I think OPR needs some sort of unconditional fees to work.
In the current LN protocol, the party that creates a payment only pays forwarding fees if the payment is successful. Unsuccessful payments, and HTLCs designed to fail (called probes), cost their creator nothing.
However, if that model was adopted by OPR, then Mallory could send a constant stream of probes through Bob to random endpoints. Each of those endpoints would fail the probes (because they wouldn’t know the payment preimage) in the normal case, but it would occasionally be the case that there would be a problem downstream and a penalty payment would propagate back to Mallory.
Even with unconditional fees, OPR would seem to create an incentive for an attacker to create disruptions within the network. Defending against such attacks could become a centralizing factor (e.g. in the same way that many popular websites use CloudFlare today because it’s one of the few services big enough to survive cheap-to-perform DDoS attacks).
If the risk-free interest rate is 2% per annum, then we would expect the forwarding fee in a mature system to approach 0.00000004% per hop for a 600 ms average settlement time (2 / (86400 / 0.6) / 365.25), or 0.0000004% per payment assuming 11 hops. The OPR fee is 5 orders of magnitude higher. Lightning Network Statistics | 1ML - Lightning Network Search and Analysis Engine - Bitcoin mainnet says the current median percentage fee (not factoring in the base fee) is “0.000072 sat/sat”, or 0.0072% per hop, or 0.08% per payment assuming 11 hops, so OPR accidental routing failures is roughly expected to double the routing cost over the current system or 10,000x increase the cost over a (very theoretical) expected floor.
That does not count the costs of non-accidental failures (i.e. attacks) or the costs of preventing attacks.
One theoretical solution, proposed years ago, is for low-value payments to be encapsulated in higher-value outputs using probabilistic payments. For example: when it would cost $10 in onchain fees to settle an HTLC, Alice offers Bob a $100 HTLC with an HTLC-Success transaction that gives him a 1% chance of being able to claim it onchain (the other 99% of the time, Alice reclaims the money). Bob accepts that and forwards $1 to Carol (perhaps also using a probabilistic payment). If the HTLC is settled offchain (either success or failure), then the $100 output is reallocated $1 to Bob and $99 to Alice. If it goes onchain, then Bob gets his 1% chance at $100 (i.e., an expected value before fees of $1) which will cost him $10 in fees if he’s successful; in the other 99% of the time, Alice gets her $100 back (expected value $99) by paying $10 in fees.
Probabilistic payments are possible today on Elements-based sidechains using OP_DETERMINISTIC_RANDOM.
It’s my belief (possibly ill-informed) that there’s been minimal work in this direction due to HTLC settlement cost griefing not being perceived as a major problem at present.
Opportunistic overpayments also has this property for almost all payments. If, in your proposal, the burn output is expected to be a multiple of the maximal OPR HTLC amount, then the same amount of funds could be used instead for an refundable overpayment that would significantly increase the odds of rapid payment success.
Most discussion about overpayments that I’m aware of has preferred a PTLC-based mechanism, which is something that’s still being developed for LN.