bug: Surplus of CoW AMM orders in settlement_scores table is zero

[fhenneke]

Problem

The surplus of JIT orders is not accounted for correctly in the autopilot when populating the settlement_scores table. The problem is present in prod and staging. Scores from the driver, however, are correct.

Impact

This error impacts reward computations as the reward script uses surplus from the settlement_scores table. This lead to small discrepancies in payments mostly to Quasilabs which were corrected manually. Other solvers are also affected, but the overall damage is small as of now.

Wrong data is currently uploaded to Dune.

To reproduce

This is easiest to see in examples of settling only one order in a batch as for this settlement (staging, auction id 10859238, block number 20261350)
The query select * from settlement_observations where block_number = 20261350 gives a surplus of 0, while the query select * from settlement_scores where auction_id = 10859238 gives a winning score of 57387933137402 which equals the surplus of that order since there are no protocol fees. The score (and surplus) are double checked using the circuit breaker which contains an independent implementation of scores (and surplus).

The problem is less obvious for other settlement (e.g. this settlement, auction id: 20272499, block number: 9116222), but the same arguments apply: surplus is stored as 46523400998314531 but the score suggests a surplus of 46624022071370740 which is larger by the surplus of the CoW AMM order.

Alternative solutions

I think that the rewards script could get by with just using the winning score directly instead of recomputing it from surplus and protocol fees. The correctness of the scores themselves is checked by the circuit breaker (with the next release).
We do upload surplus to Dune and for that it would be convenient to have a correct surplus computation in the autopilot. This is almost required since surplus for JIT orders cannot be directly computed from the database except by parsing the stored calldata.

cc @harisang

[fleupold]

cc @Mateo-mro is this on your radar (as far as I know you are finishing up CoW AMM integration work)?

[m-lord-renkse]

cc @Mateo-mro is this on your radar (as far as I know you are finishing up CoW AMM integration work)?

@MartinquaXD and I worked on the CoW AMM. I can definitely take a look into this, I am doing some CoW AMM related tasks so I can add it to the pile. The only thing is I can only start working full on it from next week, is that is ok @fhenneke .

[fleupold]

Also cc @sunce86

[sunce86]

Since cow amm orders are no longer part of the autopilot Auction, the code assumes decoded cow amm orders are not eligible for surplus: https://github.com/cowprotocol/services/blob/main/crates/autopilot/src/on_settlement_event_updater.rs#L215

To fix this, we need #2793 implemented, so that we have a mechanism how to differentiate between cow amm orders and jit liquidity orders in the autopilot.

[m-lord-renkse]

Being that the case, I will prioritize that task next week.