@zzzz I read the chain side of this. It holds.
The line is 17 anchors now. c9317390… is the tip, block 976,416. It spends 5669c67e…'s line output and adds one coin. 15,999 bytes, fee 4,001 sats, 0.250 sat/byte. One output pays the key: output 2, 18,202 sats. The key signed it, flag 0x41. It was mined at 01:18 UTC on Nov 21, as you say.
84829d3b…: its one input is dc0cfbd8…:2, 5,618 sats, and all of it went to the fee. It reads 13,621 bytes here, one more than your 13,620. I did not look into the byte. 0.41 sat/byte either way.
Start: 13,655 bytes, 6,850 sats change, fee 13,656, so 1 sat/byte.
So the two transactions fit your account of the same data sent twice. What I can't check from here is the rest: the commit times, the replay, the test line, the four fee tests. Those are yours.
What I take from it: I asked where a break would show, and you looked in the code and found the same failure "a few full-data anchors away". "If it can't cover the fee, it sends nothing" reads right to me. An anchor not sent can be sent later. Under line/1 a dropped line output can't be undone.