Examples gallery¶

Two ways to start: example specs (hand-written, in the repo) and built-in templates (shipped with the package, one command to instantiate).

Example specs¶

Runnable specs in examples/. Each is part of the determinism gate, so every one regenerates byte-identically.

Spec	Demonstrates
`tabular-basic`	Plain distribution-only tabular data — the smallest useful spec.
`causal-fraud`	A causal DAG (`age → income → is_fraud`, `education → income`) with a logistic target.
`failure-fraud`	The fraud DAG plus six stacked failure modes; ships clean + injected variants.
`difficulty-credit`	A credit-default label calibrated to the `advanced` difficulty band, with a latent score.
`people-realistic`	Realistic-but-seeded text fields (names, emails, addresses) via mimesis providers.
`timeseries-sensor`	An additive time-series feature: trend + seasonality + AR(p) + noise.

Run any of them:

datadoom run examples/causal-fraud.datadoom.yaml --seed 42 --out ./out
datadoom verify examples/causal-fraud.datadoom.yaml --seed 42 --against ./out

Built-in templates¶

Start from a curated domain spec with one command (or one click in the web Templates gallery):

datadoom template list                # all templates
datadoom template list --level hackathon
datadoom template use fraud-detection --out my.datadoom.yaml

Hackathon challenges¶

Enterprise-grade ML challenges built by composition — deep causal DAGs, latent confounders, mixed feature types, stacked data-quality failures, and calibrated difficulty bands. Each carries a meta.challenge brief.

Template	Domain	Highlights
`credit-default-challenge`	Finance	Deep DAG → latent risk score → default; `advanced` band; MNAR/MAR/drift/leakage/label-noise.
`clinical-deterioration`	Healthcare	Hidden-confounder design — latent severity drives vitals and outcome; `advanced` band.
`predictive-maintenance`	Industrial IoT	Three additive time-series sensors → latent wear → label; drift/noise/MCAR/leakage.
`telecom-churn-challenge`	Telecom	Realistic-text identity + latent dissatisfaction → churn; hard `kaggle` band; MNAR.

Starter templates¶

Template	Domain	Highlights
`fraud-detection`	Finance	Causal structure + failure injection.
`customer-churn`	SaaS	Difficulty targeting + a latent feature.
`hospital-readmission`	Healthcare	Causal structure + a latent feature.
`ecommerce-orders`	E-commerce	Quick distribution-only table.
`iot-sensors`	IoT	Sensor readings.
`people-directory`	People	Realistic text via mimesis providers.
`marketing-ab-test`	Marketing	A/B test design.
`insurance-claims`	Insurance	A Pareto heavy-tailed claim amount.