Engine v25
Engine v25 is the proprietary quantitative engine that decides ranges, hedge sizes and rebalances; this page explains what it does and how its decisions reach the chain.
Every strategy decision in Tuo comes from Engine v25, a proprietary quantitative engine developed over several years of backtesting and a live pilot. It runs off-chain as a service. The contracts hold no strategy logic at all.
What it decides
For each position, once a day, the engine answers a fixed set of questions:
| Question | Signal |
|---|---|
| Where should a new range sit and how should the capital split? | Deploy plan: tick range and capital split |
| Should an Alpha range open now? | LP entry, gated by the entry signal and a cooldown |
| Should a range close? | LP exit on a signal flip, or a forced exit after the out-of-range patience window |
| Should the position wait? | Hold, either in range or while out of range within the patience window |
| How large should the short be? | Short open sizing; hedge rebalance when coverage drifts past the trigger |
| Should the short close? | Short close, paired with the range exit it covered |
The engine is frozen. Its core is a byte-immutable package pinned by checksums, guarded by golden regression tests, and never edited in place. The service layer around it only translates state in and signals out.
How a decision reaches the chain
- The backend assembles the position's state: open ranges, idle balances, open shorts, days out of range, and indicators computed on closed daily candles to match the engine's daily loop.
- The engine returns typed signals. Each carries a status: derived when the state fully determined the signal, or stubbed when something was missing.
- The backend never executes a stubbed signal. It is quarantined for a human, with the missing input named.
- A derived signal becomes a keeper transaction. The backend snaps ticks to the pool's spacing, builds the swap route, simulates, and checks the vault's rolling budgets before signing.
- The vault checks the call again on-chain: allowlisted pool, minimum range width, price floor, action budget, loss budget. A call that fails any check reverts.
The backend runs this cycle every five minutes so that a signal is acted on within minutes of the daily candle closing, and it values the position on the same cadence.
Two rules that shape every signal
- The hedge basis differs per sleeve. Alpha ranges size their short off the full range value. The DynHedge range sizes off the exact Uniswap V3 exposure to the volatile token. See Hedging on Hyperliquid.
- A shared short is never shrunk automatically. When one short covers two ranges, it is sized for both, and reducing it requires an operator's confirmation. Shrinking it to fit one range would uncover the other.
When the venue cannot be read
If the backend cannot read the Hyperliquid account, the engine abstains from every hedge signal rather than treating the book as empty. A failed read is never mistaken for a flat book.
When the engine is down
Positions hold. The keeper only acts on a signal, so an unreachable engine means no new ranges, no resizes and no exits until it is back. Existing ranges keep earning and existing shorts keep hedging. Tuo's operators are alerted. Your withdrawal path does not depend on the engine: the exit chain is driven by your on-chain request, and the emergency exit needs no backend at all. See Operational risk.
Capital deployment during the guarded launch
During the guarded launch, capital sizing for a new position is confirmed by a Tuo operator through a signed operations interface rather than executed automatically from the engine's sizing signal. The confirmed plan enters the same queue and passes the same simulation, budget and on-chain checks as any other keeper action. Range placement, hedge sizing and exits are fully engine-driven.
What is not published
The engine's parameters, indicators and entry logic are proprietary and are not documented here. What is published is everything that bounds it: the products' composition and hedge models on Products, and the on-chain perimeter on Roles and admin controls and Parameters.
The target returns and drawdowns on the product pages come from the engine's backtests over past market data. They are backtest targets, not projections.