Source: task5-ashare-leader/output/streaming_v0/phase_b_diagnostics/task5_phase_b_diagnostics_summary.md
Last modified: 2026-05-18 18:01:29 · 9,710 bytes
Task #5 Phase B Diagnostics
Purpose: answer Kaite's IC questions with A/B evidence rather than narrative.
1. v0 -> v1.2 Attribution Ladder
| step |
scenario |
full_annualized |
full_sharpe |
full_mdd |
avg_exposure |
trades |
delta_ann_vs_v0 |
delta_sharpe_vs_v0 |
| 0 research baseline |
v0_baseline |
0.1800 |
0.8801 |
-0.2178 |
0.2964 |
445 |
0.0000 |
0.0000 |
| 1 + 50bp buy/sell slippage |
slippage_50bp |
0.1400 |
0.6806 |
-0.2432 |
0.2943 |
454 |
-0.0400 |
-0.1994 |
| 2 + 30% industry cap / limit lock = v1 |
industry_cap_30pct_slippage_50bp |
0.1369 |
0.6780 |
-0.2441 |
0.2929 |
458 |
-0.0431 |
-0.2021 |
| 3 + B3 dual-window RPS = v1.1 |
B3_rps250_125_gt80 |
0.1434 |
0.7164 |
-0.2342 |
0.2824 |
433 |
-0.0366 |
-0.1636 |
| 4 + public-fund combined bonus = v1.2 |
fund_combined_bonus |
0.1476 |
0.7392 |
-0.2234 |
0.2847 |
405 |
-0.0324 |
-0.1408 |
Interpretation: the large first-order degradation is execution friction. Industry cap/limit lock adds little extra cost after friction; B3 and public-fund sponsorship recover part of the lost Sharpe.
2. M-filter Sensitivity: Best FULL Frontier
| scenario |
open_n |
close_n |
full_annualized |
full_sharpe |
full_mdd |
oos_b_annualized |
oos_b_sharpe |
bull_start_lag_days |
avg_exposure |
| open5_close5 |
5 |
5 |
0.1716 |
0.8100 |
-0.2665 |
0.0402 |
0.1589 |
36 |
0.3171 |
| open5_close2 |
5 |
2 |
0.1652 |
0.8039 |
-0.2538 |
0.0483 |
0.1908 |
36 |
0.2795 |
| open1_close5 |
1 |
5 |
0.1700 |
0.7956 |
-0.2868 |
0.0467 |
0.1776 |
30 |
0.3285 |
| open5_close3 |
5 |
3 |
0.1651 |
0.7936 |
-0.2787 |
0.0484 |
0.1914 |
36 |
0.2813 |
| open5_close10 |
5 |
10 |
0.1736 |
0.7934 |
-0.2850 |
-0.0198 |
-0.0765 |
36 |
0.3465 |
| open3_close5 |
3 |
5 |
0.1681 |
0.7889 |
-0.2665 |
0.0402 |
0.1589 |
34 |
0.3228 |
| open10_close1 |
10 |
1 |
0.1463 |
0.7725 |
-0.2225 |
-0.0568 |
-0.2700 |
43 |
0.2406 |
| open1_close2 |
1 |
2 |
0.1601 |
0.7712 |
-0.2784 |
0.0549 |
0.2089 |
30 |
0.2985 |
| open3_close10 |
3 |
10 |
0.1683 |
0.7676 |
-0.2924 |
-0.0198 |
-0.0765 |
34 |
0.3501 |
| open10_close2 |
10 |
2 |
0.1502 |
0.7632 |
-0.2700 |
-0.0577 |
-0.2674 |
43 |
0.2553 |
| open3_close2 |
3 |
2 |
0.1574 |
0.7580 |
-0.2538 |
0.0484 |
0.1914 |
34 |
0.2945 |
| open10_close3 |
10 |
3 |
0.1479 |
0.7532 |
-0.2813 |
-0.0787 |
-0.3605 |
43 |
0.2596 |
3. M-filter Sensitivity: Best OOS-B Rebound Capture
| scenario |
open_n |
close_n |
full_annualized |
full_sharpe |
full_mdd |
oos_b_annualized |
oos_b_sharpe |
oos_b_mdd |
bull_start_lag_days |
| open3_close1 |
3 |
1 |
0.1430 |
0.7048 |
-0.2250 |
0.1271 |
0.5047 |
-0.1962 |
34 |
| open5_close1 |
5 |
1 |
0.1390 |
0.7020 |
-0.2250 |
0.1116 |
0.4529 |
-0.1979 |
36 |
| open2_close1 |
2 |
1 |
0.1487 |
0.7297 |
-0.2346 |
0.0757 |
0.2918 |
-0.1947 |
31 |
| open1_close1 |
1 |
1 |
0.1476 |
0.7392 |
-0.2234 |
0.0697 |
0.2706 |
-0.1896 |
30 |
| open20_close1 |
20 |
1 |
0.0896 |
0.5410 |
-0.2475 |
0.0482 |
0.2294 |
-0.1867 |
57 |
| open20_close3 |
20 |
3 |
0.1292 |
0.7251 |
-0.2664 |
0.0472 |
0.2193 |
-0.1867 |
57 |
| open20_close2 |
20 |
2 |
0.1053 |
0.6177 |
-0.2533 |
0.0472 |
0.2193 |
-0.1867 |
57 |
| open1_close2 |
1 |
2 |
0.1601 |
0.7712 |
-0.2784 |
0.0549 |
0.2089 |
-0.1955 |
30 |
4. Exit A/B: Three-axis vs Hard Stops
| scenario |
full_annualized |
full_sharpe |
full_mdd |
oos_b_annualized |
oos_b_sharpe |
avg_exposure |
trades |
sells |
avg_sell_return |
hard_stop_sells |
delta_sharpe_vs_three_axis |
| three_axis_v1_2 |
0.1476 |
0.7392 |
-0.2234 |
0.0697 |
0.2706 |
0.3930 |
405 |
209 |
0.2668 |
0 |
0.0000 |
| hard8_only |
0.0917 |
0.3923 |
-0.5264 |
0.1708 |
0.5870 |
0.7360 |
287 |
212 |
2.3724 |
63 |
-0.3469 |
| hard10_only |
0.1076 |
0.4649 |
-0.5242 |
0.1285 |
0.4851 |
0.7492 |
259 |
197 |
2.2914 |
50 |
-0.2743 |
| hard12_only |
0.1086 |
0.4663 |
-0.5304 |
0.2383 |
0.9048 |
0.7809 |
198 |
156 |
2.9052 |
30 |
-0.2729 |
| hybrid8_three_axis |
0.1451 |
0.7200 |
-0.2391 |
0.0528 |
0.2083 |
0.3788 |
507 |
266 |
0.2876 |
126 |
-0.0193 |
| hybrid10_three_axis |
0.1369 |
0.6949 |
-0.2287 |
0.0617 |
0.2431 |
0.3855 |
459 |
238 |
0.2619 |
94 |
-0.0444 |
| trend_only |
0.1250 |
0.5923 |
-0.3852 |
0.0501 |
0.1826 |
0.4644 |
325 |
188 |
0.7131 |
0 |
-0.1469 |
5. Stress Supplement
| window |
series |
return |
max_drawdown |
start_exposure |
end_exposure |
avg_exposure |
| 2018_bear |
v1_2 |
0.0000 |
0.0000 |
0.0000 |
0.0000 |
0.0000 |
| 2018_bear |
hs300 |
-0.2634 |
-0.3188 |
1.0000 |
1.0000 |
1.0000 |
| 2018_bear |
csi800 |
-0.2837 |
-0.3261 |
1.0000 |
1.0000 |
1.0000 |
| 2024_smallcap_crash |
v1_2 |
0.0000 |
0.0000 |
0.0000 |
0.0000 |
0.0000 |
| 2024_smallcap_crash |
hs300 |
-0.0063 |
-0.0610 |
1.0000 |
1.0000 |
1.0000 |
| 2024_smallcap_crash |
csi800 |
-0.0176 |
-0.0875 |
1.0000 |
1.0000 |
1.0000 |
| 2025_apr_liberation_day |
v1_2 |
-0.0957 |
-0.1266 |
0.9124 |
0.3577 |
0.5520 |
| 2025_apr_liberation_day |
hs300 |
-0.0353 |
-0.0767 |
1.0000 |
1.0000 |
1.0000 |
| 2025_apr_liberation_day |
csi800 |
-0.0399 |
-0.0835 |
1.0000 |
1.0000 |
1.0000 |
| 2025_q4_drawdown |
v1_2 |
-0.0786 |
-0.1688 |
0.9497 |
0.5703 |
0.8747 |
| 2025_q4_drawdown |
hs300 |
-0.0388 |
-0.0631 |
1.0000 |
1.0000 |
1.0000 |
| 2025_q4_drawdown |
csi800 |
-0.0469 |
-0.0692 |
1.0000 |
1.0000 |
1.0000 |
Files: attribution ladder CSV, full M-filter grid, stop A/B outputs, and stress supplement are in this directory.