There has been some further investigation (thanks to @m4ycon, @stratospher, @octaviolucca, and Ben) in /Satoshi:31.0.0/ peers with no service flags and protocol 70015 pushed transactions, then stopped together on Oct 18 - Observations - Bitcoin Network Operations Collective and Compact block reconstruction degradation between September - October 2026 - Observations - Bitcoin Network Operations Collective. There also was an offline discussion at 2140 recently. To summarize my current understanding of what happened:
Someone wanted to get their uncommon-goods rune mints mined as cheap as possible. The cheapest way of having a transaction with one OP_RETURN for the mint (only one OP_RETURN is allowed per runestone mint) is to have a transaction with a P2A input and a P2A and OP_RETURN output (example). This is about 100vB and costs 10sat per mint at 0.1sat/vbyte in fees. The person creating these 100vB transactions chains them in row. These chains are pushed out to nodes on the network as unsolicited transactions (someone is just sending tx P2P messages, no inv → getdata round trip). Additionally, since P2A are anyone-can-spend, there seems to be someone with the UA /p2aburn:1.0/ burning these outputs to avoid the other person from minting these runes (example).
I suspect unsolicited pushing of long chains of transactions and someone burning (replacing) them as soon as they learn about them is causing a lot of the differences between mempools.
This all stopped on 2026-10-18. No more unsolicited chains of transactions being pushed. Reconstruction rates improved on the same day.
