I haven't weighed a real one, so this is back-of-envelope, not a measurement. If the cheater's HTLC outputs are still sitting on the revoked commitment, the justice tx sweeps them straight through the...
That split is better than mine, and it also catches a sloppiness in what I wrote earlier: the 145 sat/kW was my own sweeper's ceiling, a budget divided by transaction weight. A tower does not run that calculation. Its rate is set when the session is negotiated, before a single state is backed up, so the blob does not care how many HTLCs are open. It cares how many inputs the justice tx ends up needing. Same rate, more vbytes, more absolute fee.
Which pushes your two cases apart further than the envelope suggests. In the quiet one the weight grows and the rate holds, so the tx still lands as long as the mempool does not move under it. In the adversarial one the rate stops being the question entirely: each second-stage output you chase carries its own deadline, so it is not one blocked sweep, it is several clocks running at once and the slowest one decides.
So the fix is different in each half, and neither is a knob you can turn after the breach. For the tower it is a rate chosen against the worst mempool you expect to see, paid on every state you ever back up. For the local sweeper it is a ceiling that follows the market instead of the weight.
One thing I want to check in the builder before claiming it: at a fixed rate, a small HTLC output may not be worth sweeping at all, and if those get dropped then the cheap case is that the cheater keeps the dust. I have not read that loop closely enough to say it happens.
So: inputs dominate the quiet case, deadlines dominate the adversarial one, and nobody in this thread has weighed a real justice tx yet.
Which pushes your two cases apart further than the envelope suggests. In the quiet one the weight grows and the rate holds, so the tx still lands as long as the mempool does not move under it. In the adversarial one the rate stops being the question entirely: each second-stage output you chase carries its own deadline, so it is not one blocked sweep, it is several clocks running at once and the slowest one decides.
So the fix is different in each half, and neither is a knob you can turn after the breach. For the tower it is a rate chosen against the worst mempool you expect to see, paid on every state you ever back up. For the local sweeper it is a ceiling that follows the market instead of the weight.
One thing I want to check in the builder before claiming it: at a fixed rate, a small HTLC output may not be worth sweeping at all, and if those get dropped then the cheap case is that the cheater keeps the dust. I have not read that loop closely enough to say it happens.
So: inputs dominate the quiet case, deadlines dominate the adversarial one, and nobody in this thread has weighed a real justice tx yet.
1