Grid Shift — Fabric App

Microsoft FabricDataflow Gen2T-SQLDirect LakeDAXFabric Apps (preview)

Views: Overview · Monthly

The app

Grid Shift overview for Bangladesh, dark theme · Tap to view full-size

Monthly view for Bangladesh — generation by fuel, month by month · Tap to view full-size

Monthly view for Nepal — an explicit empty state, because Ember has yearly data only for Nepal · Tap to view full-size

A Fabric App (preview) built on the dataapp template, reading every number from the sm_grid_shift semantic model through the semantic-model connector. No mock data. Two views, Overview and Monthly, with light and dark themes, a phone layout and 44px touch targets.

A watchlist feature writes per-user notes to the app's own Fabric SQL database, to show the app can write as well as read.

The app layer was built with AI assistance — Claude drafted the SQL and Power Query, Claude Code wrote the app code and DAX measures — while I chose the dataset and the story, decided the architecture and warehouse model, ran every step in Fabric, validated the results against Ember's raw data and tested the per-user notes with two accounts.

Pipeline

Built end to end in Microsoft Fabric on a trial capacity in October 2026, on Ember's open electricity data (CC BY 4.0).

Load. A Dataflow Gen2 (df_ember_load) brings Ember's yearly and monthly CSVs into staging tables in a Fabric Warehouse — 104,241 yearly rows across 224 areas (2000–2025) and 181,459 monthly rows across 99 areas (Jan 2019 – Aug 2026).
Model. A star schema in the warehouse's mart schema, built with T-SQL CTAS: dim_source, dim_area, dim_year, dim_date, fact_generation_yearly and fact_generation_monthly. Row counts and duplicate checks are in the schema script.
Semantic model. sm_grid_shift, Direct Lake on OneLake, with 6 relationships and 22 DAX measures, edited as TMDL through Fabric Git integration. Summary sources (Clean, Fossil, Total) are never summed with individual fuels.
App. A Fabric App (preview) with delegated sign-in, plus its own Fabric SQL database for per-user watchlist notes following the documented @role policy pattern.

Architecture — Ember CSVs through Dataflow Gen2 into a Fabric Warehouse, a star schema and Direct Lake semantic model, then the Fabric App · Tap to view full-size

What the data says

Figures come from the semantic model and were reconciled against Ember's raw CSV.

Gas is losing share. 64.3% of Bangladesh's generation in 2025, down from a 92.5% peak in 2010.
Coal took most of it. 21.5% in 2025, from roughly 0% in 2000.
Imports matter. Net imports are 14.2% of generation.
Clean power is still about 2%. 2.1% in 2025 — the story is not a solar boom.
Carbon intensity 696 gCO2e/kWh, on total demand of 118.0 TWh.

Checks and limits

Reconciled against source. Bangladesh's 2025 values were checked between a DAX query and Ember's raw CSV, and a monthly comparison (Bangladesh coal, Aug 2026 vs Aug 2025) was checked the same way.
Per-user notes tested with two accounts in separate private browser sessions, with a database check that each note carries its author's owner ID.
Preview and sign-in. Fabric Apps are in preview, and the live app needs a Fabric sign-in with access to the workspace — so there is no public link. It ran on a trial capacity ending in early November 2026; the screenshots and the recorded demo in the repo are the lasting record.
Data limits. Ember's latest months are provisional and get revised. Pakistan's monthly series ends earlier than the others, and Nepal has yearly data only — the Monthly view says so rather than rendering an empty chart.
Three details were not in the docs I read: Direct Lake as the model's storage mode, the database dialect: mssql setting, and the devUri local dev flow.

View source on GitHub →