# DEX fast NPZ tuner v2 — focused continuation from `wide_80_5` ## Dataset NPZ: `base_CHECK_USDC_2PCT_2026_02_05_fee_replay_v2.npz` Period: 2026-02-01 → 2026-05-01 UTC Fee rate: 0.002515 Swap rows: 133,474 ## Baseline check: `wide_80_5`, capital 1000 | metric | value | |---|---:| | return_pct | 14.116615 | | mdd_pct | -42.420592 | | avg_liquidity_share_pct_when_in_range | 4.939959 | | max_liquidity_share_pct_when_in_range | 72.750944 | | score with hard 5% cap | -1000000000707.933838 | Conclusion: with `initial_capital_usd=1000` and event-level `hard_max_liquidity_share=5%`, the current baseline is rejected. A focused cap=1000 grid also had zero accepted rows. ## Focused tune rule used here Because the current script's periodic loop is slow (~18s per one quarterly periodic run), I used a segment-vectorized focused tuner for exploration and validated it against the original script on `periodic_80_5_168h`. Search space: - Static: lower `[75,80,85,90,95,96,97,98,99,99.5]`, upper `[0.1,0.25,0.5,1,2,3,5,8]` - Rebalance: lower `[80,85,90,95,98]`, upper `[0.5,1,2,5]`, hours `[24,48,72,168,336]`, modes `oor|periodic` - capital: `10`, because this is the only practical scale where hard max share 5% accepts the candidates. At capital 100, 1000: zero accepted rows under event-level hard cap. ## Best quarterly candidates, capital 10 | strategy | return_pct | mdd_pct | score | time_in_range_pct | avg_liquidity_share_pct_when_in_range | max_liquidity_share_pct_when_in_range | fees_earned_total | rebalance_costs | rebalances | |:---------------------|-------------:|----------:|--------:|--------------------:|----------------------------------------:|----------------------------------------:|--------------------:|------------------:|-------------:| | periodic_80_0.5_336h | 37.2366 | -21.1232 | 37.1166 | 53.0688 | 0.146396 | 3.48026 | 5.21616 | 0.332482 | 6 | | periodic_80_1_336h | 36.9196 | -21.2674 | 36.7996 | 53.7745 | 0.143935 | 3.45898 | 5.22274 | 0.33242 | 6 | | periodic_80_2_336h | 36.3647 | -21.6602 | 36.2447 | 55.6745 | 0.138167 | 3.41829 | 5.24209 | 0.332315 | 6 | | periodic_80_5_336h | 35.5181 | -23.2681 | 35.3981 | 62.2968 | 0.122182 | 3.30326 | 5.36518 | 0.332018 | 6 | | periodic_85_0.5_336h | 33.0457 | -19.1408 | 32.9257 | 53.0688 | 0.132075 | 3.15507 | 4.67941 | 0.332198 | 6 | | periodic_85_1_336h | 32.7822 | -19.2939 | 32.6622 | 53.7745 | 0.129932 | 3.13758 | 4.6881 | 0.332143 | 6 | | periodic_85_2_336h | 32.3239 | -19.8129 | 32.2039 | 55.6745 | 0.124863 | 3.10409 | 4.71084 | 0.332052 | 6 | | periodic_85_5_336h | 31.6605 | -21.331 | 31.5405 | 62.2968 | 0.11074 | 3.00905 | 4.83579 | 0.331791 | 6 | | periodic_90_0.5_336h | 29.0723 | -17.3093 | 28.9523 | 53.0688 | 0.118343 | 2.84024 | 4.17002 | 0.331926 | 6 | | periodic_90_1_336h | 28.8552 | -17.5466 | 28.7352 | 53.7745 | 0.116488 | 2.82607 | 4.18013 | 0.331878 | 6 | | periodic_90_2_336h | 28.4803 | -18.0302 | 28.3603 | 55.6745 | 0.112063 | 2.79889 | 4.20498 | 0.331799 | 6 | | periodic_90_5_336h | 27.9713 | -19.4494 | 27.8513 | 62.2968 | 0.0996684 | 2.72147 | 4.32878 | 0.331572 | 6 | | periodic_95_0.5_336h | 25.0327 | -15.525 | 24.9127 | 53.0688 | 0.104221 | 2.51337 | 3.65161 | 0.331646 | 6 | | periodic_95_1_336h | 24.8583 | -15.7424 | 24.7383 | 53.7745 | 0.102647 | 2.50227 | 3.66257 | 0.331606 | 6 | | periodic_95_2_336h | 24.5601 | -16.1853 | 24.4401 | 55.6745 | 0.0988556 | 2.48096 | 3.68845 | 0.331538 | 6 | | periodic_95_5_336h | 24.1881 | -17.4894 | 24.0681 | 62.2968 | 0.0881789 | 2.42 | 3.80809 | 0.331344 | 6 | | periodic_98_0.5_336h | 22.2067 | -14.261 | 22.0867 | 53.0688 | 0.0942416 | 2.28048 | 3.28864 | 0.331448 | 6 | | periodic_98_1_336h | 22.0597 | -14.463 | 21.9397 | 53.7745 | 0.0928569 | 2.27134 | 3.29985 | 0.331413 | 6 | | periodic_98_2_336h | 21.8101 | -14.8744 | 21.6901 | 55.6745 | 0.0894964 | 2.25378 | 3.32576 | 0.331353 | 6 | | periodic_98_5_336h | 21.522 | -16.0888 | 21.402 | 62.2968 | 0.0799961 | 2.2034 | 3.44067 | 0.331181 | 6 | ## Cross-period check of selected candidates | strategy | period | return_pct | mdd_pct | score | max_liquidity_share_pct_when_in_range | rebalances | |:---------------------|:---------|-------------:|----------:|-----------:|----------------------------------------:|-------------:| | periodic_80_0.5_336h | quarter | 37.2366 | -21.1232 | 37.1166 | 3.48026 | 6 | | periodic_80_0.5_336h | feb | -1.29105 | -2.51631 | -1.31105 | 0.222302 | 1 | | periodic_80_0.5_336h | mar | 0.611836 | -18.6758 | 0.571836 | 3.54379 | 2 | | periodic_80_0.5_336h | apr | 26.4792 | -15.35 | 26.4392 | 3.88883 | 2 | | periodic_80_0.5_336h | febmar | -1.25589 | -19.2821 | -1.33589 | 3.48026 | 4 | | periodic_80_5_336h | quarter | 35.5181 | -23.2681 | 35.3981 | 3.30326 | 6 | | periodic_80_5_336h | feb | -1.71501 | -3.25006 | -1.73501 | 0.212683 | 1 | | periodic_80_5_336h | mar | -1.17363 | -20.8577 | -1.21363 | 3.37788 | 2 | | periodic_80_5_336h | apr | 26.9366 | -16.701 | 26.8966 | 3.74862 | 2 | | periodic_80_5_336h | febmar | -3.42994 | -21.9547 | -3.50994 | 3.30326 | 4 | | periodic_85_0.5_336h | quarter | 33.0457 | -19.1408 | 32.9257 | 3.15507 | 6 | | periodic_85_0.5_336h | feb | -1.21853 | -2.3261 | -1.23853 | 0.200908 | 1 | | periodic_85_0.5_336h | mar | 0.454054 | -16.9867 | 0.414054 | 3.21055 | 2 | | periodic_85_0.5_336h | apr | 23.8451 | -13.8586 | 23.8051 | 3.52343 | 2 | | periodic_85_0.5_336h | febmar | -1.3376 | -17.6338 | -1.4176 | 3.15507 | 4 | | static_99.5_0.1 | quarter | 17.37 | -25.2929 | 16.7843 | 1.62951 | 0 | | static_99.5_0.1 | feb | -0.420464 | -1.89737 | -0.420464 | 0.121932 | 0 | | static_99.5_0.1 | mar | -1.89948 | -17.4355 | -1.89948 | 1.83407 | 0 | | static_99.5_0.1 | apr | 21.8628 | -9.07037 | 21.8628 | 2.3594 | 0 | | static_99.5_0.1 | febmar | -6.4602 | -22.5283 | -6.4602 | 1.62951 | 0 | | static_95_0.1 | quarter | 20.766 | -29.8034 | 11.1592 | 1.94397 | 0 | | static_95_0.1 | feb | -0.503235 | -2.26863 | -0.503235 | 0.145893 | 0 | | static_95_0.1 | mar | -2.27747 | -20.8176 | -2.27747 | 2.18713 | 0 | | static_95_0.1 | apr | 26.1363 | -10.8521 | 26.1363 | 2.81069 | 0 | | static_95_0.1 | febmar | -7.7349 | -26.939 | -11.6128 | 1.94397 | 0 | | static_75_8 | quarter | 25.0559 | -45.7961 | -16.5364 | 2.78441 | 0 | | static_75_8 | feb | -2.05029 | -5.37228 | -2.05029 | 0.210637 | 0 | | static_75_8 | mar | -6.10216 | -34.2391 | -24.5803 | 3.12934 | 0 | | static_75_8 | apr | 43.5786 | -19.2727 | 43.5786 | 4.0105 | 0 | | static_75_8 | febmar | -14.7064 | -43.9764 | -52.6593 | 2.78441 | 0 | | static_80_5 | quarter | 25.2755 | -41.5409 | -7.80626 | 2.60042 | 0 | | static_80_5 | feb | -1.46328 | -4.24308 | -1.46328 | 0.196375 | 0 | | static_80_5 | mar | -4.72993 | -30.4656 | -15.6612 | 2.92325 | 0 | | static_80_5 | apr | 38.3032 | -16.701 | 38.3032 | 3.74862 | 0 | | static_80_5 | febmar | -12.5082 | -39.2351 | -40.9783 | 2.60042 | 0 | ## Hard conclusion 1. `wide_80_5` is not valid with `initial_capital_usd=1000` under current hard max share 5% rule. 2. If hard cap remains event-level max share, realistic accepted capital is roughly <=30 USD for static narrow ranges and about 10 USD for the current broad focused search. 3. Quarterly winner at cap=10 is `periodic_80_0.5_336h`: +37.2366%, MDD -21.1232%, max share 3.4803%, 6 rebalances. 4. This result is not stable enough yet: Feb and Feb+Mar are negative; most quarterly return comes from April. Treat this as candidate generation, not final strategy. 5. Next required code change: add fast segment-vectorized periodic/oor tuner or multiprocessing initializer. The current periodic implementation is too slow for real tuning. 6. Next required modeling change: replace hard event-level max share with at least one additional metric: p95/p99 liquidity share. Event-level max share is probably too brittle because tiny-liquidity events reject almost every $1000 run.