Pre-HardStop Timing Alignment Candidate Safety Comparison Audit

Input verification:
- latest.zip SHA256 expected: 01574CC3481878A15B11B54ECC866E8618C33E9715D36AA22E8B1F6058F037D4
- latest.zip SHA256 actual:   01574CC3481878A15B11B54ECC866E8618C33E9715D36AA22E8B1F6058F037D4
- SHA256 match: True

Calculation formulas:
- hardstop_reduction_yen = virtual_exit_profit_yen - actual_final_profit_yen
- killed_future_profit_yen = actual_final_profit_yen - virtual_exit_profit_yen
- net_effect_after_kills_yen = hardstop_reduction_yen + unresolved_effect_yen - killed_future_profit_yen
- adoption rows are limited to HardStopNow=false / ClosePriorityOK=true / ClosePriorityBlocked=false executable snapshots.

Overall comparison:
plan | eligible_baskets | hardstop_rescue | hte_kill | net | maxdd | worst_log | positive_logs | max_1_log_dep | max_1_basket_dep
TA9 | 30 | 26 | 1 | 32760.00 | 2374.00 | 2616.00 | 6/6 | 29.29% | 7.25%
TA5 | 21 | 20 | 0 | 20063.00 | 1602.00 | 0.00 | 5/6 | 32.93% | 7.98%
TA4 | 16 | 16 | 0 | 14003.00 | 1358.00 | 0.00 | 3/6 | 46.75% | 9.70%

By-log comparison:
plan | period | deposit | baskets | hardstop | hte_kill | unresolved | net | share
TA9 | 2026.03.01-2026.05.29 | 100000 | 5 | 5 | 0 | 0 | 9594.00 | 29.29%
TA5 | 2026.03.01-2026.05.29 | 100000 | 1 | 1 | 0 | 0 | 1544.00 | 7.70%
TA4 | 2026.03.01-2026.05.29 | 100000 | 0 | 0 | 0 | 0 | 0.00 | 0.00%
TA9 | 2026.03.01-2026.05.29 | 50000 | 9 | 8 | 1 | 0 | 6231.00 | 19.02%
TA5 | 2026.03.01-2026.05.29 | 50000 | 8 | 8 | 0 | 0 | 6301.00 | 31.41%
TA4 | 2026.03.01-2026.05.29 | 50000 | 8 | 8 | 0 | 0 | 6301.00 | 45.00%
TA9 | 2026.04.01-2026.05.29 | 100000 | 4 | 3 | 0 | 1 | 4984.00 | 15.21%
TA5 | 2026.04.01-2026.05.29 | 100000 | 2 | 2 | 0 | 0 | 2996.00 | 14.93%
TA4 | 2026.04.01-2026.05.29 | 100000 | 0 | 0 | 0 | 0 | 0.00 | 0.00%
TA9 | 2026.04.01-2026.05.29 | 50000 | 7 | 7 | 0 | 0 | 6606.00 | 20.16%
TA5 | 2026.04.01-2026.05.29 | 50000 | 7 | 7 | 0 | 0 | 6606.00 | 32.93%
TA4 | 2026.04.01-2026.05.29 | 50000 | 7 | 7 | 0 | 0 | 6546.00 | 46.75%
TA9 | 2026.05.01-2026.05.29 | 100000 | 2 | 1 | 0 | 1 | 2729.00 | 8.33%
TA5 | 2026.05.01-2026.05.29 | 100000 | 0 | 0 | 0 | 0 | 0.00 | 0.00%
TA4 | 2026.05.01-2026.05.29 | 100000 | 0 | 0 | 0 | 0 | 0.00 | 0.00%
TA9 | 2026.05.01-2026.05.29 | 50000 | 3 | 2 | 0 | 1 | 2616.00 | 7.99%
TA5 | 2026.05.01-2026.05.29 | 50000 | 3 | 2 | 0 | 1 | 2616.00 | 13.04%
TA4 | 2026.05.01-2026.05.29 | 50000 | 1 | 1 | 0 | 0 | 1156.00 | 8.26%

TA9 HTE Kill:
- TA9 raw has 1 HTE Kill, 70 yen opportunity loss.
- The killed basket is the only TA9 row with distance_to_hardstop_yen < 0 at the executable snapshot.
- The effect ratio is small: 70 / 32297 gross HardStop benefit = 0.22%.
- Count alone should not reject TA9, but raw TA9 still needs safety filtering and priority-slot approval.

TA9 added profit source:
- TA9 rescues 26 HardStop baskets versus TA5 20 and TA4 16.
- TA9 is positive in all six logs and covers both 50k and 100k; TA5 has one zero log and TA4 is 50k-only in this dataset.
- TA9 raw net advantage over TA5 is 12,697 yen; over TA4 is 18,757 yen.

Safe filter search:
- TA9_FILTER_DISTANCE_TO_HARDSTOP_NONNEG: net 32830.00, hardstop 26, HTE Kill 0, TA5 diff 12767.00, overfit medium
- TA9_FILTER_DISTANCE_TO_HTE_LE_NEG500: net 29196.00, hardstop 21, HTE Kill 0, TA5 diff 9133.00, overfit medium-high
- TA9_FILTER_FLOATING_LE_NEG2500: net 26596.00, hardstop 18, HTE Kill 0, TA5 diff 6533.00, overfit medium-high
- TA9_FILTER_FLOATING_LE_NEG3000: net 22095.00, hardstop 13, HTE Kill 0, TA5 diff 2032.00, overfit high
- TA9_FILTER_DISTANCE_HS_0_TO_1500: net 20063.00, hardstop 20, HTE Kill 0, TA5 diff 0.00, overfit medium-high
- TA9_FILTER_SPREAD_LE_070: net 32830.00, hardstop 26, HTE Kill 0, TA5 diff 12767.00, overfit medium

Best safety adjustment:
- TA9_FILTER_DISTANCE_TO_HARDSTOP_NONNEG: distance_to_hardstop_yen >= 0
- Filtered net effect: 32830.00
- HTE Kill: 0
- TA5 advantage: 12767.00
- This uses a realtime candidate-snapshot field and does not reference final_close_reason, basket id, period, or future hardstop time.

Leave-One-Log-Out:
- TA9: min 23166.00, max 30144.00, avg 27300.00, sign reversal False
- TA5: min 13457.00, max 20063.00, avg 16719.17, sign reversal False
- TA4: min 7457.00, max 14003.00, avg 11669.17, sign reversal False

Overlap summary:
- ALL_THREE: count 16, hardstop 16, hte 0, TA9 net 14135.00, TA5 net 14135.00, TA4 net 14003.00
- TA9_ONLY: count 9, hardstop 6, hte 1, TA9 net 12697.00, TA5 net 0.00, TA4 net 0.00
- TA9_TA5_ONLY: count 5, hardstop 4, hte 0, TA9 net 5928.00, TA5 net 5928.00, TA4 net 0.00

Recommendation:
- selected_plan: TA9_SAFE_DISTANCE_TO_HARDSTOP_NONNEG
- decision_class: TA9_DESIGN_REQUIRED
- P1_implementation_allowed: false
- Rationale: TA9 has superior corrected effect and a plausible realtime safety filter, but still needs explicit POST_HEDGE_SLOT / priority-order design approval.
- Fallback: TA5_P1_CANDIDATE if priority-order change risk is not accepted.

Residual risks and data gaps:
- RECOVERY_CLOSE and BASKET_CLOSE counts are zero in the six-log sample, so Kill 0 for those paths must not be over-generalized.
- TA9 safety filter should be parity-logged in runtime because distance_to_hardstop_yen negative/positive sign is a same-tick edge.
- No code/EA logic changes were made in this audit.
