Author(s)
Apsarth Sumit
Experiences and Qualifications
I work in TypeScript and Python. My public repository nosana-adapter-foundry, on the GitHub account above, is a prototype that checks whether a deterministic artefact survives a GPU network job round trip, with evidence records that separate simulated runs from live ones. It has not yet passed its live test, and I say so in the README.
I used an AI coding assistant to write the first version of the analysis code for this proposal. I have reviewed it and will maintain it myself, and every number in this post comes from code that is public and can be rerun.
Grant Description
CIP-88 replaced the flat quote reward with a dedicated budget of 10% of protocol revenue, split across the five quotes closest to the executed price (weights of 1/e squared). The CIP passed with 36.69M votes For against a 35M quorum, and the docs were updated on 6 October.
In the discussion thread, commenters asked for per-chain absolute reward figures and for code to recompute the statistics for later weeks. The simulation in the thread also left several things unmodelled: a guaranteed floor, the gross versus net-of-partner budget basis, unverified quotes, and buy orders. The core team has a mandate to retune the allocation, so these questions are likely to matter again.
I downloaded the anonymised data published in the thread (1 to 7 September, Ethereum, Base and BNB Chain) and built a first pipeline. With one fixed configuration it reproduces the best-quote reward share for sell orders to within four points on each chain (Ethereum/Base/BNB Chain: 57.8/67.0/59.7 percent in the post against 61.1/64.2/58.8 from my pipeline). It does not match the order counts or the flat-reward cost figures, so part of this grant is closing that gap with the authors.
Early results from the same data, before any funding (in-scope orders: 13,910 Ethereum, 10,630 Base, 17,652 BNB Chain):
- Which quotes count changes how much of the budget is paid out. Under an equal split per in-scope order, only 23% of the BNB Chain budget would be allocated if just verified quotes from the main estimator pool are eligible. Including unverified quotes raises that to 85%, but the median error of the best quote rises from 2.8 to 7.7 bps. On Ethereum the same switch moves coverage from 81% to 99%, and on Base from 75% to 99%.
- The budget basis matters. Gross versus net-of-partner fees changes the 10% budget by 20 to 37 percent per chain.
- The old rule’s native-token caps suggest BNB Chain is where CIP-88 reduces payouts for most solvers, while on Ethereum no solver is worse off. This comparison is an upper bound for the old rule and needs the simulation authors’ eligibility definition to firm up.
These numbers come from my own pipeline and are not yet reconciled with the thread, which is part of Milestone 1.
As far as I can tell, no other application covers this. If I missed one, please point me to it and I will coordinate or narrow the scope.
This grant would deliver:
- An open source pipeline that implements the CIP-88 formula exactly as the docs define it, with a documented reconciliation against the simulation numbers.
- A sensitivity analysis on the open questions from the thread.
- A script that recomputes everything for any accounting period, so the core team or any solver can rerun it.
Grant category: Protocol economics analysis and tooling (solver side).
Out of scope: any change to the protocol, the services repository, or reward payouts. This grant produces analysis and tooling only, released under the MIT license from the first commit.
Type of Grant
Milestone-based.
Milestones
| # | Title | Due (after Snapshot passes) | Funding |
|---|---|---|---|
| 1 | Exact CIP-88 pipeline and reconciliation | Week 2 | 1,000 xDAI |
| 2 | Sensitivity analysis and report | Week 4 | 1,500 xDAI |
| 3 | Refresh script and documentation | Week 5 | 1,000 xDAI |
Milestone 1 (about 20 hours). A public repository with a pipeline that computes quote rewards from the published data format, following the docs formula (network and volume fee adjustment, top five, 1/e squared weights, equal split on zero error). Deliverables: the code, a README, and a reconciliation table against the thread’s figures per chain. Done when the table is published and every difference I cannot explain is listed with the exact question for the CIP authors.
Milestone 2 (about 30 hours). Analysis of the questions raised in the thread, per chain: (a) a guaranteed floor at the old flat amount, showing which solvers gain and lose; (b) the budget on gross versus net-of-partner fees; (c) the effect of including unverified quotes in the top five; (d) buy orders; (e) absolute reward per chain under the new rule against what the flat rule paid. Deliverables: a written report posted to the forum and the scripts that produce every figure. Done when a reviewer can rerun the scripts and get the reported figures.
Milestone 3 (about 15 hours). A command line script that takes one accounting period of data in the published format and outputs the per-solver and per-chain rewards and all the checks above. Deliverables: the script, usage documentation, and a run on any later period the core team provides data for. Done when the script runs end to end on the published week from a clean checkout.
Dependency and risk: the published data covers 1 to 7 September, before CIP-88 took effect, so it cannot show how solvers respond to the new rule. I will ask the core team for a later period. If none is available, Milestone 3 is completed on the published week and the post-activation run is listed as follow-up work, not as a missed deliverable.
Length
5 weeks from the date the proposal passes on Snapshot, which is also the commencement date.
Funding Request
3,500 xDAI total, no COW. The estimate is about 65 hours of analysis and engineering work (20 + 30 + 15), at roughly the rate paid for the CoWAMM trade-envy data pipeline (3,750 xDAI for about 75 hours). I am happy to adjust the scope or the amount to what the committee considers reasonable.
Payment address
I will send the payment address to my steward when the proposal is approved. The template asks for a Gnosis Chain address, but the program docs say the Grants treasury moved to Ethereum mainnet under CIP-82, so I propose to be paid on mainnet and will confirm the network with the steward first.
Other Information
- CIP-88 discussion: CIP-Draft: Quote Competition Redesign
- Docs pull request with the formula: cowprotocol/docs, pull request 664
- Snapshot vote: the CIP-88 proposal in the CoW DAO Snapshot space, which closed with 36.69M votes For
- Reviewer: I would welcome a solver-side reviewer, since this tooling is most useful to solver teams.