# Backtest and Tuning Protocol ## Build NPZ Inspect `pool_metadata.json` first. Do not assume token0 is stable. Example if token0 is USDC and token1 is WETH: ```bash python3 dex_platform/data_collectors/build_cl_fee_replay_npz_v2.py \ --events DEX_DATA//events_all.parquet \ --out-npz DEX_DATA/fast_npz/_fee_replay_v2.npz \ --pool-name \ --token0 USDC \ --token1 WETH \ --dec0 6 \ --dec1 18 \ --quote-token token0 \ --fee-rate ``` ## Capacity tune If v3 tuner exists: ```bash python3 dex_platform/backtest/cl_fee_replay_fast_npz_v3.py \ --npz DEX_DATA/fast_npz/_fee_replay_v2.npz \ --out-dir DEX_REPORTS/_capacity_tune_april_v3 \ --fee-rates real_fee: \ --time-from 2026-04-01T00:00:00Z \ --time-to 2026-05-01T00:00:00Z \ --capital-mode grid \ --capital-grid 25,50,75,100,125,150,175,200,250,300,400,500,600 \ --grid-lower 50,60,70,80,85,90,95 \ --grid-upper 1,2,3,5,8,10,15,20 \ --target-mdd-pct 20 \ --max-avg-liquidity-share-pct 3 \ --max-p95-liquidity-share-pct 5 \ --max-p99-liquidity-share-pct 10 \ --max-liquidity-share-pct 25 \ --plots ``` ## Tranche-DCA tune ```bash python3 dex_platform/backtest/cl_fee_replay_tranche_dca_v1.py \ --npz DEX_DATA/fast_npz/_fee_replay_v2.npz \ --out-dir DEX_REPORTS/_tranche_dca_april_v1 \ --time-from 2026-04-01T00:00:00Z \ --time-to 2026-05-01T00:00:00Z \ --fee-rate \ --total-capital-grid 50,75,100,150,200,300,600 \ --parts-grid 3,4,5,6 \ --lower-grid 35,40,45,50,55,60,70,80 \ --upper-grid 5,10,15,20,25,30 \ --add-mode equal,double_base,double_once,progressive \ --max-reentries-grid 1,2,3,4 \ --min-reentry-hours-grid 0,6,12,24 \ --anchor-mode current_price,rolling_low \ --support-lookback-hours 6,12,24 \ --support-offset-pct 0,2,5 ``` ## Monthly validation Test all promising configs: ```text Feb: 2026-02-01 -> 2026-03-01 Mar: 2026-03-01 -> 2026-04-01 Apr: 2026-04-01 -> 2026-05-01 Quarter: 2026-02-01 -> 2026-05-01 ``` Do not recommend April-only winners.