eMondrian Docs

DAX in eMondrian is out of beta: what works in Power BI, and what doesn't

6 min read

As of November 2026: eMondrian Community 2026.11.09.2, modules 0.0.0.102.

Since Community 2026.11.07.1, the DAX module is no longer marked beta. That’s the part of eMondrian that answers Power BI over a live connection: every visual sends a DAX query, and eMondrian evaluates it against the same model Excel and AI agents use.

“Out of beta” here means one specific promise:

A standard Power BI report on a live connection works and shows the right numbers. Anything eMondrian doesn’t support fails with a clear error. It never shows a wrong number.

It doesn’t mean “all of DAX”. This article covers what’s in the promise, how we tested it, and what is still outside it.

Disclosure: I’m the author of eMondrian, the open-source semantic layer this is about.

What a “standard report” covers

These are the visuals and interactions Power BI generates DAX for, and that eMondrian answers:

  • tables and matrices, with subtotals and totals;
  • column, bar, line and pie charts, with tooltips and tooltip pages;
  • the classic card and the multi-row card;
  • slicers on a column, including several selected values, and slicer lists that follow the other slicers, as in Analysis Services;
  • page and visual filters, Top N, sorting by a measure;
  • drill-down through a hierarchy (Year → Quarter → Month) and cross-highlighting;
  • conditional formatting: gradients, rules and data bars;
  • publishing to the Power BI Service through the standard on-premises gateway, with scheduled refresh of dashboard tiles.

Measures come from the eMondrian schema. A measure is defined once and gives the same number in Power BI (DAX), in an Excel PivotTable (MDX) and to an AI agent (MCP).

What changed on the way out of beta

A clear error instead of a quiet wrong answer. A DAX form eMondrian can’t express used to be dropped from the query, which can produce a plausible but wrong number. Before the label came off, every path that turns DAX into an engine query was audited for arguments that could be ignored this way. Each one was either fixed, with a regression test, or turned into an error that quotes the expression. That is the rule we hold every release to.

How we tested

  • Power BI Service, visual by visual. 77 cases on OnTime, run in the browser over a live connection through the on-premises gateway, at the end of October 2026. Every failure but three was fixed and retested; the three are the visuals Power BI itself won’t offer (below).
  • Power BI Service, end to end. A 12-step plan, last run on 7 November 2026 against Community 2026.11.07.2, with Power BI Desktop 2.159.883.0 and gateway 3000.338.2. Publishing, model refresh and scheduled tile refresh pass. Not tested: the steps that need a Pro licence, and the mobile app.
  • Cross-filter consistency. 421 queries in the exact shapes Power BI Desktop sends for cross-filtering, cross-highlighting and multi-select, for every member of every clickable field of a test dashboard. Each row must equal the card total under the same filter, and highlighted rows must add up to the card: 0 mismatches, last run on 10 November 2026 on modules 0.0.0.102.
  • DAX regression suite. 94 tests built from the queries Power BI visuals send — tables, matrices with subtotals, cards, Top N, slicers, cross-highlighting, drill-down — on FoodMart, with a parent-child HR hierarchy, and OnTime on ClickHouse. On 10 November 2026, on modules 0.0.0.102 against the public demo with the full OnTime data (182 million flights, 1987–2026): 94 of 94 pass. In matrices, every subtotal equals the sum of its rows, and a ratio’s subtotal is the ratio of the sums, not the sum of the ratios. Separately, 289 queries captured from Power BI Desktop are replayed on every build and must give the same results as on the previous one.
  • What these checks are compared with. Visual values were checked against hand-written MDX on the same model, and aggregate tables against SQL. That catches errors in turning DAX into the engine’s queries. It is not a comparison with an independent engine, such as the same report on a Power BI Import model; we haven’t done that yet.
  • The public demo runs the same build: https://demo.emondrian.com/xmla, databases OnTime and FoodMart.

What is still outside the promise

Three visuals Power BI won’t offer. The new Card with categories, hierarchy slicers and small multiples. Power BI decides this itself, from the Analysis Services level eMondrian reports (2016), and shows its own message, such as “Hierarchy slicers aren’t working because you’re connected to an old version of Analysis Services”. No query reaches eMondrian, so no wrong number is shown. Workarounds: a card without categories or a multi-row card, one slicer per level, separate charts.

No report-level measures. Power BI treats eMondrian as a multidimensional source, so New measure, Quick measure and Visual calculation aren’t offered on a live connection. This is the same as with Analysis Services Multidimensional. Measures, including year-to-date and prior-year, are defined in the schema as calculated members.

Not all of DAX for hand-written queries. In DAX Studio or through MCP you can write DAX yourself. The functions eMondrian supports are listed in DAX support; others fail with an error that names them. Time-intelligence functions (TOTALYTD, SAMEPERIODLASTYEAR, DATEADD), ALLSELECTED, RANKX and RELATED are not available.

Three places where eMondrian deliberately differs from DAX. They matter mostly for hand-written queries:

  • TOPN returns exactly n rows; DAX also returns the rows tied at the boundary.
  • NONVISUAL applies to totals too, so Power BI’s “non-visual totals” aren’t modelled.
  • A VAR is evaluated where it is used, not once where it is defined.

Smart narrative and Q&A aren’t available on a live connection to Analysis Services, and Power BI says so itself.

Microsoft doesn’t support this setup. Power BI connects through its Analysis Services connector, and Microsoft officially supports only a few third-party models. eMondrian isn’t one of them. It works, and we test new Power BI releases, but Microsoft support won’t help with it.

Try it

  • No install: in Power BI Desktop, Get data → Analysis Services, server https://demo.emondrian.com/xmla, database OnTime, Connect live.
  • On your own data: git clone https://github.com/eMondrian/emondrian-community and ./setup.sh, then describe your cubes or ask an agent to draft them over MCP. The DAX module needs a key; a free 30-day evaluation key is at emondrian.com/license.

If a visual in your report fails, copy its query from Performance analyzer → Copy query and open an issue. Failing real-world queries are how coverage grows.

eMondrian: Demo · GitHub · Docs

More: Power BI · DAX support · Release notes

Backlinks