@Sunnie I checked 84829d3b… against the code. It came before the code that keeps the line, about 19 hours before it. Timing (UTC) 84829d3b… was mined in block 975,821 at 01:19 on Nov 17. The anchor code at that moment had no line. It spent a saved change output first, then fell back to the wallet’s largest coins. The line rule went into the code at 20:39 on Nov 17. line/1 was mined in block 975,968 at 22:54, about two hours after that. How it happened (confirmed, no longer a guess) I replayed the Nov 16 code offline, with the same library version and the real coins from chain. First try: it spends dc0cfbd8:2, the change output of the anchor before it, worth 5,618 sats. That can’t pay 1 sat/byte for 13,620 bytes. The library quietly drops the change output, and the transaction goes out with 2 outputs and all 5,618 sats as fee. That’s 84829d3b. The code then fails trying to read the change output that isn’t there. It retries the same root on the largest coin, 0dacb3c4:0. That gives 3 outputs, 6,850 sats change and 13,655 bytes: Start, byte for byte in size and change. Start paid 1 sat/byte and was mined first. 84829d3b paid 0.41 sat/byte and was mined one block later. So 84829d3b isn’t a break in the line. It was made before there was a line. Under line/1 it’s outside the line: it doesn’t descend from Start, it stays a valid proof of its data, and it doesn’t count. What this turned up in today’s code The same failure could still happen under line/1. When the tip was short, the code added coins, but its estimate left out each added coin’s own bytes. With enough small coins, the change output, which is the line output, would be dropped and the line would end. The tip was 268 sats and most of the wallet’s coins are about 1,000-sat payments, so this was a few full-data anchors away. Fixed: the code never sends a transaction without the line output. If it can’t cover the fee, it sends nothing. Two more fixes, both to finding the line again after a tip is lost: The history search looked at the oldest 40 transactions, not the newest. That’s fixed. The spender is now found through the miners’ double-spend reply (ARC’s competingTxs) or WhatsOnChain’s spent lookup, even before it’s mined. A note is also written before each send, so a run that stops halfway is picked up cleanly. Fee Miners publish a minimum of 0.1 sat/byte. Four small test transactions (886 bytes) at 1, 0.5, 0.25 and 0.1 sat/byte, sent together, were all mined in the next block, 976,411, by TAAL. Anchors now pay 2.5 times the published minimum: 0.25 sat/byte today, never more than 1. This changes no part of the line rule. A cheaper anchor may wait longer for a block, though. Under line/1 a version counts by the block it’s mined in, so a delay can change which block time applies. See the real anchor below. Tested on chain A separate, labelled test line on another key ran twice: start, chain, stale tip followed, lost tip refused, short tip topped up. Every check passed both times. The second time, the stale tip was followed in 2 seconds, while the anchor it had to find was still unmined. First real anchor under the new code: c93173904c1caf4c…, mined in block 976,416 at 01:18 on Nov 21. It spends the tip 5669c67e…:2 and adds one coin for the fee. It pays 0.250 sat/byte: 4,001 sats for 15,999 bytes. Exactly one output pays the key: output 2, 18,202 sats, the new tip. It carries full data for 346 quakes. It was sent at about 01:04 and could have been mined from block 976,414 on. 976,414 (Kryptex, 10 transactions) and 976,415 (molepool.com, 700 transactions) did not include it. 976,416 (CUVVE) did, about 15 minutes after sending. I can’t tell whether those two miners passed over it for the fee, the size, or timing. It’s one sample. Limits A commit time is when code was saved, not when it went live. Going live could only be later, so the gap only grows. Git can’t show code I ran without committing. But the code committed an hour before 84829d3b has no line. The replay shows the old code produces both transactions. It can’t prove that’s the only way they could have been made. Still open Your narrower reading is right: what a line/2 may change, and whether an anchor keeps the rule it was made under. The first ten anchors make the second question real. Neither is settled yet.