Join and Relationship Advisor
Recommends join versus relationship, diagnoses cardinality mismatches, and predicts row-explosion or measure-doubling before it silently corrupts your dashboard. Advisory-only. Scoped to two-table pairs.
Example: You point it at Orders and Returns. It samples uniqueness on both sides of Order ID, classifies the pair as many-to-many, recommends a relationship (not a join), and flags that SUM(Sales) will double because both tables are many-to-many via Order ID. Routes the fix to calc-writer with the LOD pattern name.
What it does
Compresses the "why is my measure doubling" investigation from an afternoon of scratch-sheet forensics to a single named diagnosis with a fix path. Reads the workbook, samples cardinality, checks the locked decision tree, and reports.
When to use it
- "Recommend relationships for these tables."
- "Should this be a join or a relationship?"
- "Why is SUM(X) doubling?"
- "Grain check on Orders and Returns."
- "Fanning risk on this data source."
When NOT to use it
- Connection setup (use
data-source-connector). - Row-level data quality (use
data-quality-review). - LOD authoring (use
calc-writer). - Model documentation (use
data-model-documenter). - Change tracking (use
workbook-auditor).
What you get back
- Recommendation per table pair: join vs. relationship, cardinality, one-line rationale.
- Row-explosion or fanning-risk warnings against shelf-placed measures.
- Desktop UI fix instructions.
- LOD-fix handoff routed to
calc-writerby exact slug.
Limitations
- Two-table pairs only.
- Cardinality inferred from
get-summary-datasampling. Extract filters may skew large-table samples. - Cannot detect fanning in calcs that construct measures at query time. Only on shelf-placed measures.
- Never modifies the data model. Every fix is a Desktop UI instruction.