs: 627,817
- Usable SNPs: 487,649
- Usable SNPs(slim): 382,550
- Total matches for kit: 109,359
- Matches to slimmed SNP count: 28.59%
Living DNA v2 (this kit was simulated by using the template of actual SNPs tested by the Living DNA Affymetrix chip)
- Number of original SNPs: 698,531
- Usable SNPs: 563,991
- Usable SNPs (slim): 431,418
- Total matches for kit: 141,648
- Matches to slimmed SNP count: 32.83%
MyHeritage v2:
- Number of original SNPs: 610,128
- Usable SNPs: 494,251
- Usable SNPs (slim): 390,654
- Total matches for kit: 117,136
- Matches to slimmed SNP count: 29.98%
So the actual matching results are sort of all over the place. But clearly the kit that started out with over 2 million reported SNPs has the lowest total match count of any...by a huge margin if we look at matches by SNP volume. Deductively, the implication is that as we increase the number of unique SNPs on file, the more likely that what would otherwise be evaluated as contiguous segments are, in fact, not. That there is enough interruption in the assumed contiguity of the segment that it no longer evaluates as a match.
And, yes; I think it should be a factor of concern among those folks not using a Superkit (and I really wish we could take the GEDmatch SNP catalog up to around 20-25 million...I've got 'em, but I can't upload 'em). More SNPs nets you fewer matches, but the resulting matches are likelier more accurate because they're less affected by algorithmic mismatch assumptions. This should be another of many cautionary tales about using very small singleton segments as genealogical evidence. IMHO.
Edited: If you're using a decent calculator it probably isn't wrong...but the person punching the keys may be.