What's covered?
Guides on building AI-generated, contextually consistent test data and getting it into a training- or testing-ready format — .jsonl and HuggingFace datasets. Each one walks through a specific use case: MDM/entity data, person data, ML training sets, and export.
MDM Test Data: Synthetic Person & Entity Records
Building synthetic, correlated person/entity fields for MDM matching and merge testing.
Person Test Data Set: Synthetic, No Real PII
Synthetic person data with contextually consistent fields, no real PII.
Synthetic Data for Machine Learning
AI-generated, correlated rows exported to JSONL/HuggingFace for training and eval.
JSONL & HuggingFace Dataset Export
Exporting a generated dataset to .jsonl or an hf-datasets bundle.
Entity Resolution Test Data: Ground-Truth Answer Key
Labeled entity-resolution sample with Master ID answer key and pairwise F1 arithmetic.
Test Data Export Formats: CSV to Parquet Compared
Which export format to pick for fixtures, and what breaks if you pick wrong.
Python Test Data Generator: Faker and pytest
Custom Faker providers and deterministic pytest faker_seed wiring.
JavaScript Test Data Generator: faker-js and Fishery
Typed Fishery factories with faker-js seed and rewindSequence reset.
Java Test Data Generator: Datafaker and Instancio
Datafaker leaves with Instancio object graphs for JUnit 5 fixtures.
C# Test Data Generator: Bogus StrictMode
Bogus Faker with StrictMode and xUnit determinism.
SQL Test Data Generator: generate_series
Server-side row generation with generate_series, setseed, and CTEs.
Finance Test Data: NACHA and Payment Fixtures
ACH, pacs.008, and check-digit-valid payment fixtures for banking QA.
Healthcare Test Data: FHIR, ICD-10, NPI
Synthetic EHR and claims fixtures with code-system membership checks.
Retail Test Data: GTIN and Order Fixtures
Check-digit-valid GTINs in relationally consistent order fixtures.
CRM Test Data: Salesforce Objects and Duplicates
Labeled CRM duplicates plus Opportunity StageName derived-field traps.
Why an Unvalidated Match Rule Is a Guess, Not a Decision
An unvalidated match rule is a guess dressed as a decision. Why eyeballing a sample can't prove precision, and what measurement looks like.
Precision, Recall, and F1 for Match Rules: A Worked Example
Precision, recall, and F1 for match-rule validation, with a full worked example: candidate pairs, TP/FP/FN, and the arithmetic shown step by step.
Gate-Review Certainty Report: What Good Looks Like
What a defensible gate-review report contains: per-rule scorecard, failure list, re-test delta, sign-off line. The anatomy, not a customer result.
How to Define Truth Keys for Fuzzy Duplicates
How to choose a truth key, tag known duplicate pairs to a Master ID, and classify duplicate types before you label a match-rule truth set.
Defending Match-Rule Quality at Steering Committee
How an SI practice lead presents and defends a match-rule score at steering committee or gate review, with evidence instead of assurance.
Match-Rule Regression: Why a Passing Score at Go-Live Doesn't Stay Passing
Match-rule regression is a precision drop after a rule change, upgrade, or new data source. Re-score against the same truth set to catch it early.
Match-Ready Examples: Patient, Supplier, and B2C Synthetic Datasets
A Match-Ready Sample is synthetic data shaped like a domain: patient, supplier, or B2C. See illustrative fields and duplicate-tagging examples for each.
Hand-Labeling Cost Explainer: A Formula You Run With Your Own Numbers
Hand-labeling cost is pairs reviewed x minutes per pair / 60 x hourly rate. Two worked illustrative examples. No market figure, no typical-cost claim.
Back to the product landing page to generate a dataset.