Results Snapshot

Research-candidate full-run artifact. This page is generated from tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be. It summarizes the durable gold modeling sample and run outputs, not the older bounded access-check snapshot. It is still a research-candidate artifact: final manuscript claims require a clean committed run and author review of the tables and notes.

Introduction

The framing questions have moved to Discussion Q&A. This snapshot is organized as a paper-facing evidence map: data and target construction first, model and evaluation design second, results third, and paper-facing exports and artifact provenance at the end.

The main result tables are in the Results section. Full table and figure provenance is collected in the Appendix section, including source artifacts and claim scopes for paper-facing outputs.

Evidence Map

flowchart LR
  A["Vendor and calendar inputs"] --> B["Bronze / silver caches"]
  B --> C["Gold panel and timing map"]
  C --> D["Leakage and sample gates"]
  D --> E["Benchmark floor and advanced benchmarks"]
  D --> F["ML-tail nested information sets"]
  E --> G["Metrics, DM/MCS, Murphy diagnostics"]
  F --> G
  F --> H["CPA conditional loss-difference diagnostics"]
  G --> I["Tables and figures"]
  H --> I
  I --> J["Generated results snapshot"]

• The left branch binds vendor and calendar inputs into a timestamp-audited gold panel.
• The middle branch compares benchmark floors, advanced econometric benchmarks, and ML-tail forecasts on registered loss units.
• The right branch separates headline nested information sets, restricted model-family comparisons, unconditional DM/MCS inference, CPA diagnostics, and supporting figures.

Materials: Data And Target

Run Metadata

Field	Value
Run ID	tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be
Artifact root	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be
Claim level	research_candidate
Requested window	['2016-07-19', '2026-05-04']
Combined clean start	2018-06-20
Gold panel dates	2016-07-19 to 2026-05-01
Forecast sample dates	2018-06-20 to 2026-05-01 (1661 rows)
Git commit	0c14f1bea7cf5a604b313cfb356fc0d75b490774
Git dirty	False
FRED vintage safe	False

• combined_clean_start is the modeling lower bound; dates before it remain audit history rather than forecast evidence.
• git_dirty is recorded so dirty runs can be rejected before manuscript tables are frozen.
• fred_vintage_safe=False is an explicit limitation: FRED data are current historical values with conservative release lag, not real-time vintage observations.

Target Distribution And Tail Diagnostics

• These diagnostics are computed from the raw clean settlement-to-open target gap_t; left loss is -gap_t, and right loss is gap_t.
• The purpose is to show why the dependent variable is a tail-risk object before comparing VaR/ES forecasts.
• Positive tail-shape estimates, heavy empirical tails, and upward mean-excess patterns are empirical support for using heavy-tail approximations such as POT-GPD; they are not a finite-sample proof of Frechet max-domain attraction.
• Raw target diagnostics motivate VaR/ES and EVT modeling. They do not validate LightGBM+EVT forecasts; forecast validity must be read from standardized residual-loss EVT diagnostics and out-of-sample VaR/ES backtests.

Target Summary

Measure	Value
Clean forecast observations	1661
Date range	2018-06-20 to 2026-05-01
Mean gap	0.000522 log (+0.05%)
Standard deviation	0.010649 log (+1.07%)
Skewness	-0.0336644
Excess kurtosis	13.1331
1% quantile	-0.028697 log (-2.83%)
5% quantile	-0.014868 log (-1.48%)
Median	0.000890 log (+0.09%)
95% quantile	0.014344 log (+1.44%)
99% quantile	0.027361 log (+2.77%)
Max drawdown gap	-0.087513 log (-8.38%) on 2020-03-13
Max upside gap	0.096937 log (+10.18%) on 2025-04-10
Jarque-Bera p-value	0
Jarque-Bera statistic	11972.7

Raw-Tail EVT Diagnostics

Tail	Threshold probability	Threshold	Exceedances	Mean excess	GPD xi	GPD scale	Hill xi
left_tail_loss	0.900	0.0104713	166	0.00809322	0.204603	0.00641777	0.476705
left_tail_loss	0.925	0.0124096	125	0.00847607	0.234934	0.00648125	0.433857
left_tail_loss	0.950	0.0148675	83	0.00960529	0.212263	0.00755758	0.418714
left_tail_loss	0.975	0.0210536	42	0.0102367	0.446797	0.00607498	0.331429
left_tail_loss	0.990	0.0286967	17	0.0139643	0.284419	0.0101838	0.34578
right_tail_loss	0.900	0.0108676	166	0.00720714	0.346438	0.0047221	0.408732
right_tail_loss	0.925	0.0124484	125	0.00772818	0.426386	0.00457968	0.384777
right_tail_loss	0.950	0.0143435	83	0.00924099	0.42009	0.00558805	0.398242
right_tail_loss	0.975	0.0184656	42	0.0122865	0.407723	0.00773658	0.419593
right_tail_loss	0.990	0.0273614	17	0.0165888	0.191275	0.0135495	0.404539
absolute_gap	0.900	0.0146899	166	0.00934487	0.331036	0.0063527	0.402974
absolute_gap	0.925	0.016469	125	0.01035	0.304284	0.00729168	0.402961
absolute_gap	0.950	0.0196277	83	0.0116699	0.312179	0.00818106	0.387247
absolute_gap	0.975	0.0259904	42	0.0141731	0.297623	0.0102021	0.36961
absolute_gap	0.990	0.0363397	17	0.018341	0.213979	0.0147211	0.357836

• The GPD threshold table is computed on raw left loss, raw right loss, and the absolute gap; it should not be read as a forecast-model diagnostic.
• The Hill and GPD shape estimates are deliberately reported over multiple thresholds because tail-index estimates are sensitive in samples of this length.

Target Distribution Figures

Figure	Tail side	Source	Claim scope	Docs file
target_gap_histogram_density	target_distribution	panel/modeling_panel.parquet	target_distribution_motivation_not_forecast_validation	figures/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/target_gap_histogram_density.png
target_loss_qq_left_tail	left_tail	panel/modeling_panel.parquet	target_distribution_motivation_not_forecast_validation	figures/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/target_loss_qq_left_tail.png
target_loss_qq_right_tail	right_tail	panel/modeling_panel.parquet	target_distribution_motivation_not_forecast_validation	figures/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/target_loss_qq_right_tail.png
target_log_survival	left_right_target_distribution	panel/modeling_panel.parquet	target_distribution_motivation_not_forecast_validation	figures/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/target_log_survival.png
target_mean_excess	left_right_target_distribution	panel/modeling_panel.parquet	target_distribution_motivation_not_forecast_validation	figures/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/target_mean_excess.png
target_hill_plot	left_right_target_distribution	panel/modeling_panel.parquet	target_distribution_motivation_not_forecast_validation	figures/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/target_hill_plot.png

Gold Panel Construction

Measure	Value
Gold modeling rows	2395
Gold columns	1384
Target-audit rows	2395
Clean target rows	2199
Forecast-sample rows	1661
Rows before combined clean start	412
Target-not-clean rows	196
Mapping excluded rows	126

Target audit reason	Rows
None	2199
roll_sq_excluded	195
missing_reference_price	1

• The cache lower bound is 2016-07-19, but XLC/core predictor coverage pushes the actual forecast sample to the combined clean start.
• Target exclusion is explicit: roll/SQ windows and the single missing reference price are carried as audit evidence, not silently dropped.
• The forecast-sample reason column makes the sample boundary reproducible row by row.

Calendar And Timing Map

Measure	Value
Normal trading mappings	2268
U.S./Japan desync mappings	127
NYSE early-close mappings	32
EDT rows	1551
EST rows	844

• The map covers EST/EDT, early closes, U.S./Japan holiday desynchronization, and normal trading alignments.
• Desync rows are not treated as normal forecast rows.
• The timing map is part of the leakage-bound gold artifact, not ad hoc evaluation logic.

Feature Coverage

Source family	Block	Features	Mean missing	Max missing
Asia proxy	Asia proxy	10	0.000%	0.000%
Asia proxy options	Asia proxy	7	49.067%	49.067%
cboe_volatility	fred_core	2	0.000%	0.000%
fred_core	fred_core	9	0.000%	0.000%
FRED credit enriched	FRED credit enriched	4	61.890%	61.890%
fx_core	fx_core	2	0.000%	0.000%
JP history	JP only	23	0.000%	0.000%
JP proxy	JP proxy	8	0.000%	0.000%
JP proxy options	JP proxy	13	54.675%	68.272%
J-Quants N225 options	JP only	14	1.135%	5.298%
massive_daily	US core	40	0.002%	0.060%
massive_minute	Asia proxy	60	0.000%	0.000%
massive_minute	JP proxy	24	0.341%	4.094%
massive_minute	US late session	84	0.000%	0.000%
massive_optional	massive_optional	2	0.000%	0.000%
unknown	unknown	2	0.000%	0.000%
US core options	US core	21	49.067%	49.067%

• U.S. core, proxy ETFs, minute late-session features, CBOE VIX, FRED rates, FRED H.10 FX, and any audit-gated options-risk fields are separated by source family and block.
• Credit-spread FRED features are enriched/optional and visibly late-starting, so they do not move the core clean start.
• Feature coverage should be read together with the leakage summary; high coverage alone is not enough without timestamp validity.

Leakage Audit

Field	Value
Status	pass_with_warnings
Rows audited	730475
Failures	0
Warnings	680103
Panel row count	2395
Panel signature seed	42
Panel signature	63ea2acb4baad92e9cb757fc661e47e8852e1f9b7ef78714a2a8eb564417da13

• Zero failures means no audited row violated the hard timestamp invariant.
• Warnings are retained because they identify conservative-lag or missing-feature situations that may matter for interpretation.
• The panel signature is deterministic and binds the leakage check to the current gold panel/config.

Methods: Model Configuration And Evaluation

Pipeline Structure

Step	Layer	Purpose
1	Vendor and calendar sources	Pull or read J-Quants, Massive, FRED, CBOE, and exchange-calendar inputs.
2	Bronze and silver cache	Preserve typed vendor/cache rows, then normalize point-in-time research features.
3	Gold modeling panel	Join targets, calendar map, feature coverage, and leakage-bound signatures.
4	Leakage and coverage gates	Enforce timestamp ordering and sample eligibility before evaluation.
5	Benchmark floor and ML-tail registry	Run target-history/econometric floors and LightGBM tail-model families.
6	Metrics, inference, diagnostics	Build loss matrices, DM/MCS/Murphy diagnostics, stress windows, and result matrix artifacts.
7	Results snapshot	Summarize run-specific evidence and claim boundaries for reader review.

• Data-access and cache artifacts live under data/bronze and data/silver.
• Durable modeling evidence lives under data/gold; forecast/evaluation/reporting read from gold and reports.
• Run-specific forecasts, metrics, diagnostics, and LaTeX tables live under reports/runs/<run_id>.

Model And Evaluation Protocol

• The registered risk level is tail_level = 0.95; the nominal VaR exception rate is 5%.
• Benchmarks use target-history information only. ML-tail models add predictors through fixed nested information sets.
• Most specifications use expanding pre-forecast training histories. The rolling-quantile benchmark is the designed exception and uses the most recent 1,000 clean observations.
• LightGBM hyperparameters are held fixed across information sets and refit dates; the snapshot reports model-family evidence rather than tuning-search evidence.
• DM/MCS inference is read on average across the unconditional evaluation sample. CPA is read as a conditional loss-difference diagnostic, not as a forecasting model.

Results And Discussion

Main result tables

Benchmark Suite

Status: completed; forecast rows: 21146; metric rows: 12; failures: 0.

Benchmark layer	Status	Forecast rows	Diagnostic rows	Failures	How to read it
floor	completed	7932	12	0	Implemented benchmark evidence for target-history and econometric floor models.
advanced	completed_nonblocking	13214	786	0	Implemented nonblocking advanced benchmark forecasts; review with common-sample gates.

Model	Information set	Tail side	Rows	VaR breach rate	Exceptions	Mean quantile loss	Mean FZ loss
ewma_vol_scaled	Target history	left_tail	661	5.295%	35	0.00139273	-3.65402
ewma_vol_scaled	Target history	right_tail	661	4.539%	30	0.00131562	-3.69642
garch_t	Target history	left_tail	661	6.354%	42	0.00135048	-3.70463
garch_t	Target history	right_tail	661	3.933%	26	0.00122176	-3.78739
gjr_garch_evt	Target history	left_tail	661	5.598%	37	0.00133528	-3.73239
gjr_garch_evt	Target history	right_tail	661	5.295%	35	0.00117898	-3.80163
gjr_garch_t	Target history	left_tail	661	6.505%	43	0.0013401	-3.7092
gjr_garch_t	Target history	right_tail	661	3.782%	25	0.00117343	-3.81523
historical_quantile	Target history	left_tail	661	6.051%	40	0.00147767	-3.5142
historical_quantile	Target history	right_tail	661	6.354%	42	0.0014632	-3.43894
rolling_quantile	Target history	left_tail	661	6.051%	40	0.00147769	-3.5066
rolling_quantile	Target history	right_tail	661	6.657%	44	0.00147041	-3.43407

• Benchmark floor rows set the target-history/econometric floor that ML models should be interpreted against.
• Advanced benchmark families are nonblocking; rows with valid forecasts are empirical evidence subject to the same sample and inference gates, while unavailable rows remain diagnostics.
• The table is not a leaderboard by itself; coverage, exception counts, quantile loss, and FZ loss must be read together.
• Common-sample rows are reported directly so readers can see the effective evidence size.

ML-Tail Headline Ladder

Status: completed LGBM ML-tail models; implemented models: LGBM direct quantile, LGBM location-scale empirical, LGBM POT-GPD plain MLE, LGBM POT-GPD capped MLE, LGBM POT-GPD EVI shrink, LGBM POT-GPD EI-weighted, LGBM POT-GPD stabilized, LGBM conditional q90 POT-GPD plain MLE, LGBM conditional q90 POT-GPD stabilized, LGBM median/MAD POT-GPD plain MLE, LGBM median/MAD POT-GPD stabilized, LGBM median/IQR POT-GPD plain MLE; forecast rows: 46892; failures: 0.

Model	Information set	Tail side	Rows	VaR breach rate	Exceptions	Mean quantile loss	Mean FZ loss
LGBM direct quantile	JP only	left_tail	550	7.636%	42	0.0013698	-3.51648
LGBM direct quantile	JP + US close core	left_tail	550	11.091%	61	0.00115871	-3.66765
LGBM direct quantile	JP + US close core + JP proxy	left_tail	550	11.636%	64	0.00113639	-3.76063
LGBM direct quantile	JP + US close core + JP proxy + Asia proxy	left_tail	550	12.000%	66	0.00116524	-3.70128
LGBM direct quantile	JP only	right_tail	550	9.636%	53	0.00131328	-3.54912
LGBM direct quantile	JP + US close core	right_tail	550	12.364%	68	0.00131598	-3.3858
LGBM direct quantile	JP + US close core + JP proxy	right_tail	550	12.000%	66	0.00129244	-3.45045
LGBM direct quantile	JP + US close core + JP proxy + Asia proxy	right_tail	550	12.727%	70	0.00127406	-3.51453

• This headline table remains strict and reports only ML-tail rows that pass the registered common-sample and coverage gates.
• Location-scale empirical, plain POT-GPD, and stabilized POT-GPD are headline candidates only after their valid OOS coverage, standardized-loss, exceedance, and ES-validity gates pass.
• Differences across information blocks are candidate forecast evidence only after the common-sample, coverage, and inference diagnostics are reviewed.
• Coverage review: 8/8 headline rows differ from the expected breach rate by more than 2.5 percentage points, so quantile/FZ loss differences alone must not be read as forecast improvement.

ML-tail artifact relationship

Artifact	Rows	Role	Claim boundary
ml_tail_metrics.parquet	8	Headline ML-tail nested-information-set comparison	Eligible for headline discussion after author review.
ml_tail_metrics_per_model.parquet	88	Per-model diagnostics on each model's own valid OOS rows	Not a cross-model comparison and not a replacement headline table.
ml_tail_result_matrix.parquet	408	Restricted common-sample VaR-only and VaR-ES comparisons	Restricted evidence; direct quantile rows here are comparison anchors.

• ml_tail_metrics.parquet is the headline nested-information-set artifact. It contains the ML-tail rows that survived the strict common-sample gate in this run.
• ml_tail_metrics_per_model.parquet reports each implemented ML-tail model on its own valid OOS rows; it is useful for debugging coverage but is not a cross-model comparison table.
• ml_tail_result_matrix.parquet creates restricted common samples for VaR-only and VaR-ES comparisons across model families and within-model information-set increments.

Other result tables

All-model diagnostic scan

Suite	Model	Information set	Metric rows	OOS N mean+-sd	Breach mean+-sd	Abs cov err mean+-sd	Q loss mean+-sd	FZ loss mean+-sd	ES severity mean+-sd
benchmark_advanced	ald_taylor_var_es_asymmetric_slope	Target history	2	661 +/- 0	5.825% +/- 0.107%	0.825% +/- 0.107%	0.00127576 +/- 8.04588e-05	-3.7264 +/- 0.0333965	0.00869969 +/- 0.000405801
benchmark_advanced	ald_taylor_var_es_sav	Target history	2	661 +/- 0	5.522% +/- 0.535%	0.522% +/- 0.535%	0.00128271 +/- 3.8144e-05	-3.73056 +/- 0.00716924	0.00886176 +/- 0.000129554
benchmark_advanced	care_expectile_asymmetric_slope	Target history	2	661 +/- 0	7.867% +/- 1.498%	2.867% +/- 1.498%	0.00127718 +/- 8.34e-05	-3.70044 +/- 0.0127067	0.0073354 +/- 0.00110666
benchmark_advanced	care_expectile_sav	Target history	2	661 +/- 0	6.732% +/- 0.749%	1.732% +/- 0.749%	0.00129396 +/- 4.30045e-05	-3.68574 +/- 0.0442191	0.00850544 +/- 0.000289762
benchmark_advanced	caviar_asymmetric_slope	Target history	2	658.5 +/- 0.707107	5.922% +/- 0.208%	0.922% +/- 0.208%	0.00127214 +/- 7.44272e-05	-3.74253 +/- 0.0317651	0.00850709 +/- 0.000182887
benchmark_advanced	caviar_sav	Target history	2	660.5 +/- 0.707107	5.451% +/- 0.648%	0.458% +/- 0.637%	0.001287 +/- 4.0599e-05	-3.7291 +/- 0.0043629	0.00935712 +/- 0.000659042
benchmark_advanced	direct_fz_loss_asymmetric_slope	Target history	2	661 +/- 0	5.825% +/- 0.107%	0.825% +/- 0.107%	0.00127576 +/- 8.04588e-05	-3.7264 +/- 0.0333965	0.00869969 +/- 0.000405801
benchmark_advanced	direct_fz_loss_sav	Target history	2	661 +/- 0	5.522% +/- 0.535%	0.522% +/- 0.535%	0.00128271 +/- 3.8144e-05	-3.73056 +/- 0.00716924	0.00886176 +/- 0.000129554
benchmark_advanced	gas_t_location_scale	Target history	2	661 +/- 0	5.522% +/- 1.391%	0.983% +/- 0.738%	0.00128515 +/- 5.79748e-05	-3.73231 +/- 0.0419237	0.00939352 +/- 0.00130429
benchmark_advanced	gas_t_pot_gpd	Target history	2	661 +/- 0	5.446% +/- 0.000%	0.446% +/- 0.000%	0.00128952 +/- 4.66132e-05	-3.74198 +/- 0.00836893	0.00945829 +/- 0.000793522
benchmark_floor	ewma_vol_scaled	Target history	2	661 +/- 0	4.917% +/- 0.535%	0.378% +/- 0.118%	0.00135418 +/- 5.45215e-05	-3.67522 +/- 0.0299802	0.00916441 +/- 0.000276157
benchmark_floor	garch_t	Target history	2	661 +/- 0	5.144% +/- 1.712%	1.210% +/- 0.203%	0.00128612 +/- 9.10169e-05	-3.74601 +/- 0.0585218	0.00955344 +/- 0.00130639
benchmark_floor	gjr_garch_evt	Target history	2	661 +/- 0	5.446% +/- 0.214%	0.446% +/- 0.214%	0.00125713 +/- 0.000110519	-3.76701 +/- 0.0489599	0.00888388 +/- 6.5351e-05
benchmark_floor	gjr_garch_t	Target history	2	661 +/- 0	5.144% +/- 1.926%	1.362% +/- 0.203%	0.00125676 +/- 0.000117851	-3.76221 +/- 0.0749739	0.00940623 +/- 0.00137493
benchmark_floor	historical_quantile	Target history	2	661 +/- 0	6.203% +/- 0.214%	1.203% +/- 0.214%	0.00147044 +/- 1.02336e-05	-3.47657 +/- 0.0532175	0.0122667 +/- 0.000676839
benchmark_floor	rolling_quantile	Target history	2	661 +/- 0	6.354% +/- 0.428%	1.354% +/- 0.428%	0.00147405 +/- 5.15148e-06	-3.47034 +/- 0.0512931	0.0120561 +/- 0.000415452
ml_tail	LGBM POT-GPD EI-weighted	JP only	2	475 +/- 0	5.158% +/- 1.340%	0.947% +/- 0.223%	0.00143629 +/- 0.000161593	-3.53387 +/- 0.0716907	0.00997476 +/- 0.000842096
ml_tail	LGBM POT-GPD EI-weighted	JP + US close core	2	475 +/- 0	5.474% +/- 0.298%	0.474% +/- 0.298%	0.00113815 +/- 1.72152e-06	-3.8776 +/- 0.0952207	0.00862388 +/- 0.000602678
ml_tail	LGBM POT-GPD EI-weighted	JP + US close core + JP proxy	2	461 +/- 0	4.664% +/- 0.460%	0.336% +/- 0.460%	0.00106108 +/- 3.15359e-05	-3.94905 +/- 0.118796	0.00901382 +/- 0.000625581
ml_tail	LGBM POT-GPD EI-weighted	JP + US close core + JP proxy + Asia proxy	2	461 +/- 0	4.881% +/- 1.687%	1.193% +/- 0.169%	0.00108099 +/- 9.24507e-05	-3.82035 +/- 0.443333	0.00908733 +/- 0.000754414
ml_tail	LGBM POT-GPD EVI shrink	JP only	2	475 +/- 0	5.158% +/- 1.340%	0.947% +/- 0.223%	0.00143653 +/- 0.000161814	-3.53388 +/- 0.0721693	0.00997694 +/- 0.000849243
ml_tail	LGBM POT-GPD EVI shrink	JP + US close core	2	475 +/- 0	5.474% +/- 0.298%	0.474% +/- 0.298%	0.00113825 +/- 1.7499e-06	-3.87771 +/- 0.0950426	0.00862244 +/- 0.000602362
ml_tail	LGBM POT-GPD EVI shrink	JP + US close core + JP proxy	2	461 +/- 0	4.664% +/- 0.460%	0.336% +/- 0.460%	0.00106093 +/- 3.15722e-05	-3.9491 +/- 0.118256	0.0090113 +/- 0.000624728
ml_tail	LGBM POT-GPD EVI shrink	JP + US close core + JP proxy + Asia proxy	2	461 +/- 0	4.881% +/- 1.687%	1.193% +/- 0.169%	0.00108099 +/- 9.24771e-05	-3.82087 +/- 0.445122	0.00908606 +/- 0.000754356
ml_tail	LGBM POT-GPD capped MLE	JP only	2	475 +/- 0	5.263% +/- 1.489%	1.053% +/- 0.372%	0.00143812 +/- 0.000161728	-3.53127 +/- 0.0782643	0.00982267 +/- 0.00111114
ml_tail	LGBM POT-GPD capped MLE	JP + US close core	2	475 +/- 0	5.474% +/- 0.298%	0.474% +/- 0.298%	0.0011394 +/- 2.8868e-06	-3.88266 +/- 0.0929846	0.00858198 +/- 0.000546112
ml_tail	LGBM POT-GPD capped MLE	JP + US close core + JP proxy	2	461 +/- 0	4.772% +/- 0.307%	0.228% +/- 0.307%	0.00105996 +/- 3.17773e-05	-4.0774 +/- 0.068081	0.00874405 +/- 0.000328565
ml_tail	LGBM POT-GPD capped MLE	JP + US close core + JP proxy + Asia proxy	2	461 +/- 0	4.881% +/- 1.380%	0.976% +/- 0.169%	0.00108202 +/- 9.21118e-05	-3.82349 +/- 0.447209	0.0089689 +/- 0.000163428
ml_tail	LGBM POT-GPD plain MLE	JP only	2	475 +/- 0	5.263% +/- 1.489%	1.053% +/- 0.372%	0.00143812 +/- 0.000161733	-3.53128 +/- 0.0782704	0.00982267 +/- 0.00111104
ml_tail	LGBM POT-GPD plain MLE	JP + US close core	2	475 +/- 0	5.368% +/- 0.149%	0.368% +/- 0.149%	0.00113983 +/- 3.50431e-06	-3.86972 +/- 0.0746819	0.008734 +/- 0.000331215
ml_tail	LGBM POT-GPD plain MLE	JP + US close core + JP proxy	2	461 +/- 0	4.772% +/- 0.307%	0.228% +/- 0.307%	0.00105996 +/- 3.17781e-05	-4.0776 +/- 0.0682899	0.00874401 +/- 0.000328547
ml_tail	LGBM POT-GPD plain MLE	JP + US close core + JP proxy + Asia proxy	2	461 +/- 0	4.881% +/- 1.380%	0.976% +/- 0.169%	0.00108132 +/- 9.31224e-05	-3.82518 +/- 0.448947	0.0089437 +/- 0.000138481
ml_tail	LGBM POT-GPD stabilized	JP only	2	475 +/- 0	5.158% +/- 1.340%	0.947% +/- 0.223%	0.00143629 +/- 0.000161593	-3.53387 +/- 0.0716907	0.00997476 +/- 0.000842096
ml_tail	LGBM POT-GPD stabilized	JP + US close core	2	475 +/- 0	5.474% +/- 0.298%	0.474% +/- 0.298%	0.00113815 +/- 1.72152e-06	-3.8776 +/- 0.0952207	0.00862388 +/- 0.000602678
ml_tail	LGBM POT-GPD stabilized	JP + US close core + JP proxy	2	461 +/- 0	4.664% +/- 0.460%	0.336% +/- 0.460%	0.00106108 +/- 3.15359e-05	-3.94905 +/- 0.118796	0.00901382 +/- 0.000625581
ml_tail	LGBM POT-GPD stabilized	JP + US close core + JP proxy + Asia proxy	2	461 +/- 0	4.881% +/- 1.687%	1.193% +/- 0.169%	0.00108099 +/- 9.24507e-05	-3.82035 +/- 0.443333	0.00908733 +/- 0.000754414
ml_tail	LGBM conditional q90 POT-GPD plain MLE	JP only	1	28 +/- 0	3.571% +/- 0.000%	1.429% +/- 0.000%	0.000990703 +/- 0	-3.94912 +/- 0	0.00764752 +/- 0
ml_tail	LGBM conditional q90 POT-GPD plain MLE	JP + US close core	1	36 +/- 0	5.556% +/- 0.000%	0.556% +/- 0.000%	0.000474017 +/- 0	-4.77857 +/- 0	0.000673893 +/- 0
ml_tail	LGBM conditional q90 POT-GPD plain MLE	JP + US close core + JP proxy	1	23 +/- 0	4.348% +/- 0.000%	0.652% +/- 0.000%	0.000376538 +/- 0	-4.94318 +/- 0	6.58221e-05 +/- 0
ml_tail	LGBM conditional q90 POT-GPD plain MLE	JP + US close core + JP proxy + Asia proxy	1	19 +/- 0	5.263% +/- 0.000%	0.263% +/- 0.000%	0.000403692 +/- 0	-4.88732 +/- 0	0.000421166 +/- 0
ml_tail	LGBM conditional q90 POT-GPD stabilized	JP only	1	28 +/- 0	3.571% +/- 0.000%	1.429% +/- 0.000%	0.000983391 +/- 0	-3.95282 +/- 0	0.00871201 +/- 0
ml_tail	LGBM conditional q90 POT-GPD stabilized	JP + US close core	1	36 +/- 0	5.556% +/- 0.000%	0.556% +/- 0.000%	0.00047967 +/- 0	-4.75451 +/- 0	0.000945934 +/- 0
ml_tail	LGBM conditional q90 POT-GPD stabilized	JP + US close core + JP proxy	1	23 +/- 0	13.043% +/- 0.000%	8.043% +/- 0.000%	0.000390424 +/- 0	-4.91963 +/- 0	0.000151402 +/- 0
ml_tail	LGBM conditional q90 POT-GPD stabilized	JP + US close core + JP proxy + Asia proxy	1	19 +/- 0	5.263% +/- 0.000%	0.263% +/- 0.000%	0.000396978 +/- 0	-4.90251 +/- 0	0.00029148 +/- 0
ml_tail	LGBM direct quantile	JP only	2	577 +/- 0	8.406% +/- 1.348%	3.406% +/- 1.348%	0.0013192 +/- 6.02852e-05	-3.55586 +/- 0.0487363	0.00840518 +/- 0.0010658
ml_tail	LGBM direct quantile	JP + US close core	2	577 +/- 0	11.265% +/- 0.735%	6.265% +/- 0.735%	0.00121287 +/- 8.41547e-05	-3.56054 +/- 0.153657	0.00672302 +/- 0.000475093
ml_tail	LGBM direct quantile	JP + US close core + JP proxy	2	550 +/- 0	11.818% +/- 0.257%	6.818% +/- 0.257%	0.00121441 +/- 0.000110342	-3.60554 +/- 0.219328	0.0065262 +/- 0.000939602
ml_tail	LGBM direct quantile	JP + US close core + JP proxy + Asia proxy	2	550 +/- 0	12.364% +/- 0.514%	7.364% +/- 0.514%	0.00121965 +/- 7.69474e-05	-3.6079 +/- 0.132049	0.00621849 +/- 0.000425778
ml_tail	LGBM location-scale empirical	JP only	2	515 +/- 0	4.951% +/- 1.236%	0.874% +/- 0.069%	0.00139893 +/- 0.000149954	-3.56351 +/- 0.0782675	0.00974933 +/- 0.000687947
ml_tail	LGBM location-scale empirical	JP + US close core	2	515 +/- 0	5.243% +/- 0.275%	0.243% +/- 0.275%	0.00109504 +/- 4.20892e-06	-3.93514 +/- 0.0815901	0.00846113 +/- 0.000659386
ml_tail	LGBM location-scale empirical	JP + US close core + JP proxy	2	501 +/- 0	4.790% +/- 0.282%	0.210% +/- 0.282%	0.0010147 +/- 3.62699e-05	-4.16259 +/- 0.0464425	0.00818853 +/- 0.000719827
ml_tail	LGBM location-scale empirical	JP + US close core + JP proxy + Asia proxy	2	501 +/- 0	4.890% +/- 1.270%	0.898% +/- 0.155%	0.00103545 +/- 8.32502e-05	-3.9035 +/- 0.436634	0.00837342 +/- 0.000141066
ml_tail	LGBM median/IQR POT-GPD plain MLE	JP only	2	577 +/- 0	3.553% +/- 0.368%	1.447% +/- 0.368%	0.00126942 +/- 0.000132383	-3.75884 +/- 0.125474	0.01132 +/- 0.000407257
ml_tail	LGBM median/IQR POT-GPD plain MLE	JP + US close core	2	577 +/- 0	5.633% +/- 0.613%	0.633% +/- 0.613%	0.00105469 +/- 9.214e-05	-4.01647 +/- 0.119185	0.00813274 +/- 0.00076531
ml_tail	LGBM median/IQR POT-GPD plain MLE	JP + US close core + JP proxy	2	550 +/- 0	5.273% +/- 0.257%	0.273% +/- 0.257%	0.00101342 +/- 0.000125138	-3.66986 +/- 0.724362	0.0080073 +/- 0.00158447
ml_tail	LGBM median/IQR POT-GPD plain MLE	JP + US close core + JP proxy + Asia proxy	2	550 +/- 0	5.455% +/- 1.029%	0.727% +/- 0.643%	0.000994617 +/- 0.000128018	-4.09428 +/- 0.271202	0.00729911 +/- 0.000596698
ml_tail	LGBM median/MAD POT-GPD plain MLE	JP only	2	475 +/- 0	5.684% +/- 0.595%	0.684% +/- 0.595%	0.00136058 +/- 0.000164193	-3.69296 +/- 0.127491	0.00951711 +/- 0.00170501
ml_tail	LGBM median/MAD POT-GPD plain MLE	JP + US close core	2	475 +/- 0	7.368% +/- 0.298%	2.368% +/- 0.298%	0.00119332 +/- 7.76125e-05	-3.87811 +/- 0.125271	0.00878918 +/- 0.00120023
ml_tail	LGBM median/MAD POT-GPD plain MLE	JP + US close core + JP proxy	2	461 +/- 0	7.050% +/- 0.767%	2.050% +/- 0.767%	0.00113052 +/- 0.000118817	-3.96233 +/- 0.245708	0.00844573 +/- 0.00109905
ml_tail	LGBM median/MAD POT-GPD plain MLE	JP + US close core + JP proxy + Asia proxy	2	461 +/- 0	6.833% +/- 1.687%	1.833% +/- 1.687%	0.00111197 +/- 0.00013827	-3.97059 +/- 0.333302	0.00847972 +/- 0.00042278
ml_tail	LGBM median/MAD POT-GPD stabilized	JP only	2	475 +/- 0	5.684% +/- 0.893%	0.684% +/- 0.893%	0.0013628 +/- 0.000166391	-3.69421 +/- 0.126373	0.00964987 +/- 0.00219316
ml_tail	LGBM median/MAD POT-GPD stabilized	JP + US close core	2	475 +/- 0	7.474% +/- 0.149%	2.474% +/- 0.149%	0.00119314 +/- 7.97593e-05	-3.88149 +/- 0.125503	0.00870034 +/- 0.00137456
ml_tail	LGBM median/MAD POT-GPD stabilized	JP + US close core + JP proxy	2	461 +/- 0	7.050% +/- 0.767%	2.050% +/- 0.767%	0.00113083 +/- 0.000117246	-3.96339 +/- 0.24094	0.00846168 +/- 0.00108002
ml_tail	LGBM median/MAD POT-GPD stabilized	JP + US close core + JP proxy + Asia proxy	2	461 +/- 0	7.158% +/- 1.227%	2.158% +/- 1.227%	0.00111244 +/- 0.000139018	-3.96212 +/- 0.347415	0.00810695 +/- 0.000996253

• This table joins benchmark_metrics_per_model.parquet and ml_tail_metrics_per_model.parquet so all benchmark and LGBM tail-model variants are visible in one place.
• Mean and standard deviation are computed across registered metric rows for the same suite/model/information-set configuration; for most rows this summarizes left- and right-tail metrics.
• It is a diagnostic scan, not the formal cross-model comparison table. Cross-model claims still require common-sample result-matrix, DM, and MCS evidence because valid dates and model gates can differ.

Restricted Common-Sample Result Matrix

Family	Axis	Loss	Rows	Common N	Date range	Joint exceptions
nested information sets	information_set_increment	var_coverage	88	0 to 550	2023-03-24 to 2026-05-01	0 to 87
nested information sets	information_set_increment	var_es_fz_loss	88	0 to 550	2023-03-24 to 2026-05-01	0 to 87
nested information sets	information_set_increment	var_quantile_loss	88	0 to 550	2023-03-24 to 2026-05-01	0 to 87
tail_model_family	model_family	var_coverage	48	0 to 28	None to None	0 to 3
tail_model_family	model_family	var_es_fz_loss	48	0 to 28	None to None	0 to 3
tail_model_family	model_family	var_quantile_loss	48	0 to 28	None to None	0 to 3

• The result matrix is the right place to compare direct quantile, location-scale empirical, plain POT-GPD, stabilized POT-GPD, and ablation variants on their restricted common dates.
• It separates VaR-only losses from VaR-ES joint scoring, so VaR-only claims are not confused with ES claims.
• Restricted direct-quantile performance is only a comparison anchor for the tail-model family; it does not replace the headline direct-quantile evidence.
• DM and MCS records are emitted only where registered row-count and exception-count gates pass; otherwise the result matrix remains descriptive.

Stress And Diagnostic Windows

Suite	Rows	Window labels
benchmark	134	loss_top_decile
ml_tail	220	loss_top_decile, vix_top_decile

• Stress windows identify high-loss or high-volatility subsamples for two-sided risk diagnostics.
• These rows use reproducible full-sample classifiers in this first pass, so they should be described as diagnostics rather than a live stress classifier.
• They are useful for finding whether model behavior changes in difficult regimes before writing manuscript discussion.

Results interpretation and claim boundaries

Data and timing audit

• The gold timing map covers 2016-07-19 to 2026-05-01 and the combined clean start is 2018-06-20.
• No forecast-sample rows before 2018-06-20 enter the modeling evidence.
• The leakage check reports status pass_with_warnings with zero leakage failures and 680103 warnings.
• FRED vintage safety is recorded as False; FRED values use conservative release timing but remain current historical observations rather than ALFRED real-time vintages.

Benchmark floor and advanced benchmarks

• benchmark_metrics.parquet reports 12 common-sample rows across 6 benchmark model families and 2 tail side(s), while benchmark forecasts contain 21146 model-date rows.
• Benchmark-floor models are external target-history and econometric baselines; this section does not rank them.
• Advanced benchmark rows are implemented for 10 model families and contribute 13214 nonblocking forecast rows; these rows are claim-gated diagnostics unless a manuscript table explicitly promotes them through the same sample and inference review.
• Benchmark-floor breach rates have a median of 0.0605144, within 2.5 percentage points of the nominal level, indicating reasonable coverage calibration relative to the ML-tail models whose breach rates are reported in the nested-information-set section.

Left/right ML-tail nested information sets

• ml_tail_metrics.parquet defines the headline ML-tail comparison across nested information sets for this run.
• The headline artifact contains 4 information sets, 1 tail level(s), and 2 tail side(s); the retained headline model rows are LGBM direct quantile.
• The implemented ML-tail registry is LGBM direct quantile, LGBM location-scale empirical, LGBM POT-GPD plain MLE, LGBM POT-GPD capped MLE, LGBM POT-GPD EVI shrink, LGBM POT-GPD EI-weighted, LGBM POT-GPD stabilized, LGBM conditional q90 POT-GPD plain MLE, LGBM conditional q90 POT-GPD stabilized, LGBM median/MAD POT-GPD plain MLE, LGBM median/MAD POT-GPD stabilized, LGBM median/IQR POT-GPD plain MLE, but the headline nested-information-set comparison should be read only from ml_tail_metrics.parquet.
• The nested information sets report downside-risk and upside-risk surfaces separately. The registered artifacts show different left/right patterns, and the generator does not assume that the two sides share the same economic mechanism.
• Coverage warning: all 8 headline rows exhibit VaR breach rates (0.0763636 to 0.127273) that exceed the nominal level by more than 2.5 percentage points. Quantile-loss and FZ-loss differences across the nested information sets must be interpreted in this context; lower loss scores may partly reflect less conservative VaR estimates rather than better conditional tail calibration.
• On left_tail, the largest quantile-loss change occurs at the first information-set augmentation (adding U.S. close core); subsequent additions of Japan proxy and Asia proxy ETFs contribute diminishing incremental loss changes. This saturation pattern is descriptive and does not automatically reduce the value of the broader information set.
• The nested information sets are used to assess candidate incremental U.S.-close information under strict common-sample rules; they do not by themselves establish forecast improvement.

Restricted model-family comparison

• ml_tail_result_matrix.parquet contains restricted common-sample comparisons for 12 LightGBM tail-model families.
• The restricted common-N range is 0 to 550 and the joint-exception range is 0 to 87.
• Recorded claim scopes are restricted_model_comparison_not_headline; these rows are restricted evidence and cannot replace the headline nested-information-set comparison.
• The tail-model family comparison is severely sample-limited: the largest restricted common-N is 28 rows. No model-family ranking claim is supportable from this restricted sample; extended OOS coverage is needed before tail-model family ranking becomes meaningful.
• Result-matrix inference is recorded separately from the headline suite-level DM/MCS: restricted DM records include 120 gate-pass rows and 210 unavailable rows; restricted MCS records include 0 gate-pass rows and 144 unavailable rows. These entries are restricted common-sample diagnostics, not headline model-family rankings.
• The result matrix is a matched-date diagnostic layer. It should not be worded as one family being better than another.

Coverage and inference gates

• Coverage review flags 8/8 headline rows with breach rates more than 2.5 percentage points from nominal coverage; Kupiec p-values fall below 0.05 in 8/8 rows and Christoffersen p-values fall below 0.05 in 0/8 rows where reported.
• Model-eviction artifacts record 8 retained rows and 80 non-retained rows under the headline sample policy.
• Block-bootstrap DM and HLN Tmax MCS artifacts are unconditional forecast-comparison diagnostics; any p-value should be read on average across the unconditional evaluation sample, not as condition-specific evidence.
• Loss differentials alone do not constitute an improvement claim; coverage, exception counts, sample gates, and inference status must be reviewed together.
• Result-matrix tail-event power flags and suite-level inference gates report 36 restricted rows with insufficient tail-event power and 0/36 unavailable DM/MCS inference rows.

CPA as conditional loss-difference diagnostics

• The ML-tail nested-information-set CPA artifact is a conditional loss-difference diagnostic across 2 tail side(s), with 42 registered row(s), 42 HAC-Wald gate pass(es), and loss families var_es_fz_loss, var_quantile_loss.
• The registered cross-model CPA artifact is a conditional loss-difference diagnostic with 608 row(s), 608 HAC-Wald gate pass(es), and loss families var_es_fz_loss, var_quantile_loss.
• Quantile-loss CPA and FZ-loss CPA are downstream inference over existing loss differentials; CPA does not generate VaR/ES forecasts and does not replace DM/MCS.

Supporting diagnostics

• Supporting LaTeX diagnostic table files are present for 4/4 registered diagnostic families.
• ml_tail_dst_attenuation.parquet contains 6 DST attenuation rows; these are descriptive timing-regime forecast diagnostics. They do not establish a structural timing mechanism.
• ES severity diagnostics contain 108 finite rows with mean exceedance severity ranging from 6.58221e-05 to 0.0127453; this is conditional-on-exception evidence.
• The diagnostic 75th-percentile VaR trigger rule marks 15295 model-date rows; 1000 of those rows coincide with VaR exceptions out of 3614 total exceptions, and mean triggered exception severity is 0.013949. This is a pre-open risk-monitoring diagnostic, not hedge PnL, transaction-cost, or trading-alpha evidence.
• Stress-window diagnostics contain 354 rows, and Murphy diagnostics contain 1600 ML-tail rows.
• Feature-unavailability diagnostics contain 330 rows.
• Figure manifest references:
• Figure: coverage_breach_rates_left_tail (Source: metrics/benchmark_metrics.parquet, metrics/benchmark_metrics_per_model.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet; Claim scope: coverage_diagnostic_not_headline_claim; File: latex/figures/coverage_breach_rates_left_tail.png).
• Figure: coverage_breach_rates_right_tail (Source: metrics/benchmark_metrics.parquet, metrics/benchmark_metrics_per_model.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet; Claim scope: coverage_diagnostic_not_headline_claim; File: latex/figures/coverage_breach_rates_right_tail.png).
• Figure: benchmark_murphy_left_tail (Source: metrics/benchmark_murphy.parquet; Claim scope: murphy_diagnostic_benchmark_floor_common_grid; File: latex/figures/benchmark_murphy_left_tail.png).
• Figure: benchmark_murphy_right_tail (Source: metrics/benchmark_murphy.parquet; Claim scope: murphy_diagnostic_benchmark_floor_common_grid; File: latex/figures/benchmark_murphy_right_tail.png).
• Figure: ml_tail_murphy_left_tail (Source: metrics/ml_tail_murphy.parquet; Claim scope: murphy_diagnostic_ml_tail_nested_information_sets_common_grid; File: latex/figures/ml_tail_murphy_left_tail.png).
• Figure: ml_tail_murphy_right_tail (Source: metrics/ml_tail_murphy.parquet; Claim scope: murphy_diagnostic_ml_tail_nested_information_sets_common_grid; File: latex/figures/ml_tail_murphy_right_tail.png).
• Figure: dst_attenuation_left_tail (Source: metrics/ml_tail_dst_attenuation.parquet; Claim scope: descriptive_dst_attenuation_not_structural_causal_identification; File: latex/figures/dst_attenuation_left_tail.png).
• Figure: dst_attenuation_right_tail (Source: metrics/ml_tail_dst_attenuation.parquet; Claim scope: descriptive_dst_attenuation_not_structural_causal_identification; File: latex/figures/dst_attenuation_right_tail.png).
• Figure: es_severity_left_tail (Source: metrics/benchmark_metrics.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet; Claim scope: es_severity_diagnostic_not_model_selection_claim; File: latex/figures/es_severity_left_tail.png).
• Figure: es_severity_right_tail (Source: metrics/benchmark_metrics.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet; Claim scope: es_severity_diagnostic_not_model_selection_claim; File: latex/figures/es_severity_right_tail.png).
• Figure: trigger_diagnostics_left_tail (Source: forecasts/benchmark_forecasts.parquet, forecasts/ml_tail_forecasts.parquet; Claim scope: trigger_diagnostic_not_pnl_cost_or_alpha; File: latex/figures/trigger_diagnostics_left_tail.png).
• Figure: trigger_diagnostics_right_tail (Source: forecasts/benchmark_forecasts.parquet, forecasts/ml_tail_forecasts.parquet; Claim scope: trigger_diagnostic_not_pnl_cost_or_alpha; File: latex/figures/trigger_diagnostics_right_tail.png).
• Figure: evt_standardized_qq_left_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_qq_left_tail.png).
• Figure: evt_standardized_log_survival_left_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_log_survival_left_tail.png).
• Figure: evt_standardized_mean_excess_left_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_mean_excess_left_tail.png).
• Figure: evt_standardized_hill_left_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_hill_left_tail.png).
• Figure: evt_standardized_threshold_stability_left_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_threshold_stability_left_tail.png).
• Figure: evt_standardized_qq_right_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_qq_right_tail.png).
• Figure: evt_standardized_log_survival_right_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_log_survival_right_tail.png).
• Figure: evt_standardized_mean_excess_right_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_mean_excess_right_tail.png).
• Figure: evt_standardized_hill_right_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_hill_right_tail.png).
• Figure: evt_standardized_threshold_stability_right_tail (Source: forecasts/ml_tail_forecasts.parquet; Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim; File: latex/figures/evt_standardized_threshold_stability_right_tail.png).

Not yet claimed

• DST attenuation rows are descriptive forecast evidence; structural DST causal identification is not claimed.
• No hedge PnL, transaction-cost, or trading-alpha analysis is performed. The trigger table is a pre-open risk-monitoring diagnostic only.
• Left-tail and right-tail outputs are both economic tail-risk surfaces for futures positions; neither side should be promoted beyond the sample, coverage, and inference gates without author review.
• The current evidence does not create an automatic model-selection statement; any manuscript claim still requires author review of sample gates, coverage, loss metrics, and inference diagnostics.

Appendix: Tables, Figures, And Run Artifacts

The appendix collects generated exports and provenance. The main Results section refers back to these files when a table or figure is suitable for manuscript use.

Paper-Facing Table And Figure Gallery

Table Manifest

Table	Source artifacts	Claim scope	Tail side	File
benchmark_metrics	metrics/benchmark_metrics.parquet	benchmark_common_sample_metric_table	None	latex/tables/benchmark_metrics_table.tex
benchmark_left_tail_risk	metrics/benchmark_metrics.parquet	left_tail_benchmark_risk_table	left_tail	latex/tables/benchmark_left_tail_risk_table.tex
benchmark_right_tail_risk	metrics/benchmark_metrics.parquet	right_tail_benchmark_risk_table	right_tail	latex/tables/benchmark_right_tail_risk_table.tex
ml_tail_metrics	metrics/ml_tail_metrics.parquet	ml_tail_nested_information_set_table	None	latex/tables/ml_tail_metrics_table.tex
ml_tail_left_tail_risk	metrics/ml_tail_metrics.parquet	left_tail_ml_tail_headline_risk_table	left_tail	latex/tables/ml_tail_left_tail_risk_table.tex
ml_tail_right_tail_risk	metrics/ml_tail_metrics.parquet	right_tail_ml_tail_headline_risk_table	right_tail	latex/tables/ml_tail_right_tail_risk_table.tex
tailrisk_es_severity	metrics/benchmark_metrics.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet	es_severity_diagnostic_table	None	latex/tables/tailrisk_es_severity_table.tex
tailrisk_trigger_diagnostics	forecasts/benchmark_forecasts.parquet, forecasts/ml_tail_forecasts.parquet	trigger_diagnostic_table	None	latex/tables/tailrisk_hedge_trigger_diagnostics_table.tex
tailrisk_claim_scope	manifest.json, config/research_config.json	claim_boundary_reference_table	None	latex/tables/tailrisk_claim_scope_table.tex
ml_tail_result_matrix	metrics/ml_tail_result_matrix.parquet	restricted_model_comparison_table	None	latex/tables/ml_tail_result_matrix_table.tex
ml_tail_result_matrix_summary	metrics/ml_tail_result_matrix.parquet, metrics/ml_tail_result_matrix_dm.parquet, metrics/ml_tail_result_matrix_mcs.parquet	restricted_result_matrix_summary_table	None	latex/tables/ml_tail_result_matrix_summary_table.tex
ml_tail_dst_attenuation	metrics/ml_tail_dst_attenuation.parquet	descriptive_dst_attenuation_table	None	latex/tables/ml_tail_dst_attenuation_table.tex

• The table manifest records the generated LaTeX table files, their source artifacts, and their claim scopes.
• Tables are paper-facing exports; the Markdown tables above are snapshot summaries for browser review.

Figure 1. Target Distribution And Tail Diagnostics

• Key readings: these figures describe the raw settlement-to-open gap and the left/right loss tails.
• They motivate VaR/ES and POT-GPD modeling, but they do not validate LightGBM+EVT forecasts.

Figure: target_gap_histogram_density. Source: panel/modeling_panel.parquet. Claim scope: target_distribution_motivation_not_forecast_validation. Tail side: target_distribution. Run file: latex/figures/target_gap_histogram_density.png.

Figure: target_loss_qq_left_tail. Source: panel/modeling_panel.parquet. Claim scope: target_distribution_motivation_not_forecast_validation. Tail side: left_tail. Run file: latex/figures/target_loss_qq_left_tail.png.

Figure: target_loss_qq_right_tail. Source: panel/modeling_panel.parquet. Claim scope: target_distribution_motivation_not_forecast_validation. Tail side: right_tail. Run file: latex/figures/target_loss_qq_right_tail.png.

Figure: target_log_survival. Source: panel/modeling_panel.parquet. Claim scope: target_distribution_motivation_not_forecast_validation. Tail side: left_right_target_distribution. Run file: latex/figures/target_log_survival.png.

Figure: target_mean_excess. Source: panel/modeling_panel.parquet. Claim scope: target_distribution_motivation_not_forecast_validation. Tail side: left_right_target_distribution. Run file: latex/figures/target_mean_excess.png.

Figure: target_hill_plot. Source: panel/modeling_panel.parquet. Claim scope: target_distribution_motivation_not_forecast_validation. Tail side: left_right_target_distribution. Run file: latex/figures/target_hill_plot.png.

Figure 2. Coverage Breach-Rate Diagnostics

• Key readings: bars report realized VaR exception rates against the nominal line.
• Read this first: exception-rate deviations set the boundary for any loss-based interpretation.

Figure: coverage_breach_rates_left_tail. Source: metrics/benchmark_metrics.parquet, metrics/benchmark_metrics_per_model.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet. Claim scope: coverage_diagnostic_not_headline_claim. Tail side: left_tail. Run file: latex/figures/coverage_breach_rates_left_tail.png.

Figure: coverage_breach_rates_right_tail. Source: metrics/benchmark_metrics.parquet, metrics/benchmark_metrics_per_model.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet. Claim scope: coverage_diagnostic_not_headline_claim. Tail side: right_tail. Run file: latex/figures/coverage_breach_rates_right_tail.png.

Figure 3. Benchmark Murphy Diagnostics

• Key readings: curves report benchmark elementary-score diagnostics on a common grid.
• The plot is a scoring-family diagnostic, not a pairwise ranking statement.

Figure: benchmark_murphy_left_tail. Source: metrics/benchmark_murphy.parquet. Claim scope: murphy_diagnostic_benchmark_floor_common_grid. Tail side: left_tail. Run file: latex/figures/benchmark_murphy_left_tail.png.

Figure: benchmark_murphy_right_tail. Source: metrics/benchmark_murphy.parquet. Claim scope: murphy_diagnostic_benchmark_floor_common_grid. Tail side: right_tail. Run file: latex/figures/benchmark_murphy_right_tail.png.

Figure 4. ML-Tail Murphy Diagnostics

• Key readings: curves report the ML-tail nested information sets on a common grid.
• Interpret curve separation together with the headline coverage warning and unconditional inference gates.

Figure: ml_tail_murphy_left_tail. Source: metrics/ml_tail_murphy.parquet. Claim scope: murphy_diagnostic_ml_tail_nested_information_sets_common_grid. Tail side: left_tail. Run file: latex/figures/ml_tail_murphy_left_tail.png.

Figure: ml_tail_murphy_right_tail. Source: metrics/ml_tail_murphy.parquet. Claim scope: murphy_diagnostic_ml_tail_nested_information_sets_common_grid. Tail side: right_tail. Run file: latex/figures/ml_tail_murphy_right_tail.png.

Figure 5. DST Attenuation Diagnostics

• Key readings: bars summarize timing-regime forecast diagnostics.
• Treat this as descriptive timing evidence; left/right patterns should not be assigned a shared structural mechanism.

Figure: dst_attenuation_left_tail. Source: metrics/ml_tail_dst_attenuation.parquet. Claim scope: descriptive_dst_attenuation_not_structural_causal_identification. Tail side: left_tail. Run file: latex/figures/dst_attenuation_left_tail.png.

Figure: dst_attenuation_right_tail. Source: metrics/ml_tail_dst_attenuation.parquet. Claim scope: descriptive_dst_attenuation_not_structural_causal_identification. Tail side: right_tail. Run file: latex/figures/dst_attenuation_right_tail.png.

Figure 6. ES Severity Diagnostics

• Key readings: bars report conditional-on-exception severity diagnostics.
• Severity is reported for risk interpretation but is not a standalone model-selection claim.

Figure: es_severity_left_tail. Source: metrics/benchmark_metrics.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet. Claim scope: es_severity_diagnostic_not_model_selection_claim. Tail side: left_tail. Run file: latex/figures/es_severity_left_tail.png.

Figure: es_severity_right_tail. Source: metrics/benchmark_metrics.parquet, metrics/ml_tail_metrics.parquet, metrics/ml_tail_metrics_per_model.parquet. Claim scope: es_severity_diagnostic_not_model_selection_claim. Tail side: right_tail. Run file: latex/figures/es_severity_right_tail.png.

Figure 7. Trigger Diagnostics

• Key readings: bars report pre-open risk-trigger diagnostics by model family.
• The trigger output is a monitoring diagnostic, not an execution-performance result.

Figure: trigger_diagnostics_left_tail. Source: forecasts/benchmark_forecasts.parquet, forecasts/ml_tail_forecasts.parquet. Claim scope: trigger_diagnostic_not_pnl_cost_or_alpha. Tail side: left_tail. Run file: latex/figures/trigger_diagnostics_left_tail.png.

Figure: trigger_diagnostics_right_tail. Source: forecasts/benchmark_forecasts.parquet, forecasts/ml_tail_forecasts.parquet. Claim scope: trigger_diagnostic_not_pnl_cost_or_alpha. Tail side: right_tail. Run file: latex/figures/trigger_diagnostics_right_tail.png.

Figure 8. EVT Standardized-Residual Diagnostics

• Key readings: figures show EVT diagnostics for LightGBM location-scale standardized residuals.
• QQ, log-survival, mean-excess, Hill, and threshold-stability diagnostics validate the POT-GPD tail assumption.
• These are assumption-validation diagnostics, not forecast-performance claims.

Figure: evt_standardized_hill_left_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: left_tail. Run file: latex/figures/evt_standardized_hill_left_tail.png.

Figure: evt_standardized_log_survival_left_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: left_tail. Run file: latex/figures/evt_standardized_log_survival_left_tail.png.

Figure: evt_standardized_mean_excess_left_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: left_tail. Run file: latex/figures/evt_standardized_mean_excess_left_tail.png.

Figure: evt_standardized_qq_left_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: left_tail. Run file: latex/figures/evt_standardized_qq_left_tail.png.

Figure: evt_standardized_threshold_stability_left_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: left_tail. Run file: latex/figures/evt_standardized_threshold_stability_left_tail.png.

Figure: evt_standardized_hill_right_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: right_tail. Run file: latex/figures/evt_standardized_hill_right_tail.png.

Figure: evt_standardized_log_survival_right_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: right_tail. Run file: latex/figures/evt_standardized_log_survival_right_tail.png.

Figure: evt_standardized_mean_excess_right_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: right_tail. Run file: latex/figures/evt_standardized_mean_excess_right_tail.png.

Figure: evt_standardized_qq_right_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: right_tail. Run file: latex/figures/evt_standardized_qq_right_tail.png.

Figure: evt_standardized_threshold_stability_right_tail. Source: forecasts/ml_tail_forecasts.parquet. Claim scope: evt_standardized_residual_diagnostic_not_forecast_claim. Tail side: right_tail. Run file: latex/figures/evt_standardized_threshold_stability_right_tail.png.

Artifact Index

Artifact	Path	Exists
manifest	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/manifest.json	yes
data_vintage	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/data_vintage.json	yes
modeling_panel	data/gold/tp/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/modeling_panel.parquet	yes
target_audit	data/gold/tp/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/target_audit.parquet	yes
calendar_map	data/gold/tp/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/calendar_map.parquet	yes
feature_coverage	data/gold/tp/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/feature_coverage.parquet	yes
leakage_summary	data/gold/ls/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/summary.json	yes
benchmark_status	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/benchmark_status.json	yes
benchmark_metrics	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/benchmark_metrics.parquet	yes
benchmark_metrics_per_model	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/benchmark_metrics_per_model.parquet	yes
benchmark_forecasts	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/forecasts/benchmark_forecasts.parquet	yes
benchmark_dm_inference	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/benchmark_dm_inference.parquet	yes
benchmark_mcs	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/benchmark_mcs.parquet	yes
ml_tail_status	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_status.json	yes
ml_tail_metrics	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_metrics.parquet	yes
ml_tail_metrics_per_model	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_metrics_per_model.parquet	yes
ml_tail_forecasts	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/forecasts/ml_tail_forecasts.parquet	yes
ml_tail_result_matrix	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_result_matrix.parquet	yes
ml_tail_result_matrix_dm	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_result_matrix_dm.parquet	yes
ml_tail_result_matrix_mcs	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_result_matrix_mcs.parquet	yes
ml_tail_dm_inference	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_dm_inference.parquet	yes
ml_tail_mcs	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_mcs.parquet	yes
ml_tail_cpa_inference	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_cpa_inference.parquet	yes
cross_model_cpa_inference	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/cross_model_cpa_inference.parquet	yes
ml_tail_model_eviction	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_model_eviction.parquet	yes
ml_tail_dst_attenuation	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_dst_attenuation.parquet	yes
ml_tail_murphy	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_murphy.parquet	yes
ml_tail_feature_unavailability	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_feature_unavailability.parquet	yes
benchmark_stress_windows	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/benchmark_stress_windows.parquet	yes
ml_tail_stress_windows	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/metrics/ml_tail_stress_windows.parquet	yes
figure_manifest	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/latex/figure_manifest.json	yes
table_manifest	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/latex/table_manifest.json	yes
latex_dir	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/latex/tables	yes
claim_scope_table	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/latex/tables/tailrisk_claim_scope_table.tex	yes
es_severity_table	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/latex/tables/tailrisk_es_severity_table.tex	yes
hedge_trigger_table	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/latex/tables/tailrisk_hedge_trigger_diagnostics_table.tex	yes
dst_attenuation_table	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/latex/tables/ml_tail_dst_attenuation_table.tex	yes
result_matrix_summary_table	reports/runs/tailrisk_20160719_20260504_20260506T035942Z_commit_0c14f1be/latex/tables/ml_tail_result_matrix_summary_table.tex	yes

• All paths above are local ignored artifacts; they are reproducible outputs, not tracked source files.
• Forecast/reporting rebuilds should read these artifacts and must not call vendor APIs.
• If this page is stale, rerun just snapshot after a completed just full or pass an explicit run id to the CLI snapshot command.

Technical Infrastructure Note

• Runtime imports are explicit at the module boundary; no dynamic runtime namespace bridge is required to generate this snapshot. This infrastructure note is separate from empirical claim boundaries.