PBIP Documenter

Generate comprehensive, bidirectional documentation from Power BI PBIP/TMDL semantic models — instantly, in your browser.

No PBIP file? Try the live demo with Contoso sample data — no setup required, runs entirely in your browser.

✨ Updated 2026-04-20 — M-step parser (10 kinds), Value.NativeQuery SQL tracing, BigQuery 3-part source paths, field-parameter + calc-group badges in diagrams, broken-reference visualization, Physical-Source Index in Markdown exports. Full changelog ↓

---

What is PBIP / TMDL?

PBIP (Power BI Project) is a developer-friendly file format introduced in Power BI Desktop that stores your semantic model and reports as plain text files instead of a binary .pbix. It integrates with Git and CI/CD pipelines.

TMDL (Tabular Model Definition Language) is the text-based format within PBIP that describes every table, column, measure, relationship, and role in your semantic model — one .tmdl file per object, readable and diffable in any editor.

If you've enabled "Save as Power BI Project" in Power BI Desktop (Preview → Developer mode), your workspace folder already contains PBIP files ready for this tool.

---

Who is this for?

Power BI Developer

You write DAX measures and want to know:

• What tables and columns does each measure reference?
• Where (which pages, which visuals) is each measure displayed?
• Which measures depend on other measures?
• Are my field parameters and calculation groups visually distinct in the model diagram?
• Are any DAX references broken (renamed or deleted columns)?

What you get: Measure Catalog with full DAX + referenced columns/measures, "Used in Visuals" grouped by page, measure dependency chains, field parameter (purple) and calculation group (brown) header badges on diagrams, and red-dashed broken-reference indicators across all lineage views.

Data Engineer

You own the source systems and want to know:

• Which physical tables (e.g. dbo.FactSales) were loaded into the model?
• Were columns renamed between source and model?
• Which DAX measures and report visuals ultimately consume each source table or column?
• What do the M queries actually do, step by step?
• Which physical columns does each model column map to — even without a rename?
• Does the model use Value.NativeQuery or Table.NestedJoin/Table.Combine?

What you get: Expanded Data Sources view showing physical→model table mapping, column renames, and a "Where Used" catalog per model column. Plus: a 10-kind M-step breakdown per table, Value.NativeQuery SQL preview with extracted table references, BigQuery project.dataset.table source paths, physical-column lineage nodes for every non-calc column, per-visual back-trace to physical schema.table, and a Physical-Source Index in Markdown exports.

Product Owner / Manager

You need the big picture:

• How large is this model? How many measures, tables, visuals, data sources?
• What are the most-used measures and source tables?
• Are there any dynamic features (field parameters, calculation groups) that behave differently than PBIR JSON suggests?

What you get: Executive Summary at the top of every export with model stats, top measures by visual coverage, and top source tables by consumption.

---

Quick Start

1. Open the tool: jonathanjihwankim.github.io/pbip-documenter
2. Select your persona (Power BI Developer / Data Engineer / Product Owner) — the app highlights the most relevant section after parsing
3. Click Open Project Folder and select your PBIP project folder
4. The tool auto-discovers .SemanticModel and .Report folders
5. Browse the parsed model in the sidebar — tables, measures, relationships, data sources, visuals, and more
6. Download documentation:
   • Full Report (.html) — self-contained with DAX highlighting, collapsible sections, column usage, data source drill-down
   • Full Report (.md) — clean Markdown with tables, ASCII layout grids, ideal for Git wikis
   • JSON — machine-readable with whereUsed blocks per column and consumers blocks per data source

Manual documentation vs. PBIP Documenter

	Manual	PBIP Documenter
10 tables, 11 measures, 15 visuals	~45 minutes	< 10 seconds
Source → model column lineage	Spreadsheet by hand	Auto-detected from M queries
Visual lineage tracing	Not feasible	Built-in
Relationship diagrams	Draw by hand	Auto-generated SVG
Keeps up with model changes	Start over	Re-run instantly
Privacy	Varies	100% client-side

---

What You Get

Point the tool at your PBIP project folder and get professional, bidirectional documentation:

For Power BI Developers (forward view)

• Measure Catalog — DAX expressions with syntax highlighting, display folders, format strings, referenced columns and measures, "Used in Visuals" by page
• Table Inventory — columns with data types, descriptions, sort-by, summarize-by, and hidden status
• Relationships — from/to columns, cardinality, cross-filter direction, active/inactive
• Roles — permission levels and RLS filter expressions per table
• Diagram Legibility — field parameters (purple) and calculation groups (brown) labeled distinctly; inactive relationships dashed; parallel edges between the same tables offset so they don't overlap; broken DAX references flagged with red dashed border and ⚠ icon in all lineage views

For Data Engineers (reverse view)

• Data Sources — expanded view with physical table names (schema + table from Navigation steps), Power Query column renames, computed columns, and full consumer catalog (measures + visuals + pages). Text search + connector-type / Gateway / Parameterized chip filters. Value.NativeQuery shown with Native SQL badge + collapsible SQL preview.
• M-Step Breakdown — every let…in block decomposed into typed steps (Source / Navigation / Projection / Rename / Filter / Join / AddColumn / TypeChange / Expand / Custom) with refs, rendered as a numbered list with colored kind badges per table
• Column Usage (Where Used) — per table, every visible column shows which measures reference it and which visuals display it
• Source Trace Lineage — click a physical source table to open a forward lineage diagram: source → model table → measures → visuals
• Physical-Column Lineage — every non-calc model column has a first-class physicalColumn node; Column Impact shows the upstream physical source; BigQuery lineage captures the full project.dataset.table path

For Product Owners (summary view)

• Executive Summary — model stats at a glance: tables, measures, relationships, pages, visuals, data sources, dynamic features, broken references
• Top Measures — ranked by number of visuals they appear in
• Top Source Tables — ranked by downstream visual coverage
• Dynamic Features — field parameters and calculation groups that PBIR JSON doesn't fully represent

Interactive Diagrams

• Relationship Diagram — SVG with pan, zoom, and zoom-to-fit; star-schema layout. Field parameters shown with purple headers, calculation groups with brown headers, inactive relationships dashed, and parallel edges between the same table pair offset so they don't overlap
• Visual Lineage — full model, visual trace (including new "Physical Columns" column), measure impact, column impact (including upstream physical source), and source trace modes. Broken field references shown with red dashed border + ⚠ icon
• Visual Usage Diagram — field-to-visual mapping

Export

• HTML — fully self-contained, embeds CSS + SVG, DAX syntax highlighting, collapsible sections, table of contents
• Markdown — clean document with fenced DAX blocks, ASCII page layout grids, structured tables, Physical-Source Index (schema.table → model tables → consumer counts), and per-visual back-trace (visual → fields → model tables → physical schema.table → renames → first M step)
• JSON — machine-readable with whereUsed and consumers blocks for downstream tooling

Your files never leave your browser. All parsing happens client-side — nothing is uploaded anywhere.

---

What's New (2026-04-20) {#whats-new-2026-04-20}

Four commits over two days closed 28 of 30 findings from a multi-perspective audit across enterprise BigQuery (61 tables), Contoso (10 tables), and demo datasets. Full findings: docs/audit-2026-04-20.md.

For Power BI developers

• ERD cardinality fixed — draw.io and Mermaid exports previously defaulted every relationship to many-to-one. Now reads the correct fromCardinality/toCardinality from the TMDL parser.
• FP + CG badges in overview diagram — field parameter tables show a purple header, calculation group tables show brown. Before this, prm_measures and This Year Last Year looked identical to ordinary tables.
• Enterprise layout rebuilt — per-ring radius now sized to each ring's occupancy; 20 collision passes with 0.8 damping; parallel edges between the same table pair get a perpendicular offset so they no longer overlap. 61-table models no longer collapse into a blob.
• Broken references visualized — visual nodes with broken DAX/column references get a red dashed border + ⚠ icon across full lineage, visual trace, and impact diagrams.
• Dataset-switch state reset — _resetState() called at the top of parseModel() and loadSampleData(): clears diagram containers, lineage selects, stale warning banners, and the static M-parser cache. Switching between datasets no longer leaks stale state.

For data engineers

• M-step parser (10 kinds) — every let…in block split into typed steps: Source / Navigation / Projection / Rename / Filter / Join / AddColumn / TypeChange / Expand / Custom. Rendered as a numbered list with colored kind badges in Table Detail.
• Value.NativeQuery first-class — native SQL detected, backing connector extracted, SQL parsed by a FROM/JOIN tokenizer that emits sqlTableRefs into the lineage graph. Rendered with a "Native SQL" badge + collapsible preview in the Data Sources view.
• Table.NestedJoin / Table.Combine captured — join metadata lifted into result.joins[] with step names and key columns, feeding the M-step graph.
• BigQuery 3-part paths fixed — chained {[Name="project"]}[Data]{[Name="dataset"]}[Data]{[Name="table"]}[Data] now walks all three levels. Previously only the last Name= value survived; enterprise BigQuery lineage now shows project + dataset + table.
• Physical columns as first-class lineage nodes — physicalColumn nodes previously only created on renames; now emitted for every passthrough non-calc column too, with maps_to_physical_column edges. Column Impact gained an upstream "Physical Source" column.
• Calc-column + calc-table DAX wired — calculated columns emit references_column/references_measure edges; calculated-table partitions emit derived_from_table edges. Previously produced zero lineage edges.
• Per-visual back-trace in exports — Markdown and HTML documents now walk visual → fields → model tables → physical schema.table → renames count → first M step (name + kind), replacing the old one-liner summary.
• Physical-Source Index in Markdown — new section: schema.table → model tables → consumer counts for each physical source.

---

Browser Support

Requires the File System Access API:

Browser	Support
Chrome 86+	✅ Supported
Edge 86+	✅ Supported
Opera 72+	✅ Supported
Firefox	❌ Not supported
Safari	❌ Not supported

---

PBIP Folder Structure

The tool expects a standard PBIP project layout:

Show folder structure

MyProject/
├── MyProject.SemanticModel/
│   └── definition/
│       ├── database.tmdl
│       ├── model.tmdl
│       ├── relationships.tmdl
│       ├── expressions.tmdl        (optional — shared M expressions / parameters)
│       ├── tables/
│       │   ├── Sales.tmdl
│       │   ├── Product.tmdl
│       │   └── ...
│       └── roles/                  (optional)
│           └── Reader.tmdl
└── MyProject.Report/               (optional — enables visual analysis)
    └── definition/
        └── pages/
            └── Page1/
                ├── page.json
                └── visuals/
                    └── visual1/
                        └── visual.json

---

Also by Jihwan Kim

Tool	Description
PBIR Visual Manager	Manage isHiddenInViewMode and visual properties in PBIR reports
PBIP Impact Analyzer	Analyze what breaks when you change a measure, column, or table
PBIP Documenter	Generate bidirectional documentation from TMDL (you are here)

---

Support Development

This tool is free forever — built and maintained solo by Jihwan Kim (Microsoft MVP). If PBIP Documenter saves you even 30 minutes of documentation work, please consider sponsoring.

Sponsor Tiers

Tier	Amount	Recognition
Gold	50+ EUR/mo	Logo + link on README and app footer
Silver	10+ EUR/mo	Name + link on README + shoutout in release notes
Bronze	Monthly supporter	Name + link on README
Coffee	One-time	Name listed in Hall of Sponsors

Hall of Sponsors

Name	Tier
Alessandro Tiberti Bertin	Bronze

---

License

MIT — Jihwan Kim

Built for: Power BI · Microsoft Fabric · PBIP · PBIR · TMDL · Semantic Models · DAX · Data Governance · CI/CD · Developer Tools