The Commonwealth Budget Explorer lets readers explore illustrative budget assumptions against a stated reference. The product case is about making that exploration understandable while keeping the model’s limits visible.
The improved explorer and matching guide were accepted on 24 October 2026 and published on the research site on 25 October. You can open the explorer and read its illustrated guide.
Preserve the reason for the product
Readers needed to change assumptions and inspect the consequences. Source and accounting uncertainty did not remove that interaction requirement. Instead, the work separated the official reference, direct changes and clearly labelled illustrations.
A familiar interface from another research product offered a useful starting point. It did not establish valid fiscal equations or make domain-specific evidence interchangeable.
Technical completion exposed an experience gap
The first private explorer had fifteen controls and five tabs, with functioning calculations. Product Owner comparison with a reference explorer showed that readers still lacked enough explanation of income, spending and the starting budget position.
A list of working features had become a weak substitute for the experience of understanding a first result. The response was to compare the purpose of each section, rather than simply copy its label.
Revise the first useful experience
The accepted revision added clearer starting outcomes, a plain-language budget picture, connected controls, richer history and published survey context. Evidence and editable assumptions were brought closer together.
The experience also preserves the reader’s scenario when they explore detail and return. Undo, reset and saved scenarios support recovery. These behaviours matter because exploring uncertainty should not require reconstructing an experiment after every detour.
Challenge the domain assumptions
Independent AI challenge and deterministic checks examined source identity, accounting boundaries and potential double counting. The need for annual funding assumptions exposed a gap, leading to two explicitly illustrative funding cases rather than invented official values.
The model presents direct annual budget changes. It does not calculate additional debt, interest, productivity or service-quality effects. Survey context retains its own metadata and interpretation limits.
AI review is not credentialed fiscal certification. Agreement among tools does not remove the need for clear sources and bounded claims.
What the evidence supports
| Evidence | Supported conclusion |
|---|---|
| Owner review of the revised experience | The revised product was accepted within its stated scope. |
| Functional and accounting checks | The documented controls and model rules behaved as tested. |
| Guide and navigation checks | The illustrated guide matched the chosen product edition. |
| Recorded public release | Readers can access that product on the research site. |
None of these establishes representative reader comprehension, policy effectiveness or a calibrated economic forecast.
What this teaches about reuse
Share interaction and evidence contracts where useful, but validate each domain separately. A new product should identify what transfers, what changes and what remains unproved.
The AI-Assisted Research family case explains how product learning returns to a shared approach. This product’s acceptance does not validate every pattern used elsewhere in that family.
A practical next test
Ask an intended reader to explain the starting position, make one change, interpret the result, find its evidence and recover without coaching. Record observations separately from arithmetic checks and owner acceptance.
Use the Product Guide to frame that next bounded test. The article begins at public edition 1.0; the linked explorer retains its own product version and release record.
Help improve this page
Found a problem or a better way to explain it? Suggest a correction or improvement.