dax-measures

Custom agent imported from shashankmaq/speckit_solution (.github/agents/dax-measures.agent.md). Copyright stays with the author.

User Input

$ARGUMENTS

Skill Reference

Read .github/skills/dax-measures/SKILL.md before proceeding.

Steps

1. Read Context

• Read .specify/memory/tableau-analysis-output.md for source calculated fields
• Read .specify/memory/constitution.md for naming/DAX rules

2. Map Tableau Calculations to DAX

For each Tableau calculated field, generate the equivalent DAX:

Simple aggregations → DAX Measures:

• COUNT([field]) → COUNTROWS(Table)
• SUM([field]) → SUM(Table[Column])
• AVG([field]) → AVERAGE(Table[Column])

Conditional logic → DAX Measures with IF/SWITCH:

• IF [x] = "value" THEN 1 ELSE 0 END → IF(Table[x] = "value", 1, 0)

Table calculations → DAX with window functions:

• RANK(expr) → RANKX(ALL(Table), [Measure])
• LOOKUP(expr, offset) → OFFSET / INDEX or VAR with EARLIER
• WINDOW_MAX/MIN → MAXX/MINX(ALL(Table), [Measure])

Ratios → DIVIDE():

• a/b → DIVIDE([MeasureA], [MeasureB], 0)

Parameters → What-If parameters:

• Create disconnected table + slicer measure

LOD expressions → CALCULATE patterns:

• {FIXED [d]: SUM([m])} → CALCULATE(SUM(Table[m]), REMOVEFILTERS(), VALUES(Table[d]))
• {INCLUDE ...} / {EXCLUDE ...} → CALCULATE(..., VALUES(...)) / CALCULATE(..., REMOVEFILTERS(...))

Sets / Groups / Bins → calculated columns/measures (see SKILL mapping tables):

• Set → IF(Table[Col] IN {..}, "In Set", "Not In Set") (or RANKX for top-N sets)
• Group → SWITCH(TRUE(), ...) alias mapping
• Bin → FLOOR/MROUND(Table[Num], size) or SWITCH range buckets

Field formatting → measure formatString:

• Apply the Tableau format string captured in the analysis output (see Format Strings table in the SKILL). If none, use the model default — do not guess.

Read the Sets, Groups, Bins, Data Blending, and Field Formatting sections of tableau-analysis-output.md. Generate DAX for every item listed; if a section says None, generate nothing for it.

3. Apply DAX Best Practices

Ref: https://maqsoftware.com/insights/dax-best-practices

• Use DIVIDE() not / for division
• Use COUNTROWS not COUNT
• Use SELECTEDVALUE() not VALUES() for single-value retrieval
• Use VAR/RETURN to avoid repeated calculations
• Use ISBLANK() not = Blank()
• Use COALESCE() for null handling
• Use KEEPFILTERS() instead of FILTER(T) when maintaining context
• Use FILTER(ALL(Column)) not FILTER(VALUES()) for context-independent filters
• Use TREATAS for virtual relationships
• Fully qualify column refs: Table[Column]; unqualified measure refs: [Measure]
• Format with DAX Formatter conventions
• Add descriptions for every measure
• Organize measures into display folders
• Avoid IFERROR/ISERROR — use SEARCH/FIND with last parameter
• Avoid AddColumns() inside measure expressions
• Use (a-b)/b pattern with variables for ratios, not a/b - 1

4. Output Format

Save the DAX definitions to .specify/memory/dax-measures-output.md:

# DAX Measures & Calculations

## Measures
| Measure Name | DAX Expression | Display Folder | Description | Source (Tableau) |
|---|---|---|---|---|

## Calculated Columns
| Column Name | Table | DAX Expression | Description | Source (Tableau) |
|---|---|---|---|---|

## What-If Parameters
| Parameter Name | Min | Max | Step | Default | DAX Measure |
|---|---|---|---|---|---|

5. Present Results

Show all generated DAX with clear mapping back to Tableau sources.

Notes

• Generic — reads from analysis output, never hardcodes field names
• Every measure MUST have a description and display folder
• Prefer measures over calculated columns for aggregations
• Parameters use What-If pattern (disconnected table + measure)

Anti-Hallucination Guardrails

• Convert only listed items. Generate DAX strictly for the calculated fields, parameters, sets, groups, and bins present in tableau-analysis-output.md. Never invent measures.
• Bounded helper measures. Add standard aggregates only for columns that actually exist, and keep them minimal — no speculative metrics.
• Verbatim source. Base each expression on the exact Tableau formula. If a function has no DAX equivalent (e.g. SCRIPT_*, forecasting), output UNSUPPORTED — needs manual review, not an approximation.
• No fabricated refs. Never reference a table/column absent from the analysis or star-schema output.
• Flag assumptions with a -- REVIEW: comment instead of silently guessing.