Static fees charge the same.
LP risk does not.

Sentinel tests whether Uniswap v4 can price adverse-selection risk better: low fees in normal conditions, higher LP compensation when market conditions become dangerous.

Fund the evidence, not the assumption.

This grant is designed to prove or disprove the economics before production capital is committed. Every deliverable is public, reproducible and falsifiable — including the decision not to deploy.

Applicant: independent solo builder • Contact: mario18g@gmail.com • Ask: $35,000 • Duration: 11 weeks
Working V0 prototype: github.com/chaosxcode/sentinel-hook • Prior work: HookGuard

1. The problem

LP risk changes faster than fee logic. LVR research formalizes the adverse-selection cost LPs face when stale AMM prices are traded against by better-informed arbitrageurs. A 2021 study of 17 Uniswap v3 pools (43% of sampled TVL) measured roughly $199.3M in fees against $260.1M in impermanent loss over its study period — an aggregate gap of $60.8M.

Figure 1 — Historic evidence from arXiv:2111.09192. A historic baseline, not a claim that all LPs lose today.

No fake performance claims: Gate 1 exists specifically to measure whether this problem is strong and predictable enough in current pools to justify Sentinel.

2. First users and target market

Figure 2 — The LP experience. Deterministic pool logic; no offchain operator, no wallet classification.

Initial research focuses on high-volume WETH/stablecoin markets (research filters: trailing 30-day volume ≥ $50M, median TVL ≥ $5M — our filters, not Uniswap requirements). A first pilot would target a new WETH/USDC dynamic-fee pool on Unichain only after all three gates pass.

3. Mechanism

Five signals. One bounded risk score.

T = clip(0.30·V + 0.25·S + 0.20·D + 0.15·I + 0.10·A, 0, 1)

Figure 3 — The V0.1 model. All inputs are pre-swap state only.

These weights are hypotheses, not claimed results. Training data selects parameters; validation freezes them; a locked holdout tests them exactly once. A reviewer should never be able to ask “where did 0.30 come from?” and catch us pretending an initial guess is empirically proven.

4. Why Uniswap v4

v4 supports hook-controlled dynamic fees natively in the swap lifecycle: beforeSwap can return an LP-fee override on a dynamic-fee pool, validated by v4 core under the protocol’s fee rules. Sentinel therefore places the risk decision inside the pool itself — transparent logic executing in the same transaction as the swap — rather than relying on a trusted offchain service to recommend or enforce a fee.

For V1, external oracles stay out of the production fee path entirely — Uniswap’s v4 security framework notes that offchain dependencies expand the failure surface.

5. Security and anti-gaming

Uniswap’s hook security guidance warns that dynamic fees can create trader extraction, manipulation through liquidity splitting, and new MEV vectors — and that worst-case adversarial flows must be modeled, not just averages. Sentinel treats that as the design spec:

Figure 4 — Every anticipated attack paired with a testable defense. Gate 3 attacks this list adversarially.

Plus: no arbitrary external calls in the swap path, fixed-size O(1) state (no unbounded iteration), hard admin ranges with versioned configuration, and minimized upgradeability.

6. HookLab: the experiment

Every strategy receives the same historical event stream and only information available before each trade. An external reference price may be used after the swap to measure LVR outcomes — but no strategy ever sees future or reference information when choosing a fee.

Figure 5 — Four strategies, one scorecard. The volatility+surge baseline is deliberately strong (Aegis-like public design).

Core metrics: LP net economics (fees earned − measured adverse-selection/LVR cost), trader effective fee, slippage, fee-regime distribution, modeled volume retention, gas overhead — reported as pool-month distributions with clustered-bootstrap confidence intervals, so one exceptional market period cannot flatter the whole system.

7. Three hard gates

Figure 6 — Pre-registered pass/fail criteria with defined failure actions.

Gate 1 — Prove the problem

Study ≥3 core pools. Require positive measured adverse-selection cost on ≥70% of active study days, top 10% risk windows explaining ≥30% of loss, and pre-swap score correlation of Spearman |ρ| ≥ 0.15 with ex-post adverse-selection cost on validation data, distinguishable from zero under the pre-registered clustered-bootstrap analysis. Failure: publish the methodology and terminate the live-hook thesis.

Gate 2 — Prove the economics

On locked holdout data: reduce measured LVR by ≥10% vs the best static baseline, improve LP net economics by ≥5% vs static and ≥3% vs the strongest dynamic baseline, remain positive in ≥60% of pool-months, and keep trader friction bounded (volume-weighted effective fee rise ≤1bp without pre-registered justification; modeled volume retention ≥95%). Failure: do not request production-pilot funding.

Gate 3 — Prove safety

≥100,000 stateful fuzz sequences with zero invariant violations, median swap-path overhead ≤40k gas vs the comparable no-hook path, all fee bounds and fallbacks intact, and no materially profitable manipulation in the pre-registered attack suite. An independent security review is required before meaningful capital sits behind the hook — the Uniswap Foundation Security Fund is the intended audit path. Failure: no live capital.

These bars are intentionally demanding. They are our proposed pre-commitments, not established industry thresholds — that is the point: they are falsifiable.

8. Study design and data

Data hierarchy

• Primary (v4 era): raw Uniswap v4 PoolManager events — Initialize, Swap, ModifyLiquidity — because UF’s own v4 data guide warns hooks can make derived datasets incomplete in certain return-delta configurations.
• Primary (pre-v4 training era): Uniswap v3 pool events for the 2023–2024 window.
• Structured layer: the official Uniswap/v4-subgraph (pools, swaps, liquidity modifications, day/hour snapshots).
• Cross-check: Dune / Allium as convenience layers, never the sole source of truth for hooked pools.
• Reference prices: high-frequency public CEX market data (e.g. exchange-published historical trade archives), used strictly ex-post for LVR/adverse-selection measurement — never visible to any strategy pre-swap.

The v3 → v4 transfer question, addressed head-on

Signals explored on v3 history must re-validate on live v4 data (Feb 2025 onward) before parameters freeze for the holdout. If signal behavior does not transfer across protocol versions, that is a Gate 1/Gate 2 failure — reported, not hidden. The July 31, 2026 cutoff gives the study a complete final month before the current grant preparation period. The split is our proposed experimental design, not a Uniswap requirement.

9. Pre-committed result templates

These are the exact charts the final report will publish. They ship empty on purpose — the grant funds filling them in, whatever the answer turns out to be.

Figures 7–8 — Deliberately labeled “NOT A RESULT” so no illustrative bar can ever be mistaken for empirical Sentinel performance.

10. 11-week delivery plan

Figure 9 — Proposed project milestones, not Uniswap-mandated timelines. Open source from week 1.

11. Adoption path

Open-source benchmark → test deployment (Unichain Sepolia) → hook registry → security review / UF Security Fund → routing enablement → small audited pilot → public live dashboard → broader deployment only after evidence.

The first proposed live target is a new WETH/USDC dynamic-fee pool on Unichain, conditional on all three gates passing. UF’s Hooks Routing Rebate Program (up to 85% of routed gas costs subsidized, starting with Unichain) gives hooked pools a concrete distribution path. Proposed pilot planning target: $100K–$250K TVL for ~30 days — an internal target, not a claim of guaranteed partner capital, and not part of this $35K concept-stage scope.

Core hook, HookLab and research outputs remain open source. No token. No hidden paid service required to verify the hook. Phase II (audit, production hardening, monitoring, integrations) is funded separately only if the evidence warrants it.

Budget

13. Who is building this

A solo, self-taught builder-operator — and this application is deliberately structured so the Foundation funds verifiable output, not credentials. Two pieces of that output are already public:

Everything funded here is public and reproducible: if the work is good, the repos prove it; if a gate fails, the report says so. Specialist security depth is explicitly deferred to the independent review / UF Security Fund path rather than claimed. Contact: mario18g@gmail.com

14. Why this fits Uniswap now

Technical appendix — extraction anchors

Anchors from the official Uniswap/v4-subgraph network configuration. Events decoded: Initialize, Swap, ModifyLiquidity. Subgraph queried via The Graph’s GraphQL API with attribute-based pagination.

Primary references

• Uniswap Developer Funding
• Uniswap Foundation v4 Hook Design Lab
• Uniswap v4 Security Framework
• Uniswap Foundation v4 Data Guide
• Official v4 Subgraph
• Loss-Versus-Rebalancing (LVR) paper
• Historic Uniswap v3 LP study
• Hooks Routing Rebate Program

Can Uniswap v4 price LP risk better?

We are not asking the Foundation to believe the answer. We are asking for the resources to measure it, build it, attack it and publish the result — whichever way it comes out.

Fund the evidence.

mario18g@gmail.com