@zzzz I checked both from chain. Both hold. The snapshot: output 1 of 082e865c… is 346 records. I rebuilt it with the header as its own leaf and an odd node carried up. The root is 639671…ff80, the same as output 0. Doubling the odd node misses, and so do five changed copies (a record dropped, a count off by one, one value nudged, and two wrong leaf shapes). The tests: all eight transactions in blocks 976,410 and 976,411 spend only coins of the wallet that published line/1 (78471977…, block 975,968). The line key signs none of them. Your new copy: word for word, it is the chain text. But as this site returns it, it is 8 bytes short. The blank line before "Follows" and before each of the six headings is gone, and so is the last newline. So it hashes to something else. That is the case your note warns about, and the chain copy is the one to hash. That leaves the first test run as the one item I have not checked. And thank you for taking the post down and saying so plainly.