Drag the canvas. Hover a table or field to inspect it.
46%
dim_dateDimension
One calendar date
dim_timeDimension
One intraday time bucket
dim_videoDimension
One historical version of a video
dim_channelDimension
One historical version of a channel
dim_viewerDimension
One privacy-scoped viewer version
dim_deviceDimension
One device/application capability combination
dim_geoDimension
One policy-approved geography member
dim_surfaceDimension
One product placement surface
fact_playback_eventFact
One accepted playback event
fact_watch_sessionFact
One playback attempt and session version
fact_impressionFact
One eligible video impression
fact_engagement_eventFact
One accepted engagement action
fact_search_eventFact
One search query or result interaction
fact_ad_eventFact
One ad lifecycle event
fact_ad_impressionFact
One eligible ad impression
fact_qoe_eventFact
One playback-quality incident or sample
fact_video_metric_dailyFact
Video × date × metric version
fact_channel_metric_dailyFact
Channel × date × metric version
fact_live_concurrencyFact
Live stream × region × minute
fact_experiment_assignmentFact
One subject × experiment assignment
fact_shorts_sessionFact
One Shorts feed viewing attempt
fact_audience_retentionFact
Video × segment bucket × metric version
dim_experimentDimension
One versioned experiment definition
dim_adDimension
One ad creative/campaign version
dim_content_formatDimension
One governed content-format definition
bridge_identity_linkBridge
One permitted identity link and validity interval
metric_registryRegistry
One metric definition version
02
Interview Answer
I start by defining the grain of every fact so one row has an unambiguous meaning—for example, fact_watch_session is one playback attempt and version, while fact_video_metric_daily is one video, date, and metric version. I connect those facts to conformed date, video, channel, viewer, device, geography, surface, experiment, and ad dimensions. Video and channel use Type 2 history so facts retain the metadata and monetization state valid when the event occurred. Identity links remain consent-aware and reversible. Live, Shorts, and VOD share dimensions but use separate facts where concurrency, loops, swipes, and retention curves have different semantics. Finally, a versioned metric registry owns formulas such as qualified views instead of duplicating logic across dashboards.
