Real public dataset — independent e-commerce analysis

Different data.
Different evidence.

12,330 sessions · 1,908 purchase sessions · 15.47% purchase rate.
95% Wilson interval: 14.85%–16.12%.

C. Sakar & Y. Kastro (2018), Online Shoppers Purchasing Intention Dataset, UCI Machine Learning Repository. DOI 10.24432/C5F88Q. CC BY 4.0. Downloaded 9 October 2026. Original measurements retained; aggregate analyses and graphics added by this project.

Measured missing cells: 0. Identical feature rows: 125, retained because there is no identity key to confirm duplicate sessions. Product-page duration maximum 63,973.5, compared with p99 8,701.1; extremes are reported rather than silently trimmed.

These associations cannot estimate an AI reward effect.

The purchase label Revenue is Boolean, not dollars. No durable IDs, exact timestamps, ordered event stream or transaction amounts are supplied. Retention, ordered funnels, ARPU and experiment lift cannot be estimated. Returning_Visitor is a recorded source category, not a measured repeated visit in this file. PageValues can contain outcome-related information, so it is unsuitable for pre-treatment causal adjustment.

Charts show descriptive subgroup conversion and individual Wilson intervals. Small traffic categories are visibly uncertain. Month names are source categories in an unspecified historical year; they do not describe a current time series. Differences may reflect selection and confounding, and are never transferred to the AI SaaS simulation.

Original CSV SHA-256:
b3055ee355f59134d851d32641183cb4a8b45def7124d2f50442a042f358e0d9

Reproduce with python -m scripts.external and python -m scripts.external_report. Original source, attribution and separate analysis JSON are included in the project package.