Results Snapshot

Generated by just snapshot from artifacts/full_with_peer at 2026-05-11T05:44:55+00:00.

Discussion

Research question. Can filing-origin public SEC/PCAOB information predict whether an issuer later enters observable public review-and-correction channels, and how does this public reporting-risk construct relate to, but differ from, legacy detected-misstatement benchmarks?
Data. The workflow combines the legacy gvkey x data_year detected-misstatement benchmark, the public SEC/PCAOB lake, the gold issuer_origin_panel and filing_origin_panel, and an external gvkey-CIK-year bridge for overlap validation.
Models. The core public cascade uses XGBoost over metadata, XBRL, text/notes, auditor, oversight, and all-feature sets. Peer-compatible Dechow, Perols, Bao, and Bertomeu-style suites are included when the peer-enabled study directory is present.
Metrics. The common metric vocabulary is PR-AUC relative to prevalence, ROC-AUC, Brier, Brier Skill Score, ECE, top-k precision, top-decile lift, and Bao-style top-fraction precision, sensitivity, specificity, BAC, and NDCG.
Sellable claim. The strongest current framing is a measurement-and-ranking paper on filing-origin public reporting-risk states. It does not support causal claims, unobserved true-fraud occurrence claims, or same-estimand performance rankings over prior fraud-prediction papers.
Current best public-cascade specification. all + rolling_7y with reported mean PR-AUC 0.2430.
Bridge boundary. Construct overlap is candidate_farr; WRDS or equivalent institutional bridge evidence remains preferred for final manuscript-grade integrated claims.

Run Metadata

Field	Value
Study directory	`artifacts/full_with_peer`
Snapshot mode	`full`
Study manifest timestamp	`2026-05-11T04:56:56+00:00`
Public-lake report timestamp	`2026-05-11T03:51:32+00:00`
Peer comparison mode	`full`
Bridge status	`external_crosswalk_available`
Construct-overlap validation tier	`candidate_farr`
Raw benchmark input	`/Volumes/ExternalSSD/data/reporting-risk-cascade/raw_dataset_misstatement.parquet`
Public issuer panel	`/Volumes/ExternalSSD/data/reporting-risk-cascade/public_lake/gold/issuer_origin_panel.parquet`
Bridge crosswalk	`/Volumes/ExternalSSD/data/reporting-risk-cascade/external/gvkey_cik_year.csv`

Component Status

Component	Status	Tier	Output
benchmark	`complete`		artifacts/full_with_peer/benchmark
public_cascade	`complete`		artifacts/full_with_peer/public_cascade
bridge_probe	`external_crosswalk_available`		artifacts/full_with_peer/bridge_probe
peer_comparison	`complete`		artifacts/full_with_peer/peer_comparison
public_peer_comparison	`complete`		artifacts/full_with_peer/public_peer_comparison
construct_overlap	`complete`	`candidate_farr`	artifacts/full_with_peer/construct_overlap

Evidence Map

flowchart LR
    L["Legacy benchmark<br/>timing, drift, missingness,<br/>peer-compatible metrics"]
    P["Public filing-origin cascade<br/>comment threads, amendments,<br/>8-K Item 4.02, AAER support"]
    B["Bridge gate<br/>gvkey-CIK-year coverage<br/>candidate_farr unless WRDS supplied"]
    V["Construct-overlap checks<br/>co-occurrence, lift,<br/>reciprocal ranking, event time"]
    S["Snapshot docs<br/>generated from artifacts<br/>checked by just snapshot"]
    L --> B
    P --> B
    B --> V
    V --> S

Public Lake and Gold Panel Scale

Layer	Artifact	Rows	Notes
Silver	`filing_dim`	21,786,118	normalized public filing index
Silver	`issuer_dim`	968,124	normalized issuer dimension
Silver	`xbrl_core_fact`	18,010,256	controlled XBRL core facts
Silver	`xbrl_fact_summary`	362,013	accession-level fact coverage
Silver	`note_summary`	345,490	Notes summary mode
Silver	`comment_thread`	125,381	SEC comment-thread signal
Silver	`correction_event`	90,117	amended-filing/correction signal
Gold	`issuer_origin_panel`	205,719	annual issuer-year modeling table
Gold	`filing_origin_panel`	21,786,118	filing-origin provenance table

Public Cascade Readiness

Field	Value
Main sample rows	90,342
Fiscal-year span	2011-2023
Domestic US GAAP only	`True`
Task positive counts	`{"8k_402": 2008, "aaer_proxy": 19, "amendment": 17241, "comment_thread": 24840}`
Zero-positive tasks	`none`
Task status counts	`{"fit": 520, "skipped_one_class_train": 120}`
Readiness level	`xbrl_ratio_baseline`
Best reported feature set	`all`
Best reported train window	`rolling_7y`
Best reported mean PR-AUC	0.2430

Public Task Metrics

Task	Positives	Mean prevalence	Mean PR-AUC	Mean ROC-AUC	Rows
`comment_thread`	24,840	0.2615	0.3654	0.6327	160
`amendment`	17,241	0.1552	0.2530	0.6271	160
`8k_402`	2,008	0.0221	0.0506	0.6544	160
`aaer_proxy`	19	0.0012	0.0490	0.6503	40

Public Feature-Family Metrics

Feature set	Features	XBRL ratios	XBRL coverage	Mean PR-AUC	Mean ROC-AUC	Rows
`all`	79	11	15	0.2730	0.7382	104
`metadata`	28	0	0	0.2578	0.7156	104
`xbrl`	42	11	15	0.2001	0.6371	104
`auditor`	6	0	0	0.1673	0.5527	104
`oversight`	1	0	0	0.1499	0.5514	104

Legacy Benchmark Timing Diagnostics

Field	Value
Rows	82,908
Firms	9,156
Years	2001-2019
Positive rate	0.0297
Positive rows without timing proxy	2,309
Timing claim status	`proxy_imputed_lag`

Label mode	Best window	Mean PR-AUC	Mean ROC-AUC	Top-100 precision	Retained positive share
`naive`	`rolling_5y`	0.0729	0.7301	0.0879	1.0000
`proxy_imputed_lag_1y`	`rolling_5y`	0.0451	0.6576	0.0621	0.8232
`proxy_imputed_lag_2y`	`expanding`	0.0394	0.6666	0.0543	0.9037
`proxy_imputed_lag_3y`	`expanding`	0.0340	0.6532	0.0471	0.8423
`proxy_imputed_lag_5y`	`expanding`	0.0322	0.6320	0.0379	0.6874
`proxy_drop_observed`	`rolling_7y`	0.0229	0.5549	0.0243	0.0604

Peer-Compatible Literature Benchmarks

These rows are present only when the peer-enabled study has run. They are model-family transfer and metric-language alignment, not exact replications of the original-paper samples.

Model	Rows	Mean PR-AUC	Mean ROC-AUC	Max PR-AUC	Mean Brier
`bertomeu_style_xgb`	336	0.0427	0.6601	0.1710	0.0162
`perols_logit`	336	0.0315	0.6156	0.0759	0.1775
`perols_bagged`	336	0.0311	0.6271	0.0868	0.1809
`perols_linear_svm`	336	0.0306	0.6131	0.0745	0.1862
`perols_stacking`	336	0.0302	0.6075	0.0708	0.1967
`perols_mlp`	336	0.0297	0.5888	0.0716	0.2022
`bao_inspired_tree_ensemble`	336	0.0283	0.6251	0.0628	0.0165
`dechow_variable_logit`	336	0.0235	0.5225	0.0672	0.2466
`perols_entropy_tree`	336	0.0227	0.5810	0.0444	0.2245

Public-Label Peer Transfer

Model	Rows	Mean PR-AUC	Mean ROC-AUC	Max PR-AUC	Mean Brier
`bertomeu_style_xgb`	480	0.2247	0.6453	0.5082	0.1124
`bao_inspired_tree_ensemble`	480	0.2245	0.6449	0.5063	0.1124
`perols_bagged`	480	0.2119	0.6319	0.4561	0.2238
`perols_stacking`	480	0.2074	0.6184	0.4454	0.2224
`perols_linear_svm`	480	0.2056	0.6166	0.6283	0.2242
`perols_logit`	480	0.2056	0.6217	0.5472	0.2252
`perols_mlp`	480	0.2002	0.6070	0.4413	0.2324
`perols_entropy_tree`	480	0.1997	0.6125	0.4264	0.2287
`dechow_variable_logit`	480	0.1584	0.5348	0.3084	0.2500

Public Peer Task Summary

Task	Rows	Mean prevalence	Mean PR-AUC	Mean ROC-AUC	Max PR-AUC
`comment_thread`	1,440	0.2615	0.3284	0.5915	0.5082
`amendment`	1,440	0.1552	0.2326	0.6097	0.3854
`8k_402`	1,440	0.0221	0.0516	0.6432	0.6283

Bridge and Construct-Overlap Validation

Metric	Value
raw_rows	82,908
raw_firms	9,156
matched_raw_rows	81,218
matched_raw_firms	9,075
row_coverage_rate	0.9796
firm_coverage_rate	0.9912
raw_positive_rows	2,460
matched_positive_rows	2,433

Direction	Model	Target	PR-AUC	ROC-AUC	Top-decile lift
Public cascade score -> legacy positives	`public_cascade`	`8k_402`	0.0326	0.6828	2.9462
Legacy/peer score -> public labels	`bertomeu_style_xgb`	`label_8k_402_365`	0.0436	0.7033	3.0477

Key readings:

Public labels and legacy detected-misstatement labels are related but non-identical constructs.
Public-cascade scores can rank legacy positives in the matched overlap; legacy/peer scores can also rank severe public correction labels.
candidate_farr bridge evidence is useful for internal validation, but should be labeled clearly until a WRDS-grade bridge is available.

Selected Artifact Index

This index lists high-signal artifacts referenced by this generated snapshot.

artifacts/full_with_peer/study_summary.md (present)
artifacts/full_with_peer/study_run_manifest.json (present)
artifacts/full_with_peer/benchmark/benchmark_summary.md (present)
artifacts/full_with_peer/benchmark/rolling_metrics.csv (present)
artifacts/full_with_peer/public_cascade/public_cascade_summary.md (present)
artifacts/full_with_peer/public_cascade/public_cascade_metrics.csv (present)
artifacts/full_with_peer/peer_comparison/peer_comparison_summary.md (present)
artifacts/full_with_peer/peer_comparison/legacy_model_family_metrics.csv (present)
artifacts/full_with_peer/public_peer_comparison/public_model_family_summary.md (present)
artifacts/full_with_peer/public_peer_comparison/public_model_family_metrics.csv (present)
artifacts/full_with_peer/bridge_probe/bridge_probe_summary.json (present)
artifacts/full_with_peer/bridge_probe/coverage_report.csv (present)
artifacts/full_with_peer/construct_overlap/construct_overlap_summary.md (present)
artifacts/full_with_peer/construct_overlap/public_score_legacy_ranking.csv (present)
artifacts/full_with_peer/construct_overlap/reciprocal_alignment.csv (present)
artifacts/full_with_peer/opacity_validation_refresh/opacity_diagnostics_summary.csv (present)