[Bug] Time truncation with conversion metrics

[wtremml18]

Is this a new bug in metricflow?

• I believe this is a new bug in metricflow
• I have searched the existing issues, and I could not find an existing issue for this bug

Current Behavior

When I query sl query --metrics 7d_conversion_metric --group-by metric_time__month, the monthly granularity overrides the conversion metric and everything is trunced within the calculation. This should not be the intended behavior.

We essentially want to produce 7 day conversion grouped at the monthly level. As an added bonus, we would also want to exclude accounts where the conversion period is incomplete (e.g., base time < today() - conversion window) [pretty easy from a filter perspective, but should be default behavior]

Here is a snippet of code produced from the sl query

INNER JOIN
          product_funnel_events product_funnel_events_src_10000
        ON
          (
            subq_11.account_id = product_funnel_events_src_10000.account_id
          ) AND (
            (
              subq_11.metric_time__month <= DATE_TRUNC('month', product_funnel_events_src_10000.first_event_time)
            ) AND (
              subq_11.metric_time__month > DATEADD(day, -7, DATE_TRUNC('month', product_funnel_events_src_10000.first_event_time))
            )
          )

You can see that the conversion granularity isn't being respected because of the monthly grouping

Here is the metric definition for additional context

  - name: activation_7d
    description: "Conversion rate from signup to activation in 7 days"
    type: conversion
    label: Visit to buy conversion rate (7-day window)
    type_params:
      conversion_type_params:
        entity: account_id
        base_measure:
          name: num_signups
          fill_nulls_with: 0
          join_to_timespine: true
          filter: |
            {{ Dimension('pk_onboarding_activation__website_status') }} = 'LIVE' 
            and {{ Dimension('pk_onboarding_activation__spam_flag') }} = false
        conversion_measure:
          name: num_accounts_activated #Change this to a Filter instead of CASE WHEN
          fill_nulls_with: 0
          join_to_timespine: true
        window: 7 days

The query should function more like this (intentionally different syntax for clarity)

SELECT
      DATE_TRUNC('month', signup_time) AS metric_time__month
      , count(distinct base.account_id) as num_signups
      , count(distinct case when datediff('day', signup_time, activation_time) <=7 then 1 else 0 end AS num_activations
      , num_activations / num_signups as activation_7d
    FROM base
    LEFT JOIN conversion
        ON base.account_id = conversion.account_id
        AND signup_time <= activation_time
        AND dateadd('day', 7, signup_time) > activation_time
        AND signup_time <= dateadd('day', -7, getdate()) -- or max date of available data

Expected Behavior

If the conversion granularity is "days", grouping by a different granularity shouldn't impact the conversion calculation. It should aggregate by the base metric's time dimension.

Steps To Reproduce

1. Create conversion metric with 7 day conversion
2. Query using --group_by metric_time__month --compile

Relevant log output

Environment

- OS: Mac
- dbt: dbt_cloud
- metricflow: CLI version 0.38.28

Which database are you using?

snowflake

Additional Context

No response

[plypaul]

@wtremml18 This might have been resolved by earlier work - I created a repro case in our test suite (#1899), but I don't see the behavior you're describing. Mind taking a look at the PR and reopening this issue if that's not the case?