Discovery quality mediation and durable user value

Causal Inference
Mediation
Recommendation Systems
Metrics
A mediation project asking whether broader discovery exposure creates future value directly or indirectly through same-day satisfaction depth.
Published

April 30, 2026

Decision Question

How much of the effect of broader discovery exposure on longer-term user value flows through same-day satisfaction depth?

Causal Setup

  • Treatment: broader discovery exposure.
  • Mediator: same-day satisfaction depth.
  • Outcome: future user value and engagement.
  • Challenge: CTR may not capture durable satisfaction.

Methods

  • Metric construction and validation
  • Mediation estimands and assumptions
  • Direct, indirect, and total effects
  • Robustness and sensitivity checks
  • SEM-style and ML mediation extensions

Portfolio Takeaway

The project frames metric design as a causal problem: a good discovery metric should explain durable value, not merely immediate response.

Selected Figures

26 Final Effect Decomposition

27 Final Robustness Ranges

28 Final Advanced Model Comparison

Notebook Sequence

The links below open rendered HTML versions of the notebooks. The source .ipynb files are kept in the matching folder under notebooks/projects/.

  1. Discovery Quality Problem Setup and EDA
  2. Metric Construction and Validation for Discovery Quality
  3. Mediation Estimands and Assumptions for Discovery Quality
  4. Direct, Indirect, and Total Effects for Discovery Quality
  5. Robustness and Sensitivity for Discovery Quality Mediation
  6. Advanced SEM and ML Mediation Models
  7. Final Report and Figures for Discovery Quality Mediation

Generated Artifacts

Limitations

These are notebook-driven causal analyses, not production guarantees. Each project should be read with its identification assumptions, support diagnostics, measurement choices, and sensitivity checks in view.