DeFi risk automation / public repo

RangeGuard

A stateful bot that watches Uniswap V3 ranges and opens or closes directional hedges on Hyperliquid testnet.

At a glance

What I owned

Period: March 2026 to present
Role: Product and full-stack implementation
Team: Independent build
Users: Concentrated-liquidity operators testing how to manage directional exposure after a range exit
Status: Public source; hedge open and close verified on Hyperliquid testnet

Contribution

Built pool monitoring, hedge execution, encrypted key storage, SQLite state, dashboard controls, circuit breakers, and the anti-whipsaw redesign.

Tradeoff

Respond quickly enough to a real range exit without paying fees and slippage on every short-lived boundary crossing.

Evidence

Public repository, dated testnet walkthrough, observed trade trace, and explicit documentation of what linear perp hedges cannot cover.

Source: Public repository
Venue: Hyperliquid testnet
Observed trace: 11 hedge entries
Modeled saving: 63.5% on that trace

One position, several state boundaries

The bot watches a Uniswap V3 pool, confirms that price has left the configured range, sizes a directional hedge, executes it on Hyperliquid testnet, persists the result, and closes the hedge after price returns.

RangeGuard architecture showing pool monitoring, policy gates, hedge execution, persisted state, and safety controls — recreatedPool state moves through monitoring and policy gates before execution; position state and safety controls persist below.

The first strategy reacted too literally

A dated ETH/USDC testnet trace contained 11 hedge entries near the upper range boundary. Ten of 11 were inside the later 0.3% noise buffer. Opening and closing on those crossings turned correct event detection into repeated fees, slippage, and loss.

RangeGuard boundary plot showing ten noisy hedge entries clustered inside the later confirmation buffer and one confirmed exit — recreatedTen of eleven observed entries sat inside the later confirmation buffer.

The redesign came from the failure

The balanced policy combined a range buffer, five-tick debounce, and ten-minute cooldown. Applied to the recorded trace, it kept one entry instead of 11 and reduced modeled total loss from $11.16 to $4.07, a 63.5% reduction on that dataset.

RangeGuard decision trace showing buffer, debounce, cooldown, and one accepted hedge from eleven observed entries — recreatedBalanced policy on the recorded testnet trace: 11 observed entries, one accepted hedge.

The hedge has a hard limit

Linear perps can reduce directional delta after a range exit. They cannot hedge the LP position's nonlinear gamma, so RangeGuard is useful for testing stateful risk controls and P&L-path stabilization, not for claiming that impermanent loss disappears.

Comparison of a linear perpetual hedge with the nonlinear gamma curve of a concentrated-liquidity position — recreatedA perp can offset direction; the LP position's curvature remains.

Project linksGitHub