A report page has limited space and an expensive first question: where should I look? Filling that space with charts is easy. Establishing a hierarchy that helps someone understand a result, compare it and investigate an exception takes a more deliberate design.
The brilliant-v1 report policy starts with that hierarchy. It gives the page one decision, a primary outcome and supporting information with distinct jobs. It also keeps design policy separate from proof. A calm page can still contain the wrong number, and a geometry pass cannot establish business correctness.
One outcome, then the explanation
The policy's default is four contextual KPI slots: an outcome, two drivers and a guardrail. Show fewer when the model supports fewer. Do not duplicate current, prior-year and growth variants of one measure simply to fill the row. Six logical cards is the policy ceiling, not a target.
Each KPI needs context. Name the period, unit, baseline and variance when available. Distinguish the data-through date from the refresh timestamp. A trend needs qualified chronological observations. Missing baselines or trends remain unavailable rather than being fabricated from a metric name.
Polarity also needs governance. Higher cost is not automatically adverse, and a positive variance is not automatically favourable. Preserve the arithmetic sign and apply a business-approved interpretation. Unknown polarity stays neutral. A descriptive breakdown shows contribution or association; it does not establish causation.
A frame with room for the comparison
The preview brilliant-v1 frame uses a 1280 × 720 canvas. Its header is 72 pixels high, outer content margins are 24 pixels, gutters are 16 pixels and the KPI band is 136 pixels. These are product-policy choices, not universal usability thresholds.
| Region | Design purpose |
|---|---|
| Header | A native title, navigation and compact period/scope controls. |
| KPI band | The primary outcome and contextual comparisons. |
| Dominant visual | The main analytical question, with enough space to read it. |
| Supporting panels | A ranked comparison and actionable detail. |
The engine owns placement. Whole canvas pixels and cumulative rounding keep siblings aligned to the frame edge. The final placement manifest governs the emitted page and artwork. Hand-editing PBIR positions bypasses that relationship and makes later checks harder to explain.
Keep versions distinct. The existing default frame has a 60-pixel header and 10-pixel gutters. Brilliant-v1 does not silently replace that frame or a harvested client reference. Inspect the installed contract and supported page-spec route before selecting a policy.
Give each page a different job
Overview orients the reader. Trends & Drivers supports comparison over time. Breakdown exposes composition. Detail & Exceptions supplies an action queue. Drillthrough starts from selected context and needs a clear way back. Those roles keep the normal page sequence coherent without forcing every question onto the landing page.
Sample schematic · Fictional values · Preview design direction
The website schematic is an illustration, not a qualified tenant export. Its purpose is to make the hierarchy readable. Real field bindings, interactions and numerical observations still need their own evidence.
Use colour and artwork with a specific purpose
The policy uses light neutrals, one blue accent and labelled teal/orange sentiment. Scenario colour and good/bad interpretation have different meanings. Do not let one colour carry both. Text, icons, patterns and explicit labels preserve meaning when colour is unavailable or ambiguous.
The palette alone cannot establish accessibility. Some comparison tones need a darker essential boundary, markers or an accessible alternative. Check the actual adjacent colours and rendered host. Font settings use Power BI points, while frame coordinates use canvas pixels. Shrinking type to make a long label fit defeats the hierarchy.
Static backgrounds can carry panels, subtle depth and decorative rules. Values, titles, filter context and controls remain native or live. Artwork must follow the final placements, and it needs separate export and mobile review. The curriculum explores generated and designer workflows; the product brief does not establish a shipped automatic background generator.
Qualify the numbers separately
A release needs distinct evidence for geometry, semantic queries, accessibility, interactions and rendered appearance. Independent expectations judge the checked values. Bounded Windows OCR is a preview route for authored numeric-label regions, with explicit limitations. It does not establish every visual's correctness.
The design default is useful because it gives each piece of information a job. The proof trail is useful because it says which claims the evidence can support. Explore the sample report pages, then study the proposed F4 report-design course to practise the schematic without confusing a good layout with a qualified result.