Two comments. First, running computational phasing with one parent tested can significantly help research matches and align them to the correct maternal or paternal lines. However, we have no good data that indicate computational phasing can or should eliminate any need for triangulation, or that very small segments can be taken at face value even with both parents phased.
Perhaps the best numbers we have for this come from a compilation by Tim Janzen of data collected by John Walden. The information looked at segment "survivability" from phasing, meaning whether or not phasing showed a segment to be invalid because it did not match the parents.
- 9cM segments: proved false 15% of the time when one parent was phased; with full trio phasing they proved false 20% of the time.
- 8cM segments: proved false 22% of the time when one parent was phased; with full trio phasing they proved false 38% of the time.
- 7cM segments: proved false 37% of the time when one parent was phased; with full trio phasing they proved false 58% of the time.
- 6cM segments: proved false 58% of the time when one parent was phased; with full trio phasing they proved false 74% of the time.
- 5cM segments: proved false 71% of the time when one parent was phased; with full trio phasing they proved false 86% of the time.
Second, with segment size, we again come to the fact that centiMorgans are mathematical estimates of genetic relationship based upon assumptions of gametogenic crossover points along each chromosome. The cM isn't a physical measurement, and what's reported is never exact...in part because of the nature of the linear equations involved; in part because of the objective of indicating a 0.01 probability of crossover at any given section of any chromosome; in part because the tested SNPs amount to only about 0.02% of the base pairs in the genome, so the physical segment start and end points are determined by the tested SNPs, not by the base pairs that might, in fact, match or not match; in part because some level of SNP mismatch is typically allowed by the reporting entity (e.g., the 1:300 error rate allowed by 23andMe or the the SNP mismatch bunching limit allowed at GEDmatch), and in SNP-poor chromosomal regions the possibility exists that there may be many thousands of base pairs within a reported segment that don't actually match; and in part because the female and male genomes do not look at all the same when it comes to centiMorgan computation.
Since females experience crossover about 40% more frequently than males during gametogenesis, their genome maps in terms of centiMorgans look markedly different. All we see reported from the testing companies is a sex-averaged value; that's really all they can tell us. But in the realm of very small segments, that breaks down to a point where a sex-averaged evaluation is basically useless. Here's one real-world example that I happen to have at hand:
Chromosome 9 from bp 32,216,761 to 53,574,578. The sex-averaged value is 7.56cM. On the female genome, it works out to 13.6cM. But on the male genome, that segment is only 1.52cM. I doubt anyone feels we should be working with 1.5cM segments.
I don't believe this is simply a matter of revising text on a Help page.