Pre-built synthetic datasets.
Skip the platform. Skip the PoC of the PoC.
Production-ready CSVs across fintech, retail, security, talent and compliance — labelled, statistically audited, ready for pd.read_csv(). Skip the 6-month vendor sales cycle. Skip the platform you have to learn. Skip the Kaggle dataset everyone overfit in 2018.
Why not just use…
The four alternatives — and why they break for production.
We're not competing with Mostly AI or Kaggle. We're a different category: pre-built, audited, downloadable. Here's the comparison most buyers run mentally.
Public datasets
Kaggle 2013 fraud, UCI, CICIDS 2017, Olist
- Works for
- Free. Cited in every tutorial. Good enough for a teaching slide.
- Breaks at
- Stale (most are 2013–2018). PCA-anonymised, so no feature engineering. Every job candidate has memorised them. No fresh fraud / attack typologies.
- NexusMock instead
- Fresh schemas with real engineered features, refreshed quarterly.
DIY with an LLM
"Just ask GPT to make 100k rows"
- Works for
- Free. Fast iteration. Looks plausible at first glance.
- Breaks at
- No statistical guarantees. Distributions drift, correlations break, labels are nonsense. Your fraud_type column will not survive a Random Forest.
- NexusMock instead
- Deterministic generation pipeline + automated Benford / AUC / integrity audit per release.
Synthetic-data platforms
Mostly AI, Gretel, Tonic, Hazy, Synthesized.io
- Works for
- Powerful. Audit-grade. Match real distributions when configured right.
- Breaks at
- You buy the factory, not the output. 50k€+/year, weeks of onboarding, you still have to design the schema. Wrong tool for "I need a CSV by Friday".
- NexusMock instead
- We already ran the platform for you. You get the CSV.
Custom vendor / consultancy
Bespoke synthetic-data engagement
- Works for
- Any schema you want. Tailored to your portfolio.
- Breaks at
- 3–6 month sales cycle. 25–100k€ minimum. NDA + DPA + procurement gates. By the time it ships, your roadmap has moved.
- NexusMock instead
- Self-serve under the corporate-card threshold. Custom variant in 5 days for 999€ if you really need it.
Catalog
5 verticals, one quality bar.
Every release passes the same integrity, statistical, and detectability checks. Pick the one you need — or join a waitlist to vote on what we build next.
Fraud Transactions
Credit-card transactions with five fraud typologies pre-labelled. For anti-fraud ML, rules tuning, and sandbox traffic.
E-commerce Customer Behavior
Synthetic shopper sessions with clickstream, basket composition, and conversion outcomes. For recommender systems, churn models, and CRO experiments.
SOC Event Logs
SIEM-ready synthetic event streams with six attack typologies labelled. ECS-aligned schema, drop into Splunk/Sentinel/Elastic today.
Synthetic CVs
Resume data for AI hiring tools, matching engines, and fairness audits. Demographically balanced, EU AI Act Annex IV audit-ready, no real candidates.
Crypto KYC Transactions
On-chain transaction graphs with AML-labelled clusters: mixer use, layering, structuring, sanctioned counterparts. For chain analysis and MiCA compliance.
Featured — available now
Real fraud patterns. Zero real cardholders.
Synthetic credit-card transactions with five fraud typologies pre-labelled. Train your detection models today — without RGPD, HIPAA, PCI scope, or eighteen months of legal review.
The quality bar
The same standard for every dataset on the shelf.
Real numbers, real datasets
The ROC curves on the left are computed from output/nexusmock_*_pro_100k.csv with a plain Random Forest baseline (no tuning, 12 features, 30% test split). Fraud sits at 0.92 because we ship a friendly_fraud noise floor. SOC at 0.95 because 30% of the events are ambiguous false-positive territory.
The numbers in our quality cards are not marketing. They are the actual outputs of deep_validate.py, the same script shipped inside every paid ZIP as QUALITY_REPORT.md.
Integrity, audited.
No nulls. No duplicates. No type contamination. Every release runs an automated integrity suite before it ships.
Statistically realistic.
Distributions, correlations and conditional dependencies match published benchmarks for the vertical. Where there's no public benchmark, we say so.
Labelled and learnable.
Every dataset ships with a model-ready baseline AUC. If a simple gradient boosting can't beat 0.85, the dataset doesn't ship.
Documented to ship.
Every package includes a data dictionary, a quality report, a starter Jupyter notebook, and a clear commercial licence. Open the ZIP, read 5 minutes, decide.
Pricing philosophy
Priced to be approved on a corporate card.
Most teams need to ship a model this quarter, not next year. Our tiers are deliberately under common procurement thresholds: 29€ / 79€ / 249€. Pay, download, build. No DPA, no PO, no quarterly call.
Need a custom volume, vertical, or onboarding contract? info@nexusmock.com.
What's next
We build the verticals you ask for.
4 datasets are in the pipeline. Waitlist signups set the priority — the one with the most demand ships first.
- RetailE-commerce Customer BehaviorJoin waitlist →
- CybersecuritySOC Event LogsJoin waitlist →
- HR / TalentSynthetic CVsJoin waitlist →
- ComplianceCrypto KYC TransactionsJoin waitlist →
Don't see your vertical? Tell us what you need → We'll quote a custom dataset in 24 hours.