Hello everyone, Is there a problem with user filter? this metric in the screenshot is supposed to have the percentage of users with at least 2 purchases, but the results is the count of all purchases. Why is that? Thank you in advance!

@future-teacher-7046 or @helpful-application-7107 can you take a look?

narcos-waiting.jpg

I think this metric should be fine? I can't replicate the error. Can you show the results you're getting that indicate that the result is the count of all purchases? Can you run a metric analysis on the fact metric page?

Hello @helpful-application-7107 here you can find a screenshot of a recent experiment where the metric fails. Also how it's defined in the fact table, and for some reason on the metric page itself, there is a capping settings but it's not there when you click edit (because it is not count so it makes sense, but I don't know why the settings appeares like it's a count). Also, the metric analysis shows 323% while it's binomial 😄

Thanks this is helpful, I can try to take a look today.

great, thank you for taking the time!

It does look like a bug...

you're right, I am self-hosting the version 3.4 and the query is missing the user filter as you can see here.

__factTable as ( -- Fact Table (purchases)
    SELECT
      user_id as user_id,
      CAST(m.timestamp as DATETIME) as timestamp,
      -- FT-GB: Transaction (binomial) (desktop)
      CASE
        WHEN (
          purchase_device = 'desktop'
          AND transaction_type != 'cc_validation'
        ) THEN 1
        ELSE NULL
      END as m0_value,
      -- FT-GB: 2nd Transaction (after exposure) (binomial) (desktop)
      CASE
        WHEN (
          transaction_type != 'cc_validation'
          AND purchase_device = 'desktop'
        ) THEN 1
        ELSE NULL
      END as m1_value,
      -- FT-GB: 3rd Transaction (after exposure) (binomial) (desktop)
      CASE
        WHEN (
          transaction_type != 'cc_validation'
          AND purchase_device = 'desktop'
        ) THEN 1
        ELSE NULL
      END as m2_value,
      -- FT-GB: >3 Transactions (after exposure) (binomial) (desktop)
      CASE
        WHEN (
          transaction_type != 'cc_validation'
          AND purchase_device = 'desktop'
        ) THEN 1
        ELSE NULL
      END as m3_value,
      -- FT-GB: Transactions (count) (desktop)
      CASE
        WHEN (
          purchase_device = 'desktop'
          AND transaction_type != 'cc_validation'
        ) THEN 1
        ELSE NULL
      END as m4_value,
      -- FT-GB: Transactions Credits (count) (desktop)
      CASE
        WHEN (
          purchase_device = 'desktop'
          AND transaction_type != 'cc_validation'
        ) THEN m.package_value
        ELSE NULL
      END as m5_value
    FROM
      (
        SELECT
          user_id,
          user_pseudo_id as anonymous_id,
          TIMESTAMP_MICROS(event_timestamp) AS timestamp,
          purchase_device,
          item_category,
          transaction_type,
          package_value,
          purchase_source
        FROM
          `fact.purchases`
        WHERE
          event_date BETWEEN PARSE_DATE('%Y%m%d', '20241212') AND PARSE_DATE('%Y%m%d', '20250311')
      ) m
    WHERE
      m.timestamp >= '2024-12-12 08:36:00'
      AND m.timestamp <= '2025-03-11 08:09:17'
  ),

Yes, unfortunately you'll need to upgrade to use user filters. We try not to ship versioned numbers with bugs like this but this one slipped through the cracks.

Thank you, Luke, for checking it out. I will ask our developers to update it. Have a great day!

What SQL engine are you using? sorry about this!

no problem. We are using BigQuery

Hello 🙂 Joining the thread to continue this topic, as Ahmed will be ooo for next 2 weeks.