Review the semantic model structure and DAX, report findings by category, and
propose corrected DAX where relevant.
Instructions
Get the model definition. Accept a TMDL folder/export, a .bim/JSON model
definition, or a pasted description of tables, columns, relationships, and
measures. If given only a screenshot or vague description, ask for the
relationships and measure list directly. A review needs the actual
expressions, not a summary of them.
State this limit up front, in the first response: this is a static review
of the model definition. It cannot connect to the live model, cannot see
data volumes or cardinality, and cannot measure real query performance.
Findings about likely performance impact are structural inference, not
profiling results, so say so.
Check the model against each category below. Skip a category cleanly if the
model doesn't contain what it checks (e.g. no bi-directional relationships
to review).
Report findings grouped by category, worst-impact first:
| Category |
Object |
Issue |
Fix |
| Relationships |
Sales to Customer |
Bi-directional, both dimension tables |
Set to single-direction; use CROSSFILTER(..., BOTH) inside the one measure that needs it |
For any DAX fix, show the corrected expression in full, not a diff.
On request, rewrite the flagged DAX or produce a prioritized backlog the
user can hand to whoever maintains the model.
Review categories
Relationships
- Bi-directional relationships between two dimension tables, or wherever a
single-direction relationship plus a
CROSSFILTER(..., BOTH) inside the one
measure that needs it would give the same answer with less filter-context
ambiguity across the rest of the model.
- Many-to-many relationships without a documented reason.
- Inactive relationships with no
USERELATIONSHIP reference anywhere in the
model's measures. That's dead weight.
- Missing or inconsistent relationship cardinality (e.g. many-to-many where
one-to-many is intended).
Calculated columns that should be measures
- A calculated column doing row-by-row aggregation logic that a measure would
compute at query time instead. It costs storage and compression for no benefit.
- A calculated column duplicating a value already derivable via a measure or a
Power Query step upstream.
Date handling
- No dedicated date table, or a date table not marked as a date table.
- Time-intelligence functions (
TOTALYTD, SAMEPERIODLASTYEAR, etc.) applied
against a column that isn't contiguous or isn't a proper date column.
- Role-playing date scenarios (order date vs. ship date) handled by duplicating
measures instead of duplicating the date table.
Naming and formatting
- Table, column, or measure names that are cryptic (
T1, CustNo, Amt)
instead of self-explanatory.
- Inconsistent data types for the same concept across tables (a date stored as
text in one table, a real date in another).
- Missing or inconsistent format strings on the same kind of measure (currency,
percentage).
- No descriptions on key measures. Harmless for report authors, but it also
means Copilot in Power BI can't ground answers on that measure correctly.
Star schema shape
- Fact tables mixed with dimension attributes in the same table (snowflake or
fully flat design where a star schema would simplify filtering).
- Missing surrogate keys or relationships built on unstable natural keys.
DAX correctness and clarity
CALCULATE with filter arguments that silently override intended context.- Iterators (
SUMX, FILTER) used where a plain aggregation would do.
- Division without
DIVIDE(), risking divide-by-zero errors.
- Measures that reimplement logic another measure already provides. That's a
correctness risk if the two drift.
Guardrails
- Never claim to have measured performance. Frame everything as "based on the
model definition" or "typically causes."
- Never invent table row counts, data volumes, or refresh times not given by
the user.
- Note whichever storage mode applies (Import, DirectQuery, Composite,
Direct Lake) if it's stated or evident, since it changes what actually
matters: a bi-directional relationship is cheap in one mode and expensive
in another.
Tone
Direct, structural. State the issue, the why, and the fix. Skip the theory
lecture unless asked.
1---2name: power-bi-model-review3description: Use this skill whenever the user shares a Power BI or Analysis Services semantic model (a TMDL export, a BIM/JSON model definition, or a pasted list of tables, relationships, and measures) and asks for a model review, a performance check, or help fixing DAX, before writing or correcting any DAX measure for that model.4---56Review the semantic model structure and DAX, report findings by category, and7propose corrected DAX where relevant.89## Instructions10111. Get the model definition. Accept a TMDL folder/export, a `.bim`/JSON model12 definition, or a pasted description of tables, columns, relationships, and13 measures. If given only a screenshot or vague description, ask for the14 relationships and measure list directly. A review needs the actual15 expressions, not a summary of them.16172. State this limit up front, in the first response: this is a **static review18 of the model definition**. It cannot connect to the live model, cannot see19 data volumes or cardinality, and cannot measure real query performance.20 Findings about likely performance impact are structural inference, not21 profiling results, so say so.22233. Check the model against each category below. Skip a category cleanly if the24 model doesn't contain what it checks (e.g. no bi-directional relationships25 to review).26274. Report findings grouped by category, worst-impact first:2829 | Category | Object | Issue | Fix |30 | --- | --- | --- | --- |31 | Relationships | Sales to Customer | Bi-directional, both dimension tables | Set to single-direction; use `CROSSFILTER(..., BOTH)` inside the one measure that needs it |3233 For any DAX fix, show the corrected expression in full, not a diff.34355. On request, rewrite the flagged DAX or produce a prioritized backlog the36 user can hand to whoever maintains the model.3738## Review categories3940**Relationships**41- Bi-directional relationships between two dimension tables, or wherever a42 single-direction relationship plus a `CROSSFILTER(..., BOTH)` inside the one43 measure that needs it would give the same answer with less filter-context44 ambiguity across the rest of the model.45- Many-to-many relationships without a documented reason.46- Inactive relationships with no `USERELATIONSHIP` reference anywhere in the47 model's measures. That's dead weight.48- Missing or inconsistent relationship cardinality (e.g. many-to-many where49 one-to-many is intended).5051**Calculated columns that should be measures**52- A calculated column doing row-by-row aggregation logic that a measure would53 compute at query time instead. It costs storage and compression for no benefit.54- A calculated column duplicating a value already derivable via a measure or a55 Power Query step upstream.5657**Date handling**58- No dedicated date table, or a date table not marked as a date table.59- Time-intelligence functions (`TOTALYTD`, `SAMEPERIODLASTYEAR`, etc.) applied60 against a column that isn't contiguous or isn't a proper date column.61- Role-playing date scenarios (order date vs. ship date) handled by duplicating62 measures instead of duplicating the date table.6364**Naming and formatting**65- Table, column, or measure names that are cryptic (`T1`, `CustNo`, `Amt`)66 instead of self-explanatory.67- Inconsistent data types for the same concept across tables (a date stored as68 text in one table, a real date in another).69- Missing or inconsistent format strings on the same kind of measure (currency,70 percentage).71- No descriptions on key measures. Harmless for report authors, but it also72 means Copilot in Power BI can't ground answers on that measure correctly.7374**Star schema shape**75- Fact tables mixed with dimension attributes in the same table (snowflake or76 fully flat design where a star schema would simplify filtering).77- Missing surrogate keys or relationships built on unstable natural keys.7879**DAX correctness and clarity**80- `CALCULATE` with filter arguments that silently override intended context.81- Iterators (`SUMX`, `FILTER`) used where a plain aggregation would do.82- Division without `DIVIDE()`, risking divide-by-zero errors.83- Measures that reimplement logic another measure already provides. That's a84 correctness risk if the two drift.8586## Guardrails8788- Never claim to have measured performance. Frame everything as "based on the89 model definition" or "typically causes."90- Never invent table row counts, data volumes, or refresh times not given by91 the user.92- Note whichever storage mode applies (Import, DirectQuery, Composite,93 Direct Lake) if it's stated or evident, since it changes what actually94 matters: a bi-directional relationship is cheap in one mode and expensive95 in another.9697## Tone9899Direct, structural. State the issue, the why, and the fix. Skip the theory100lecture unless asked.
Microsoft