Protocol
Simulation
A full-book simulation runs continuously on the same width policy the operator uses. It does not predict token prices. It shows what the LP book would have paid relative to HODL and relative to a continuous rebalancer.
What is being simulated
Every pool the desk lists is replayed with the exact state machine from width selection: jump-diffusion features in, w* or sit-out out, rebalance when the band dies, emissions and swap margin applied as they are on-chain. The output is a path of vault mark-to-market, not a predicted token chart.
Because the simulator and the bot share a policy, a rank you see on the desk is a historic payoff of this strategy, not a backtest of a different one.
How to read the table
A sound way to pick a pool is:
- Look at 12–24 hours of simulation, not the last print.
- Prefer a pool that has stayed high in the rank, not one that spiked because APR printed a 4-digit annualised number on a quiet hour.
- Restrict to a token pair you are willing to hold. If you would not inventory the residual mix after a sit-out, do not LP it.
Highest APR is often highest rebalance cost and highest σ_eff. Those pools can sit at the top of a naive yield list and the bottom of Π(w). The simulation is there so you do not have to do that filter by hand.
What the numbers mean
| Column idea | Definition |
|---|---|
| vs HODL | V_strategy − V_HODL, same start mix, never traded. |
| vs rebalancer | V_strategy − V_reb, where the rebalancer trades back to 50/50 on a short clock and pays spread every time. |
| Rank | Ordering of those excesses, persistence-weighted over the last 12–24 hours. The desk tabs are Simulation Ranks: Ladder (the table) and Simulation Ranks: Snake (how those ranks moved). |
| Jump-adjusted rank | The same ordering after discounting for jump risk. A quiet 12–24 hours is not enough if the pair has a habit of blowing through LP ranges. |
If both excesses are negative, the honest read is: this pair, this week, preferred holding or doing nothing. The desk will sit out for you if you still deposit; you do not have to.
Jump-adjusted ranks
Ladder and Snake can be read raw, or jump-adjusted (theadj control on the desk). The raw rank looks at a short window — typically 12–24 hours — of how the LP book beat HODL and the continuous rebalancer. A smaller-cap pair can look excellent in that window simply because it was quiet.
Quiet is not the same as well-behaved. The jump-adjusted rank looks at a longer print of the tokens and asks how prone the pair is to jumps: moves large enough that a concentrated range is left immediately, so the book pays rebalance and IL instead of earning fees. If a token has a history of those gaps, the method discounts the LP ranking even when the last day printed a high score.
That is the same jump term as in the width maths (λ and J in the jump-diffusion). The simulation is not forecasting the next jump. It is down-weighting pairs whose past jumps would have invalidated the range the strategy would have been sitting in.