Financial · LQS Verified v3.1

Training data you can cite in a model-risk file.

Labeled datasets for fraud detection, credit scoring, sentiment, and risk modeling. Every one ships with a cryptographically-signed LQS cert your MRM team can verify against SR 11-7 / OCC 2011-12 and ECOA fair-lending review. Multi-oracle consensus, 95% CI, contamination-checked against public evals.

Formats
4
Parquet · CSV · SQLite · Arrow
Tasks
6+
Sentiment · fraud · credit · tick · crypto · alt-data
MRM framework
SR 11-7
Fed · OCC 2011-12 · cert-citable
Fair lending
ECOA
Subgroup-equity dim · per-group CI
MRM + fair-lending fields mapped into every cert
SR 11-7
Fed · model risk
OCC 2011-12
MRM guidance
ECOA
fair lending
GDPR
Art. 28 + SCCs
Ed25519
signed cert
Featured datasets

Financial data with verifiable provenance.

Live marketplace listings filtered to finance. Every card shows signed LQS score, subgroup-equity flag (for credit / underwriting use), adversarial-stability score (for fraud), and contamination screen against FinanceBench / SEC EDGAR.

Tasks covered

From market microstructure to earnings sentiment.

Market sentiment
Financial news articles, earnings-call transcripts, and analyst reports labeled with bullish/bearish sentiment and market-impact scores.
schema · direction · magnitude · horizon
Crypto & blockchain
Crypto OHLCV data, on-chain transaction graphs, and social-sentiment datasets labeled for price-movement prediction and wash-trade detection.
venues · CEX + DEX + L1/L2
Fraud detection
Transaction datasets with labeled fraud/legitimate cases. Adversarial-stability scored — resilient against input perturbations from actual fraud adversaries.
field · adversarial_stability
Price & tick data
Historical OHLCV, tick-by-tick, and order-book snapshots with event labels for backtesting and model training. Survivorship-bias disclosures required.
granularity · 1ms – 1d
Credit & risk
Anonymized loan-application datasets with default labels and credit-scoring features. ECOA-aligned subgroup-equity metrics on every cert.
field · subgroup_equity · ECOA
Alternative data
Web-scraped, satellite, and proprietary alt-data sources with financial-performance correlation labels. Source lineage + provenance chain preserved.
sources · web · geo · transactional
Parquet · columnar (tick data) CSV · headers required SQLite · .db Arrow · zero-copy
What LabelSets adds

Cert fields your MRM team can cite.

Financial AI sits under SR 11-7, OCC 2011-12, ECOA, and the EU AI Act. LabelSets certs carry the model-risk, fair-lending, and adversarial-stability evidence your validation package needs.

SR 11-7 citable
Independent third-party attestation closes the "training-data lineage" gap in model-validation packages. Ed25519 signature + timestamp survives audit.
framework · SR 11-7 · OCC 2011-12
ECOA fair-lending dim
Per-subgroup accuracy breakdowns baked into the cert for ECOA / Reg B fair-lending audits. Protected-class balance and per-group CI captured.
field · subgroup_equity · ECOA
Adversarial stability (fraud)
Every fraud dataset perturbed by the LQS scorer with adversarial input variants. Stability score embedded — so your fraud model holds up against real adversaries.
field · adversarial_stability
Downstream-F1 projection
LQS v3.1 projects expected F1 at 10× data volume — so your validation team can decide whether to procure more of a given seller's supply.
field · f1_projection_10x
Benchmark contamination
Every financial dataset hashed against FinanceBench, SEC EDGAR splits. Overlap flagged at the cert level so backtest numbers hold up.
screens · FinanceBench · EDGAR
Ed25519-signed provenance
Every cert carries a public-key signature + fingerprint. Buyers verify at /verify. Revocation registry handles post-facto license or PII flags.
fingerprint · aa4c070af907e2ea
FAQ

Questions MRM teams actually ask.

Stock and crypto price data, earnings-call transcripts with sentiment labels, financial news with market-impact labels, fraud and anomaly-detection datasets, credit-scoring data, and alternative-data sources.
CSV, Parquet, SQLite, and Arrow. Parquet is recommended for large tick-data datasets due to efficient columnar compression and fast read times with pandas and polars. SQLite is preferred when relational joins are part of the workflow.
Yes, if you hold the data rights. Financial datasets with personal information (names, account numbers) must be anonymized before uploading. Upload, pass verification, set a price, and earn 85% of every sale.
Suitability depends on the specific dataset — check the listing description for coverage dates, frequency, and survivorship-bias disclosures. Sellers are required to document data provenance clearly on the cert. Contamination screening against FinanceBench and SEC EDGAR is automatic.
Yes. The LQS cert is designed to drop into the dataset-quality section of SR 11-7 / OCC 2011-12 model-validation packages. Independent third-party attestation of training-data lineage is a frequent audit gap — the signed cert addresses that directly. Buyers should confirm with their MRM team.

Browse all financial datasets.

Live marketplace filtered by LQS score, subgroup-equity flag, adversarial-stability score, and format. Or list proprietary market data under NDA — enterprise private listings available.

Browse datasets → Talk to enterprise
Related categories
NLP / Text Computer Vision Audio & Speech Medical Imaging Autonomous Vehicles