StartOS 0.4.0.1, Core Lightning 26.06.8:3, node alias IRONMONKEEEEE_CLN.
I tried several times to turn Splicing off under Actions & Config, Experimental Features. The selection does not stick. The control returns to the middle dash. Dual Funding and Liquidity Ads is set to Disabled and that one does stick.
That matters because splicing is no longer experimental in this CLN release. It is on by default, and the old experimental-splicing flag is accepted and ignored. The StartOS toggle cannot be used to turn it off.
What happened on the node:
Four channels had remote-initiated splices. Each one completed, returned briefly to CHANNELD_NORMAL, then CLN received a protocol error and the channel went on-chain. onchaind logged “Could not find resolution for output 2/3: did we cheat?” and then “onchaind complete, forgetting peer.” No sweep was created. The outputs stayed in listchainmoves with debit_msat 0 and never appeared in listfunds.
Three of those outputs have since been spent on the revocation branch, not the normal 144-block CSV path. That is 3,043,561 sats. A fourth output, 1,122,274 sats, is still unspent and can be taken the same way.
Funding spends:
- f54914b3b3bea085ad13ba28414423c833ac42142c66db62a4a767b04129bb14
- ffa608bc6c672ed7d6dfd161ca537bbc1579df7c2da575e52806cabcdac30ff8
- c4e0b155116be1d96f7d5fc540a9f0a97fb2bc857a44b7ff45c4c7c58ed8c076
- 7594e7214fdb843af4c488a9e862aa4f5213ffdb2bd9e8cda91717bee8c5984a (output 3 still unspent, channel 969475x193x1)
I have not restored a backup, deleted the database, shut the node down, or broadcast a manual transaction. I have a service backup from September 21, which predates these splices. I also emailed security@blockstream.com and filed a GitHub issue on Core Lightning.
Two questions for Start9:
- Is the splicing toggle known not to persist on 0.4.0.1, and is there a supported way to force splicing off?
- Should this service backup be left untouched until CLN confirms the recovery path?
I can add logs if useful. The live output is the urgent part.