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. The full research plan, pre-registered gates and budget are in the grant application.
This is the V0 skeleton referenced in the grant application. It intentionally contains no trained risk model and no performance claims — the grant funds producing that evidence, whatever the answer turns out to be.
What V0 does prove is the safety scaffolding the five-signal model will run inside, working end to end on the real v4 stack:
| Safety property | Where | Verified by |
|---|---|---|
| Hard fee bounds (floor 0.01%, ceiling 1.00%) | SentinelHookV0.sol |
fuzz: fee never leaves bounds across 6,400 randomized swaps |
| Rate-limited fee changes (≤ 0.05% per update) | _stepToward |
fuzz + unit tests |
| Hysteresis + cooldown (no threshold bouncing) | _beforeSwap |
test_CooldownAndHysteresisEaseBack |
| Pre-swap observations only — a trade cannot set its own fee | _beforeSwap |
test_BigMoveElevatesFee_RateLimited |
| Safe base-fee fallback on unknown state | _beforeSwap |
test_UnknownPoolFallsBackToBaseFee |
| Stale-signal recalibration | _beforeSwap |
test_StaleStateRecalibratesTowardBase |
| Dynamic-fee pools only | _beforeInitialize |
test_RevertsOnStaticFeePool |
| Fixed-size O(1) state, no external calls in swap path | whole contract | code review |
Measured V0 gas overhead: ~13.9k per swap vs an identical no-hook pool
(test_GasOverheadVsNoHookPool) — well inside the grant's Gate 3 target of
≤ 40k.
The V0 placeholder rule is a single tick-movement signal with two fee tiers. It is deliberately dumb. The funded work replaces it with the five-signal model (volatility, size pressure, deviation, imbalance, acceleration), benchmarked in HookLab against static and volatility-based baselines on historical data, with pre-registered pass/fail gates.
These are exactly the gaps the funded research closes:
- The movement signal compares consecutive swaps only — slow sustained moves
and quiet-period gaps are not modeled (an attacker can wait out
STALE_AFTER). The full model uses notional-weighted EWMA windows. - No size-pressure, deviation, imbalance or acceleration signals yet.
- Parameters (
ENTER_TICKS, tier fees, cooldown) are unvalidated placeholders, not trained values. - No manipulation-suite testing beyond the invariant fuzz — that is Gate 3.
Unichain Sepolia: 0xcbd5bac7b96770d7f18b97d05d6518a4d0913080
(deploy tx 0xc3e8…97f4) —
CREATE2-mined so the address encodes the hook's permission bits
(beforeInitialize | afterInitialize | beforeSwap = 0x3080).
A demo dynamic-fee pool runs on Unichain Sepolia against the deployed hook
(test tokens STA /
STB,
pool created in 0x2ec0db…).
The choreography in script/04_DemoSwaps.s.sol:
- A calm swap pays the 0.05% base fee.
- A large swap moves the pool −355 ticks — and still pays the base fee, because the fee is decided from pre-swap state: a trade cannot price itself.
- The next five swaps see that movement and ramp the fee — one rate-limited
step per update, every step a public
FeeUpdatedevent:
| Event | Fee change | Tick move seen | Transaction |
|---|---|---|---|
| 1 | 500 → 1000 | −355 | 0x6e8f4b… |
| 2 | 1000 → 1500 | +4 | 0xb7880c… |
| 3 | 1500 → 2000 | −3 | 0xe021b8… |
| 4 | 2000 → 2500 | +3 | 0x2ba01e… |
| 5 | 2500 → 3000 | −3 | 0xdc971a… |
The pool ends at the 0.30% elevated tier (asserted on-chain by the script), held there by hysteresis until the cooldown lets it ease back to base. Note events 2–5: tick moves of ±3–4 keep the target elevated while the fee climbs — rate limiting and hysteresis behaving exactly as the tests promise.
Built from Uniswap v4-template.
git clone --recurse-submodules https://github.com/chaosxcode/sentinel-hook
cd sentinel-hook
forge build
forge test -vvDeploy scripts (script/00_DeployHook.s.sol onward) follow the template's
HookMiner flow; the first deployment target is Unichain Sepolia.
- Grant application: https://chaosxcode.github.io/sentinel-hook/
- Prior work — HookGuard: transparent risk scanner for Uniswap v4 hooks: https://github.com/chaosxcode/hookguard
MIT