Search projects, hackathons, sponsors and tech

    ← All projects

    SUBFLOOR

    A signed price floor enforced inside 1inch SwapVM settlement, so an agent cannot sell below it

    Demo

    About

    SUBFLOOR is a vault where an autonomous agent trades your whole portfolio and one thing is impossible by arithmetic: a fill below the worst price you signed. The floor is not enforced by the strategy, the agent, or anyone watching the mempool. It lives in the settlement function itself, the code no program can reach around, and a machine-checked proof covers every program that could ever run, not just the ones we tested. The authority to set the floor sits on a hardware device in your hand, so the machine that trades is never the machine that defines "worst." A public index recomputes every fill against every floor, so the guarantee is not our claim about our own execution, it is anyone's query. It runs live on mainnet with our own money. A detector can be wrong. A bound cannot. Prior art, named because it is real: CoW Protocol's settlement contract already enforces on-chain that no order clears worse than what it specifies. But that is taker-side and per discrete order, one signed order at a time. Nothing covers a delegated maker running continuous two-sided programmatic strategies on standing inventory, which is exactly what an agent-run book is. This property exists for taker orders. Nobody gives it to delegated makers. SUBFLOOR implements ERC-8377 (Reference-Relative Slippage Bounds), a draft standard I authored (ethereum/ERCs PR #1935, public since Aug 2026). The specification is public prior art; every line of implementation here was written during the event, and none of the ERC's reference implementation is reused.

    How it's made

    SUBFLOOR is a vault where an autonomous agent trades your whole portfolio and one thing is impossible by arithmetic: a fill below the worst price you signed. The floor is not enforced by the strategy, the agent, or anyone watching the mempool. It lives in the settlement function itself, the code no program can reach around, and a machine-checked proof covers every program that could ever run, not just the ones we tested. The authority to set the floor sits on a hardware device in your hand, so the machine that trades is never the machine that defines "worst." A public index recomputes every fill against every floor, so the guarantee is not our claim about our own execution, it is anyone's query. It runs live on mainnet with our own money. A detector can be wrong. A bound cannot. Prior art, named because it is real: CoW Protocol's settlement contract already enforces on-chain that no order clears worse than what it specifies. But that is taker-side and per discrete order, one signed order at a time. Nothing covers a delegated maker running continuous two-sided programmatic strategies on standing inventory, which is exactly what an agent-run book is. This property exists for taker orders. Nobody gives it to delegated makers. SUBFLOOR implements ERC-8377 (Reference-Relative Slippage Bounds), a draft standard I authored (ethereum/ERCs PR #1935, public since Aug 2026). The specification is public prior art; every line of implementation here was written during the event, and none of the ERC's reference implementation is reused.