Pulse Path's reporting owner needs a defensible active-subscription dataset. Build current subscription evidence, retain unknown plans, and distinguish subscription count from user count before recommending the metric definition.
Pulse Path's operations and finance teams report different active-account totals. One query reads every raw subscription version, another removes missing plan names, and a third counts users rather than subscriptions.
Those results answer different questions. You need a model whose grain and missing-value rules are explicit before deciding which count belongs in the handoff.
Use one current row per non-NULL subscription_id, normalize status, and keep an unknown plan as an explicit category. Then build evidence at subscription, user, and plan grain. Keep the measures named precisely.
A user may hold more than one subscription, so a distinct-user count isn't a substitute for a subscription count. The final checked recommendation uses those definitions rather than whichever total happens to be lower.
select subscription_id, user_id, coalesce(lower(trim(plan)), 'unknown') as plan, lower(trim(status)) as status, monthly_amount, updated_at from {{ source('pulse', 'subscriptions') }} where subscription_id is not null qualify row_number() over (partition by subscription_id order by updated_at desc)=1
A worked pattern with fictional records. Adapt the model name and business rule to each task.