How would they even measure their accuracy? By definition, there's not even a baseline to measure against, right?
There are a number of ways that we can check for errors. First, there are many different sequencing and alignment tools, each with different characteristics. For example, by cross checking long read sequencing from a nanopore sequencing deveice and more common Illumina paired end sequencing, we can see where they agree or disagree and then further check with another validated method like Sanger sequencing, if we’re really confused about which is correct. Also, we already know a bit about biology, so we can check the sequence for obviously wrong patterns. Like if our sequencing says the ferret has a mutation that would destroy a critical protein’s function (e.g. a frameshift or premature stop codon) but the ferret looks fine, then we can reasonably infer that the sequencing was wrong somehow. Finally, you’re right that there’s not a “baseline”. All processes in biology are inherently lossy. That said genome sequencing uses pieces of the cellular machinery (DNA polymerase) that can copy gene sequences with incredibly high fidelity, so we rely on biology’s incredible achievement to be reasonably sure that we’re getting the “right” answers.
This is the best of what the intenet can be.
And to the reconstruction: I don't think it does in this case. As they talk about working with frozen cells and we are not in the realm of frozen Mammuts who's genomes has degraded and needs to be reconstructed. In that case yes, we would talk about smaller pieces.
No process is 100% clean in biology, that's how we get random mutation and evolution of species
My guessis that they already used the same tools with sheeps or rats in a freezer. Combining that info with some other tool to measure how much dqmage the sample gpt [1] may give an estimation.
[1] I guess the exact damage depends on how fast it froze and the changes of tenperature and other difficut to know details.