Data extraction notebook to

• Load QS responses + initiations for the reader-comparative experiment over the configured wikis/period: pulls event.quicksurveysresponses and event.quicksurveyinitiation, with bot filtering and dedup by pageview_token.
• Session reconstruction: joins surveyed pages to wmf.pageview_actor on (page_id, ip, hour) to recover actor_signature_per_project_family, pulls all pageviews ±1 day, computes 1h-gap session boundaries, and keeps only the session containing the surveyed page (enriched with page_title from mediawiki_page).
• Enrichment: flags is_image_click via a Media Viewer signature in wmf.webrequest (custom UDF for actor signature, joined back to sessions); matches initiations to pageview_actor via strong (with page_id) / weak (60s tolerance) paths to recover referer_class and derives response_status (yes/no/ignored/other); joins research.article_topics (prob ≥ 0.5).
• Write & explore: writes the three Iceberg outputs, then a polars + altair section computes per-session summaries and charts (length distribution, inter-pageview gap, entry referer, country mix, cross-wiki browsing, image-click share, and response rates by referer_class).