pbi-replicatelisted
Install: claude install-skill starmynd-org/powerbi-to-ai-visuals
# Rebuild the report, faithfully
The point of this step is trust. Before anyone accepts a redesign, they need to
see the same report, with the same numbers, outside Power BI. So the page is
*compiled from the report definition* - never eyeballed from a screenshot.
```bash
python3 tools/replicate_html.py --model build/model --out build/replica
```
Requires `build/model/model.json` and `build/model/report-spec.json` from
`pbi-extract`.
## What the compiler carries across, and why each one bites
**Visual-level filters.** In the sample report the matrix filters out
`[Not Applicable]` business types and the two charts beside it do not. Miss it
and every subtotal in that visual is wrong while the rest of the page looks
right. Filters live per visual in `report-spec.json`.
**The saved slicer selection.** A report ships filtered. This one opens on
FY2019. Rebuild it unfiltered and every number differs from the screenshot the
user is holding.
**Sort-by columns.** `Product[Category]` sorts by a `Sorting` column pulled
from another table with `RELATED`. That is the only reason the order reads
Bikes, Components, Clothing, Accessories. Alphabetical order is the tell that a
rebuild was done by eye.
**Theme colour indexes.** A visual stores `ThemeDataColor: {ColorId, Percent}`,
not a hex value. `ColorId` indexes a palette whose slot 0 is the theme
background and slot 1 the foreground, so the data colours start at slot 2:
`dataColors[ColorId - 2]`. Verified on this report - the heade