1% of the human genome (3billion bp) is still 30M bases, which could easily contain 1 or 2 genes which you may have genotyped, depending on the resolution of your genotyping chip (usually 10M markers spread out over the entire genome, and usually clustered around variants/genes of interest).
So yeah, even 5 generations down these differences could be detected at the variant level. Even at 10, you could still _theoretically_ find differences.
I stress theoretically, because the chance of you having all the individuals from a pedigree genotyped at the same time using the same tech (and thus the same subset of common genotypes to compare) is slim to none.
I'd say 5 generations is a rough rule of thumb where the signal is not that much higher than the noise.
1. Circumstances where both mtDNA and Y chromosome are shared between biological father and pedigreed father. I think the Biblical term is Levirate marriage.
2. Circumstances without Y where maternity is secretly disputable and paternity uncontentious. I think the researcher called this secret adoption.
At distances longer than several generations, is there noise in the measure of autosomal matches such that you cannot detect these?