Tableau troubleshooting 03: Commissioning, delivery and acceptance—so you can change what you receive
When commissioning a Tableau tool, the client explains how it will be used after completion, the developer delivers what is needed for that use, and the recipient checks that the agreed requirements have been met. We believe these activities—commissioning, delivery and acceptance—need to work together.
Developers cannot meet demands that grow beyond what the budget and schedule allow. Yet if only the finished views are handed over, recipients may not notice that materials needed for later corrections or updates are missing. Both sides need to anticipate the work that will follow completion and agree on what to deliver and what to check.
What changes arise as a tool stays in use?
Rokugeisha’s Takuya Iwasaki worked on the development of e-CSTI as a Cabinet Office Fellow in 2019–2020, including work on the client side of commissioning visualization tools. From April 2020, he served as visualization lead at the National Graduate Institute for Policy Studies, with responsibility for the overall operation and management of e-CSTI under a commission from the Cabinet Office. His team both delivered visualization tools to the Cabinet Office and received tools made by other partner companies for integration into the site.
Operating the site involved more than receiving tools and publishing them. Sometimes several rounds of revisions with developers were needed before publication. Further changes arose as the tools continued to be used.
These changes can broadly be grouped into three types:
- Display adjustments before publication: Adjusting text size, unintended line breaks and other display details arising from differences between the development environment and the environment in which the tool is actually used.
- Bug fixes and functional improvements after publication: Correcting operational problems or data errors in the existing tool, or improving its functionality.
- Updates with a new year’s data: Incorporating data from the next survey round or business year into the visualization. Changes to fields or classifications may also require changes to data preparation.
The materials needed depend on the change. A small display adjustment may require only editing the workbook. Correcting data or updating to a new year can require going back to the source data and the steps used to prepare it.
There were occasions when an attempted update revealed that necessary source data were not available to the team, requiring contact with the original developer. The views would open, but the work could go no further. We introduced this problem in Part 2, “It opens, but you cannot change it.”
These experiences taught us to build with later data updates in mind, as well as getting the current display right. Preserve the materials and steps that lead from source data to the views, and make them available to whoever works on the tool next. This is the starting point for Rokugeisha’s delivery package.
At Rokugeisha, we organize the package into five folders
When building or revising Tableau tools, we use the following structure as the basis for the materials we hand over. The five folders follow the path from receiving data through preparation to visualization.
| Folder | What it contains and why |
|---|---|
| 01 Original supplied data | The source data provided to us, preserved without changes. These are the starting point for investigating questions about the figures. |
| 02 Data prepared for processing | Data organized for use in production, along with mapping tables for names, classifications and other items. Preparation for adding new data also starts here. |
| 03 Processing flows and code | Tableau Prep flows or Python code used to process the data, together with their outputs. These allow the processing steps to be examined and rerun for an update. |
| 04 Tableau | The visualization workbook and the data it reads. These are used to check the display and edit the views. |
| 05 Change and validation records | Records of what changed, what was checked and the results. They help recipients understand the revisions and the scope of validation. |
A README explains how to use the package: what to open first, how input data pass through the processing steps to reach the views, and what to do in what order when updating. It explains how the materials work together, rather than simply listing files.
At acceptance, we share our validation results and check whether recipients can use the package to carry out their planned work. If they plan to update it with a new year’s data, we check together whether they can do so in their own environment.
Keeping the data, processing steps and explanations together also helps with AI-assisted checks. At Rokugeisha, we have confirmed that this package can be used with tools such as Codex and Claude Code to check whether updates can be carried out correctly by following the documented steps. This applies only where the data may be shared with services that use cloud-based large language models.
Along with the finished Tableau tool, we want to provide the materials and procedures needed to make further changes. Recipients should be able to work on the corrections and updates that arise as the tool continues to be used. These five folders form our foundation for that handover.