The income statement as its own visualBeta

A P&L is not a chart. It is a table with a memory: rows that belong together, subtotals that follow from their children, formula rows such as EBITDA or gross margin, cost rows that compute negative and are shown positive — and a “below plan” that is bad for revenue and good for cost. Rebuilding that with a matrix and a handful of conditional formats costs an afternoon every time, and the result is still rigid.

That is the gap the new visual fills. You bind the account dimension and the measures — and you get the statement including structure, sign logic and variances.

Why IBCS-inspired

Because management reports have a language that should not be reinvented every time. The visual adopts the notation rather than interpreting it: actual solid dark grey, previous year grey, plan white with an outline, forecast hatched. Variances as bars next to the number, with one identical scale per Δ column across all rows — otherwise the eye compares bar lengths that have nothing to do with each other. Labels outside, a positive sign spelled out, no gridlines, white ground.

One deviation is deliberate and documented: instead of the classic IBCS green, favourable variances use a teal tone — readable with red–green colour-vision deficiency too. If you want the pure doctrine, switch the format pane to “IBCS classic green”; the note in legend and footer disappears with it.

What the beta already does

The heart of it is the calculation engine: unbalanced account hierarchies of any depth — either as parent-child or, more commonly, as level columns L1..Ln from a star schema. Subtotals are created automatically from the paths; formula rows reference other rows by name ([Gross profit]+[Operating expenses]), formula on formula included, with circular-reference detection. Orphan accounts land visibly in “Not assigned” instead of quietly disappearing — a data-quality signal, not a bug.

• The value driver tree grows out of the formula graph (DuPont) — and, since recently, out of a plain dimension hierarchy with no formula row at all: with several roots, a virtual total root carries them as children. Tidy layout, operator circles on the edges, and ⌖ re-roots the tree on any card.
• The card zoom pulls one node onto the whole surface: combo chart with months and bridge, scenario grid, “contributes to” and “driven by” — and it scales itself into the viewport.
• Period-aware YTD: if actuals only run to August, the reference is summed to August as well. Before that, the figure compared eight months of actuals against twelve months of plan — arithmetically correct, factually apples to oranges. The label now says so: “YTD _Aug”.
• Periods as they really arrive: “2026-01”, date columns, German and English month names, with and without a year. One parser for all of it, one ordering for every consumer — plus a “data order” switch for fiscal years. Whatever cannot be read is reported rather than guessed.
• Selection, cross-filtering, drillthrough: a click on a value column sets the Power BI selection of that row — for a subtotal the union of all leaves beneath it, for a formula row the leaves of its operands. A right click opens the native context menu with the page's drillthrough targets; in the card zoom, “↗ Drill” does both in one gesture.
• Readable on any screen: a font size preset from HD through Full HD to UHD, plus fine scaling — and it applies to everything, from the title line to the footnote.

Built with AI — and documented in the open

Like ChartKitchen before it, this visual was built AI-assisted: steered by one person with a controlling and BI background, written by an AI agent, verified through tests and a public git history. That is not stated as marketing but as context — anyone who wants to know how load-bearing such a thing is should be able to read it up. The calculation engine is deliberately free of Power BI dependencies and tested in isolation, because a reconciliation error in a P&L would be the fastest way to lose trust: 28 engine test blocks, headless render tests of every case and more than 260 interaction checks run on every version.

The detailed reckoning — what this state would have cost the classic way, and why it is not the AI alone that makes the difference — is in the position paper “Ten days to a market-ready state”.