Criteo Uplift Prediction Dataset

Source

• Dataset metadata snapshot: source.md
• Metadata JSON: metadata.json
• Canonical external source: https://ailab.criteo.com/criteo-uplift-prediction-dataset/

Action-Time-Series Notes

• Dataset type: marketing treatment-effect benchmark.
• Temporal structure: logged ad-exposure/customer records; not a rich sequential state trajectory.
• Actions or interventions: binary treatment/control exposure plus visit/conversion outcomes.
• Suitability: useful for one-step treatment-response or uplift modeling; weak for world models because next-state dynamics are thin.

Foundation TSFM Relevance

Agenda slot	Verdict	Evidence	Missing pieces
Causal structure, counterfactuals, and control	partially closes	The source records binary treatment/control exposure with visit and conversion outcomes.	One-step uplift labels do not provide state-transition rollouts or rich action histories.
Benchmark level	warning	It is a useful treatment-effect benchmark, but the modeling target is customer response, not system-state evolution.	Needs sequential state, repeated interventions, and temporal counterfactual evaluation.

Links Into The Wiki

• Foundation Time-Series Model Research Agenda
• Action-Conditioned Time-Series Datasets
• World Models