C1: Create tfact_grade — course run grade fact table

[quazi-h]

What are the relevant tickets?

Closes #2125
Epic: #2073

Description (What does it do?)

Adds tfact_grade — a new incremental transaction fact table in the dimensional star schema.

Grain: One row per (user × course run) grade record.

Platforms covered: MITx Online, MITx Pro, edX.org

Source models:

• int__mitxonline__courserun_grades — filtered to courserun_platform = 'MITx Online' to exclude edX.org-hosted MITx Online course grades (which are captured via the edxorg source and would otherwise be double-counted)
• int__mitxpro__courserun_grades
• int__edxorg__mitx_courserun_grades — uses courserungrade_user_grade (column name differs from other platforms); timestamps unavailable so grade_date_key and grade_created_on are null for edX.org rows; grade_id is a surrogate from (user_id, courserun_readable_id)

Key implementation details:

• Incremental strategy: delete+insert on grade_key; watermark computed via a pre-computed CTE per platform (avoids Trino correlated-subquery fan-out anti-pattern, consistent with tfact_certificate and tfact_enrollment)
• edX.org rows are fully re-ingested on each incremental run (null timestamps; delete+insert is idempotent)
• Defensive ROW_NUMBER dedup guard (Trino does not support QUALIFY)
• Foreign keys: dim_user, dim_course_run (is_current = true), dim_platform, dim_date
• edX.org rows include MicroMasters-linked grades (micromasters_program_id IS NOT NULL); a separate MicroMasters platform source can be added in a future phase

How can this be tested?

1. Compile: cd src/ol_dbt && ol-dbt run --select tfact_grade (or dbt compile --select tfact_grade)
2. Run against QA: ol-dbt run --select tfact_grade --target dev_qa
3. Validate row counts and pass rates:

SELECT
    platform,
    COUNT(*) AS grade_count,
    ROUND(AVG(CAST(is_passing AS double)), 2) AS pass_rate
FROM ol_warehouse_production_dimensional.tfact_grade
GROUP BY platform;
-- Expected: pass_rate ~0.5–0.7 for mitxonline; edxorg rows have null grade_date_key

4. Verify FK integrity (run dbt tests): dbt test --select tfact_grade

Additional Context

The platform-filtering fix for MITx Online (filtering to courserun_platform = 'MITx Online') was identified during a pre-implementation rubber duck review. int__mitxonline__courserun_grades contains grades for both MITx Online-hosted and edX.org-hosted course runs; without this filter, edX.org course grades would appear twice in the fact table (once from mitxonline, once from edxorg).

This model is a new leaf node — it consumes existing intermediates and has no impact on any existing downstream models (confirmed via OpenMetadata lineage analysis).

[Copilot]

Pull request overview

Adds a new incremental transaction fact table tfact_grade to the dimensional star schema, consolidating course run grades from MITx Online, MITx Pro, and edX.org. The model follows the patterns established by tfact_certificate and tfact_enrollment: per-platform CTEs are unioned, joined to dim_user/dim_course_run/dim_platform, and incrementally loaded via delete+insert keyed on a surrogate grade_key. A pre-computed per-platform watermark CTE avoids correlated-subquery fan-out, and a ROW_NUMBER dedup guard substitutes for Trino's missing QUALIFY.

Changes:

• New model tfact_grade.sql with platform-aware FK joins and an incremental watermark per platform.
• New YAML entry in _fact_tables.yml describing the table, grain, FK relationships, and column-level tests.
• MITx Online source filtered to courserun_platform = 'MITx Online' to prevent double-counting edX.org-hosted MITx Online grades.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/ol_dbt/models/dimensional/tfact_grade.sql	New incremental fact model unioning mitxonline/mitxpro/edxorg grades with FK lookups, watermarked incremental filter, and ROW_NUMBER dedup.
src/ol_dbt/models/dimensional/_fact_tables.yml	Documents tfact_grade columns, relationships, and tests, including unique combination of (grade_id, platform).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rachellougee]

dbt build fine but a bit slow. I left a comment but not a blocker.

[rachellougee]

No need to change it now, but the OR clause in the dim_course_run join here has a performance cost . This was likely copied from tfact_enrollment and tfact_certificate which have the same pattern.

All three platforms could join dim_course_run on (courserun_readable_id, platform) if we populate courserun_readable_id in above CTEs, the same natural key used to build courserun_pk, eliminating the OR entirely.