[Feature] Support metric_time in Metric Filters with Parent Query Granularity Inheritance

[wtremml18]

Is this your first time submitting a feature request?

• I have read the expectations for open source contributors
• I have searched the existing issues, and I could not find an existing issue for this feature
• I am requesting a straightforward extension of existing metricflow functionality, rather than a Big Idea better suited to a discussion

Describe the feature

Summary

Enable the ability to specify metric_time as part of a metric filter so the filter joins on the same granularity of the parent metric. This ensures metric filters respect the time context of the query.

Motivation

Currently, metric filters can reference a single entity or dimension, but they do not have a way to inherit the granularity of the query. This creates a core issue where the parent metric grouped by a time-granularity (e.g., mont), but the filter is based on "all time" data.

Calculating something like 'Product Usage by Paid Users group by Day/Week/Month' is not possible with the current filter implementation. SCDs are not a viable approach because they can't dynamically adjust to the query timeframe and add unnecessarily complex SQL logic to the compiled code.

By allowing metric_time to be included in metric filters, we can:

• Ensure that metric filters operate at the same time granularity as the metric being queried.
• Maintain backward compatibility by only applying this behavior when metric_time is explicitly included.

Proposed Behavior

• When metric_time is included in a metric filter group-by, the filter is applied using the same time granularity as the queried parent metric.
• If metric_time is not specified in the metric filter, existing behavior remains unchanged.
• If metric_time is included but the parent metric does not have a time dimension, it is silently ignored.

Use Case Example

Consider the question "How many SMS Customers are sending Emails by month":

- name: distinct_paid_sms_email_senders
    label: SMS Customers sending Emails
    type: derived
    type_params:
      expr: paid_sms_email_senders
      metrics: 
        - name: distinct_paid_sms_accounts
        - alias: paid_sms_email_senders
          filter: |
            {{ Metric('sum_emails_sent', group_by=['account_id']) }} > 0

The data will change post-facto if an account starts sending emails because the filter is applied over all time. This yields a meaningless metric with historical values that constantly fluctuate.

By changing the filter to {{ Metric('sum_emails_sent', group_by=['account_id','metric_time']) }} > 0, this allows analytics to create a metric that is historically static and outputs the intended results of "How many Paid SMS accounts sent emails last June".

Describe alternatives you've considered

1. SCD II would be an alternative approach, but adds a ton of complexity. It would also yield analytically unintuitive results, since the SCD would likely be at a smaller granularity. Entities are typically aggregated if they were active at any point during the month, (e.g., "Monthly Active Users") and an SCD may unintentionally filter values if the dimension is stored at a daily granularity.
2. The other alternative is to hard-calculate the metric in an intermediate table. This does not scale and leads to conflicting results when you calculating common metric-patterns like "per paid user".

Who will this benefit?

This will benefit any analyst who is using the Semantic Layer to create dashboards or uses core metrics as part of query filters.

Are you interested in contributing this feature?

Yes - PR 1658. Likely NOT production ready

Anything else?

No response