Existing model-mail aggregates provide supplementary context for scenario analysis. No new model run or email delivery is performed by this reference builder.
Korean market
Mail artifact date: 20261022 · Signal date: 20261020
SMTP success marker: — · This does not establish recipient inbox placement.
How to use this with scenarios
For RT1/RT3, check whether observed upside conditions agree with signal direction. For RT5, inspect inverse or weakness indicators. For RT2/RT4/RT6, also check reversal, recovery and volatility conditions. A high inverse score is not automatically a buy signal. Do not pool results from different assets, dates or holding horizons. These references do not automatically alter MiDM choices or B3 probabilities; no combined decision rule has been validated.
Models connected to the mail / observation table
| Role | Model | Inference connection | Timing type | Direction | Lag | Correlation | Source BH q | Source independent EV | Source independent win rate | Strict all-period pass |
|---|---|---|---|---|---|---|---|---|---|---|
| Operational Phase7 | P7__all_market_bcdef_v7__regime_10d_thr10 | Positive coincident; source labels significant; not re-audited here | Leading | Inverse | +4 sessions | -0.196 | 0.6757 | -1.10% | 36.36% | Not passed |
| Phase14-26 reference | P14__14-26_nonft_heavy_balacc_large_nonft2080ti_e60__regime_10d_thr10 | Positive coincident; source labels significant; not re-audited here | Lagging | Inverse | -2 sessions | -0.150 | 0.0262 | 1.16% | 46.84% | Not passed |
| Phase14-23 reference | P14__14-23_nonft_heavy_prf_wide_resnet_nonft2080ti_e60__regime_10d_thr10 | Positive coincident; source labels significant; not re-audited here | Lagging | Inverse | -2 sessions | -0.196 | 0.0262 | 0.36% | 44.40% | Not passed |
Research registry ranking reference
These are ranks copied from the mail, not a newly validated comparison of all current models under matched conditions. Unavailable or excluded models are not treated as that mailing’s daily inference.
| Registry rank | Model | Top N | Inference connection | Source independent EV | Source independent win rate | Source indicator label | Periods passed | Strict all-period pass |
|---|---|---|---|---|---|---|---|---|
| 1 | 40-2-KR-lightgbm_full | 3 | Unavailable | 0.95% | 45.19% | Inverse coincident; statistically unconfirmed in source | 3 | Not passed |
| 2 | 40-10-KR-all_validation_weighted_stack | 3 | Unavailable | 0.98% | 51.96% | Positive coincident; statistically unconfirmed in source | 2 | Not passed |
| 3 | 40-3-KR-mlp_full | 3 | Unavailable | 0.65% | 49.90% | Inverse coincident; source labels significant; not re-audited here | 2 | Not passed |
| 4 | 39-2 scratch__KR__xgboost | 5 | Unavailable | 1.41% | 46.16% | Inverse lagging; source labels significant; not re-audited here | 2 | Not passed |
| 5 | 40-1-KR-xgboost_full | 3 | Unavailable | 2.02% | 52.54% | Inverse coincident; source labels significant; not re-audited here | 1 | Not passed |
| 6 | P32__32-2_resnet | 3 | Unavailable | 1.19% | 54.60% | Inverse leading; statistically unconfirmed in source | 1 | Not passed |
| 7 | P32__32-10_resnet | 3 | Unavailable | 1.70% | 51.11% | Inverse coincident; statistically unconfirmed in source | 1 | Not passed |
| 8 | 40-9-KR-deep_validation_weighted_ensemble | 3 | Unavailable | 0.69% | 51.21% | Positive coincident; statistically unconfirmed in source | 1 | Not passed |
| 9 | P14__14-26_nonft_heavy_balacc_large_nonft2080ti_e60__regime_10d_thr10 | 3 | Unavailable | 1.16% | 46.84% | Inverse lagging; source labels significant; not re-audited here | 1 | Not passed |
| 10 | 40-6-KR-lstm_core | 3 | Unavailable | 0.59% | 46.94% | Inverse coincident; source labels significant; not re-audited here | 1 | Not passed |
Post-recommendation outcomes · 5/20-session aggregates
| Model | Horizon (sessions) | Recommendations | Realized | Open | Entry pending | Realized net win rate (%) | Net trade EV, 30bp (%) | Net excess trade EV, 30bp (%) | Latest tracked signal |
|---|---|---|---|---|---|---|---|---|---|
| P14__14-26_nonft_heavy_balacc_large_nonft2080ti_e60__regime_10d_thr10 | 5 | 156 | 136 | 20 | 0 | 44.85294117647059 | -0.7772751024659589 | 0.9094819630404866 | 20261020 |
| KR_COMBO__P14_14-26_X_14-23__regime_10d_thr10 | 5 | 69 | 59 | 10 | 0 | 49.152542372881356 | -0.5046686716320266 | 0.41166907851773804 | 20261020 |
| P14__14-23_nonft_heavy_prf_wide_resnet_nonft2080ti_e60__regime_10d_thr10 | 5 | 152 | 132 | 20 | 0 | 48.484848484848484 | -1.6001211208050878 | 0.20869965222526385 | 20261020 |
| P7__all_market_bcdef_v7__regime_10d_thr10 | 5 | 879 | 838 | 41 | 0 | 39.97613365155131 | -0.25563000726811463 | -0.7384580126285526 | 20261020 |
| KR_COMBO__P14_14-26_X_14-23__regime_10d_thr10 | 20 | 69 | 26 | 43 | 0 | 34.61538461538461 | -1.9196197192994555 | 0.0384824073338696 | 20261020 |
| P14__14-23_nonft_heavy_prf_wide_resnet_nonft2080ti_e60__regime_10d_thr10 | 20 | 152 | 62 | 90 | 0 | 25.806451612903224 | -5.3434471760790725 | -0.48139064173283674 | 20261020 |
| P14__14-26_nonft_heavy_balacc_large_nonft2080ti_e60__regime_10d_thr10 | 20 | 156 | 64 | 92 | 0 | 32.8125 | -5.2822230259871095 | -1.7885917041624844 | 20261020 |
| P7__all_market_bcdef_v7__regime_10d_thr10 | 20 | 879 | 696 | 183 | 0 | 35.201149425287355 | -1.3083708049774858 | -4.41809480431248 | 20261020 |
Realized-only net win rates and EV are inherited aggregates. Open and entry-pending recommendations are excluded from realized results, creating maturity and sample-composition differences. Net trade EV uses the source’s 30bp round-trip cost. Registry independent EV, simulated NAV and excess returns use different source aggregates; identical periods or execution contracts must not be assumed. BH q and indicator labels are inherited, not re-tested here. Mailing performance does not establish current or per-RT predictive superiority.
US market
Mail artifact date: 20261022 · Signal date: 20261021
SMTP success marker: 2026-10-22T07:14:32.763650+09:00 · This does not establish recipient inbox placement.
How to use this with scenarios
For RT1/RT3, check whether observed upside conditions agree with signal direction. For RT5, inspect inverse or weakness indicators. For RT2/RT4/RT6, also check reversal, recovery and volatility conditions. A high inverse score is not automatically a buy signal. Do not pool results from different assets, dates or holding horizons. These references do not automatically alter MiDM choices or B3 probabilities; no combined decision rule has been validated.
Models connected to the mail / observation table
| Display rank | Model | Top N | Source evaluation period | Source simulated NAV return | Source excess vs SPY | Win rate | EV | Sharpe | MDD | Timing type | Direction | Lag | Lag correlation | Source BH q | All-period pass |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 29-1-ft_transformer_down_reversal_full__ft_transformer | 3 | 20260601~20261021 | 50.68% | 48.24% | 54.39% | 2.25% | 2.54 | -11.56% | Coincident | Inverse | +0 sessions | -0.234 | 0.0262 | Not passed |
| 2 | 29-5-mlp_deep_down_reversal_abs_liq__mlp_deep | 3 | 20260601~20261021 | 34.13% | 31.70% | 54.45% | 1.79% | 1.96 | -17.77% | Lagging | Inverse | -2 sessions | -0.119 | 0.0262 | Not passed |
| 3 | 29-13-lightgbm_balanced_balanced_updown_core__lightgbm_balanced | 3 | 20260601~20261021 | 30.85% | 28.41% | 51.93% | 1.45% | 1.90 | -9.13% | Coincident | Inverse | +0 sessions | -0.330 | 0.0262 | Not passed |
| 4 | 29-5-resnet_down_reversal_abs_liq__resnet | 3 | 20260601~20261021 | 28.93% | 26.50% | 58.51% | 1.66% | 1.52 | -26.03% | Coincident | Inverse | +0 sessions | -0.142 | 0.0262 | Not passed |
| 5 | 29-4-ft_transformer_down_reversal_abs_full__ft_transformer | 3 | 20260601~20261021 | 25.64% | 23.21% | 51.42% | 1.31% | 1.46 | -16.35% | Lagging | Inverse | -2 sessions | -0.171 | 0.0427 | Not passed |
| 6 | 29-1-resnet_down_reversal_full__resnet | 5 | 20260601~20261021 | 22.92% | 20.49% | 51.16% | 1.21% | 1.52 | -13.10% | Coincident | Inverse | +0 sessions | -0.070 | 0.7630 | Not passed |
| 7 | 29-4-mlp_deep_down_reversal_abs_full__mlp_deep | 5 | 20260601~20261021 | 21.92% | 19.49% | 51.80% | 1.19% | 1.34 | -15.93% | Lagging | Inverse | -2 sessions | -0.127 | 0.0902 | Not passed |
| 8 | 29-5-xgboost_down_reversal_abs_liq__xgboost | 5 | 20260601~20261021 | 20.84% | 18.41% | 52.43% | 1.01% | 1.29 | -15.62% | Coincident | Inverse | +0 sessions | -0.211 | 0.0262 | Not passed |
| 9 | 29-13-mlp_deep_balanced_updown_core__mlp_deep | 5 | 20260601~20261021 | 19.93% | 17.50% | 50.21% | 1.01% | 1.52 | -11.29% | Coincident | Inverse | +0 sessions | -0.146 | 0.1471 | Not passed |
| 10 | 29-14-resnet_balanced_updown_rank__resnet | 3 | 20260601~20261021 | 17.37% | 14.93% | 51.77% | 0.96% | 1.15 | -14.81% | Lagging | Positive | -2 sessions | 0.055 | 0.9765 | Not passed |
Research registry ranking reference
These are ranks copied from the mail, not a newly validated comparison of all current models under matched conditions. Unavailable or excluded models are not treated as that mailing’s daily inference.
| Registry rank | Model | Top N | Inference connection | Source independent EV | Source independent win rate | Source indicator label | Periods passed | Strict all-period pass |
|---|---|---|---|---|---|---|---|---|
| 1 | P29__29-2-resnet_down_reversal_core___resnet | 3 | Not included | 1.37% | 55.97% | Inverse lagging; source labels significant; not re-audited here | 0 | Not passed |
| 2 | P29__29-1-resnet_down_reversal_full___resnet | 3 | Included | 1.42% | 55.15% | Inverse coincident; statistically unconfirmed in source | 0 | Not passed |
| 3 | P29__29-1-ft_transformer_down_reversal_full___ft_transformer | 3 | Included | 1.42% | 56.14% | Inverse coincident; source labels significant; not re-audited here | 0 | Not passed |
| 4 | P29__29-13-mlp_deep_balanced_updown_core___mlp_deep | 3 | Included | 1.53% | 54.43% | Inverse coincident; statistically unconfirmed in source | 0 | Not passed |
| 5 | P29__29-1-mlp_deep_down_reversal_full___mlp_deep | 3 | Not included | 1.31% | 55.15% | Inverse coincident; statistically unconfirmed in source | 0 | Not passed |
| 6 | P29__29-1-mlp_wide_down_reversal_full___mlp_wide | 3 | Not included | 1.29% | 55.32% | Inverse lagging; statistically unconfirmed in source | 0 | Not passed |
| 7 | P29__29-13-ft_transformer_balanced_updown_core___ft_transformer | 3 | Not included | 1.51% | 54.76% | Inverse coincident; source labels significant; not re-audited here | 0 | Not passed |
| 8 | P29__29-5-ft_transformer_down_reversal_abs_liq___ft_transformer | 3 | Not included | 1.24% | 54.33% | Inverse coincident; source labels significant; not re-audited here | 0 | Not passed |
| 9 | P29__29-13-lightgbm_balanced_balanced_updown_core___lightgbm_balanced | 3 | Included | 1.17% | 54.76% | Inverse coincident; source labels significant; not re-audited here | 0 | Not passed |
| 10 | P29__29-1-xgboost_down_reversal_full___xgboost | 3 | Not included | 1.28% | 53.99% | Inverse coincident; source labels significant; not re-audited here | 0 | Not passed |
Post-recommendation outcomes · 5/20-session aggregates
| Model | Horizon (sessions) | Recommendations | Realized | Open | Entry pending | Realized net win rate (%) | Net trade EV, 30bp (%) | Net excess trade EV, 30bp (%) | Latest tracked signal |
|---|---|---|---|---|---|---|---|---|---|
| 29-1-ft_transformer_down_reversal_full__ft_transformer | 5 | 212 | 196 | 12 | 3 | 57.6530612244898 | 2.6867391060218453 | 3.152451071867421 | 20261021 |
| 29-3-mlp_deep_down_reversal_rank__mlp_deep | 5 | 5 | 5 | 0 | 0 | 80.0 | 7.004290106188587 | 2.420360151607308 | 20260729 |
| 29-13-mlp_deep_balanced_updown_core__mlp_deep | 5 | 132 | 106 | 20 | 5 | 53.77358490566038 | 2.0128381967420443 | 1.6677199808599255 | 20261021 |
| 29-13-lightgbm_balanced_balanced_updown_core__lightgbm_balanced | 5 | 201 | 186 | 12 | 3 | 52.1505376344086 | 0.7244511044838393 | 1.4458066168030466 | 20261021 |
| 29-5-resnet_down_reversal_abs_liq__resnet | 5 | 127 | 111 | 12 | 3 | 59.45945945945946 | 1.3786626142362484 | 0.6298373873767901 | 20261021 |
| 29-5-mlp_deep_down_reversal_abs_liq__mlp_deep | 5 | 142 | 126 | 12 | 3 | 51.587301587301596 | 0.8510902542296748 | 0.4805117236072608 | 20261021 |
| 29-1-resnet_down_reversal_full__resnet | 5 | 210 | 185 | 20 | 5 | 43.24324324324324 | -0.31799735148678016 | 0.07991007043112702 | 20261021 |
| 29-14-resnet_balanced_updown_rank__resnet | 5 | 25 | 9 | 12 | 3 | 55.55555555555556 | -1.0206842542736636 | -0.39331205229556626 | 20261021 |
| 29-4-mlp_deep_down_reversal_abs_full__mlp_deep | 5 | 29 | 3 | 20 | 5 | 0.0 | -5.158193438017296 | -4.633147563832902 | 20261021 |
| 29-5-xgboost_down_reversal_abs_liq__xgboost | 5 | 30 | 5 | 20 | 5 | 80.0 | 4.886060842314808 | — | 20261021 |
| 29-4-ft_transformer_down_reversal_abs_full__ft_transformer | 5 | 19 | 3 | 12 | 3 | 66.66666666666666 | 0.331307564232971 | — | 20261021 |
| 29-13-mlp_deep_balanced_updown_core__mlp_deep | 20 | 132 | 106 | 20 | 5 | 60.37735849056604 | 5.509976704066473 | 9.64685878474636 | 20261021 |
| 29-3-mlp_deep_down_reversal_rank__mlp_deep | 20 | 5 | 5 | 0 | 0 | 60.0 | 10.470030855384548 | 6.390142154833057 | 20260729 |
| 29-1-ft_transformer_down_reversal_full__ft_transformer | 20 | 212 | 151 | 57 | 3 | 60.264900662251655 | 6.770715880950326 | 5.882604491021041 | 20261021 |
| 29-13-lightgbm_balanced_balanced_updown_core__lightgbm_balanced | 20 | 201 | 140 | 57 | 3 | 52.142857142857146 | 2.0844176913982833 | 4.144588914225267 | 20261021 |
| 29-5-mlp_deep_down_reversal_abs_liq__mlp_deep | 20 | 142 | 81 | 57 | 3 | 58.0246913580247 | 4.134003507588026 | -7.181764338368393 | 20261021 |
| 29-1-resnet_down_reversal_full__resnet | 20 | 210 | 113 | 90 | 5 | 47.78761061946903 | 0.3785329334310288 | -9.187309591258074 | 20261021 |
| 29-4-mlp_deep_down_reversal_abs_full__mlp_deep | 20 | 29 | 3 | 20 | 5 | 66.66666666666666 | -0.5053950141708103 | — | 20261021 |
| 29-5-resnet_down_reversal_abs_liq__resnet | 20 | 127 | 55 | 68 | 3 | 49.09090909090909 | -1.8581067880721702 | — | 20261021 |
| 29-14-resnet_balanced_updown_rank__resnet | 20 | 25 | 9 | 12 | 3 | 44.44444444444444 | 1.4187994591753406 | — | 20261021 |
| 29-4-ft_transformer_down_reversal_abs_full__ft_transformer | 20 | 19 | 0 | 15 | 3 | — | — | — | 20261021 |
| 29-5-xgboost_down_reversal_abs_liq__xgboost | 20 | 30 | 0 | 25 | 5 | — | — | — | 20261021 |
Realized-only net win rates and EV are inherited aggregates. Open and entry-pending recommendations are excluded from realized results, creating maturity and sample-composition differences. Net trade EV uses the source’s 30bp round-trip cost. Registry independent EV, simulated NAV and excess returns use different source aggregates; identical periods or execution contracts must not be assumed. BH q and indicator labels are inherited, not re-tested here. Mailing performance does not establish current or per-RT predictive superiority.
Reference refreshed: 2026-10-22T09:12:57+09:00 · Stored local aggregates only. Mail bodies, stock-level predictions, raw data and generation code are not published.