Hi ZmnSCPxj,

You are of course correct. I had considered the effect of reorgs, but the
email seemed to be getting too lengthy to mention that too.

You would need a few spare blocks in which Bob won't be accused of bribery
as a safety margin, which does reduce the time frame in which Alice can get
her transaction confirmed in order to have a valid bribery fraud. This
seems workable if the time frame was long enough (over a few hours should
be sufficient, assuming we consider reorgs of over 3-4 blocks to be
unlikely), but could indeed be problematic if the time frame is already
short to begin with.

Nadav

On Thu, Jun 25, 2020 at 7:04 AM ZmnSCPxj <ZmnSCPxj@protonmail.com> wrote:

> Good morning Nadav,
>
> > > I and some number of Lightning devs consider this to be sufficient
> disincentive to Bob not attacking in the first place.
> >
> > An additional disincentive could be introduced in the form of bribery
> proofs for failed attempts.
> >
> > If we assume that "honest" users of the LN protocol won't reveal their
> timelocked transactions before reaching the timelock expiry (they shouldn't
> anyway because standard full node implementations won't relay them), we can
> prove that Bob attempted bribery and failed to an outside observer by
> showing Bob's signed timelocked transaction, spending an output that was in
> reality spent by a different transaction prior to the locktime expiry,
> which should not be possible if Bob had waited.
>
>
> Unfortunately this could be subject to an inversion of this attack.
>
> Alice can wait for the timelock to expire, then bribe miners to prevent
> confirmation of the Bob timelocked transaction, getting the Alice
> hashlocked transaction confirmed.
>
> Now of course you do mention "prior to the locktime expiry" but there is
> now risk at around locktime.
>
> Particularly, "natural" orphaned blocks and short-term chainsplits can
> exist.
> Bob might see that the locktime has arrived and broadcast the signed
> timelocked transaction, then Alice sees the locktime has not yet arrived
> (due to short-term chainsplits/propagation delays) and broadcast the signed
> hashlocked transaction, then in the end the Alice side of the short-term
> chainsplit is what solidifies into reality due to random chance on which
> miner wins which block.
> Then Bob can now be accused of bribery, even though it acted innocently;
> it broadcasted the timelock branch due to a natural chainsplit but Alice
> hashlocked branch got confirmed.
>
> Additional complications can be added on top to help mitigate this edge
> case but more complex == worse in general.
> For example it could "prior to locktime expiry" can ignore a few blocks
> before the actual timelock, but this might allow Bob to mount the attack by
> initiating its bribery behavior earlier by those few blocks.
>
> Finally, serious attackers would just use new pseudonyms, the important
> thing is to make pseudonyms valuable and costly to lose, so it is
> considered sufficient that LN nodes need to have some commitment to the LN
> in the form of actual channels (which are valuable, potentially
> money-earning constructs, and costly to set up).
>
> Other HTLC-using systems, such as the "SwapMarket" being proposed by Chris
> Belcher, could use similar disincentivizing; I know Chris is planning a
> fidelity bond system for SwapMarket makers, for example, which would mimic
> the properties of LN channels (costly to set up, money-earning).
>
> Regards,
> ZmnSCPxj
>