Live shadow · 63,052 real swaps · figures from committed reports
PegGuard — measured IL insurance for Uniswap v4
LVR is the priceable part of impermanent loss. PegGuard makes the flow that causes IL pay for it — a dual-mode dynamic-fee hook that charges a premium only on the correcting (toxic) leg, priced off a de-trended oracle deviation. Every constant is calibrated from real on-chain backtests, and every number on this page comes from a committed shadow report. We report what we measured — including where it loses.
99.3%
Premium precision (live)
63,052
Real swaps replayed
+6.23 bps
Net in quiet flow
−0.70 bps
Net, blended (honest)
5.62 bps
Base fee for break-even
We charge the arbitrageur, not your traders
Original idea
In HARVEST mode the premium falls only on the correcting-direction leg — the toxic/arb flow that moves price toward fair value and inflicts LVR on LPs. Measured premium-weighted precision is 92.4% in calm and 98.9% in volatile windows; 99.3% on the live pool. Every other UHI9 fee hook we reviewed (Devia, FluxShield, BackStop, Goldgard, ΔShield) charges a symmetric absolute-deviation fee that taxes benign two-sided flow equally.
Correcting / toxic leg
99.3%
of premium dollars land here — the flow that actually causes impermanent loss.
vs
Benign / two-sided flow
≈ 0
retail and uninformed swaps pay only the base fee — no premium.
Rivals price |deviation| and charge both directions; one even charges its own arbitrage leg the minimum fee. PegGuard prices the signed, de-trended deviation, so only the leg that harms LPs pays.
Honest live economics, by regime
Impact · measured
Over 24.6 h and 63k real Arbitrum WETH/USDC swaps, the blended net is −0.70 bps — but that single number hides the structure. Quiet and normal flow are clearly LP-positive at the current 5 bps base; the negative blend is driven entirely by the high-volatility toxic tail and the un-priced oracle-fallback leg. We publish the breakdown rather than the headline.
Quiet +6.23 and normal +2.43 bps are profitable; high-vol −3.81 and fallback −1.41 bps pull the blend to −0.70. The premium is precise but capped (50 bps) — it cannot out-earn a 5-bps-plus markout tail. The fallback leg charges zero premium and is pure drag we can remove with wider feed coverage.
Where the premium pays for itself
Unique execution
Net LP economics by swap size, same live window. Large toxic prints carry low markout per dollar while the premium scales with deviation — so the ≥$50k bucket recovers 156% of its markout and is net +6.74 bps. The thin, benign retail buckets are where markout out-runs the premium.
Net bps by trade size — +6.74 bps on ≥$50k prints vs negative sub-$50k retail. This is the structural signature: PegGuard wins on large, sharp toxic flow and underperforms on small benign churn at a 5 bps base.
Every constant is calibrated — the base fee is the lever
Measured, not guessed
The research finding the live data confirms: profitability is a base-fee-vs-toxicity problem. The "required base fee for break-even" is each pair/regime's measured toxicity. The calm window already clears at 5 bps (+1.89 bps net); the live blend needs 5.62 bps; WETH/USDT is deeper. The premium is a precise top-up on a correctly-priced base.
Required base fee for 0-net (the toxicity ladder) vs the current 5 bps tier. Calm 3.12 · vol 5.48 · live WETH/USDC 5.62 bps.
Net bps by window/pair: calm +1.89, vol −0.47, live WETH/USDC −0.70, WETH/USDT −1.63. Profitability is regime- and pair-dependent — more capital and cheaper chains do not change the sign; the correct base tier does.
GUARD breaker & the Reactive control plane
Correct RSC use
For stable pairs there is no harvestable fuel (cross-venue deviation 0.000–0.35 bps sits below ~0.2 bps feed noise), so GUARD runs only a depeg circuit breaker. The volatility sentinel is a Reactive Smart Contract that watches the mainnet canonical pool — a signal the hook's own chain cannot observe — and flips the fee regime cross-chain.
40 bps / 180 s
Selected sentinel trigger, calibrated by replay
0
Calm false positives (6 volatile triggers)
74 min
Lead before the heavy-bleed window
$2,399
Bled before firing at a looser 60 bps trigger
Why this is the correct Reactive integration: the RSC does work the contract economically cannot do itself — observe mainnet price discovery and flip the pricing regime — and the hook independently enforces monotone safety (tighten instantly, loosen only after a delay/TTL), so a forged or buggy callback can only make the hook more conservative. A 60-s-late flip costs ~0.4% of episode bleed, which is why Reactive belongs in the control plane and Pyth's 2–5 s pull belongs on the hot path.
Architecture & safety invariants
Functionality
1
Never reverts a swap
Any oracle fault — stale, confidence anomaly, unseeded basis — degrades to the symmetric base fee. The hook never blocks a trade.
2
Slow-basis de-trending
62% of the raw pool-vs-CEX gap is slow venue/quote basis; taxing it inflates fake capture from 17% to 224%. A trailing EMA (τ=450 s) removes it before pricing.
3
Relative confidence gate
Pyth confidence (~10 bps) dwarfs the ~1 bps signal, so an absolute gate never fires; PegGuard gates on confidence vs its own EMA at 3×.
4
Monotone control plane
Tighten instantly, loosen with delay. A Reactive Network outage degrades gracefully via TTL expiry.
5
Research-as-regression tests
52 Foundry tests replay real swap fixtures against event-level truth baselines; a parity-locked Python shadow pipeline reproduces them within ±1pp before any live run.
6
Honest live forward test
24.6 h / 63k swaps on real flow, 98.2% truth coverage. We report the negative blended net plainly — measurement integrity over optimism.