Benchmark: 24 real process files
On 3 November 2026 we converted 24 real process files, from client projects, and counted what came through. Every page was converted, with no structural defect. Here are the figures, how we counted, and their limits.
Results on 24 real files
Median of the 24 files, with the lowest and highest file in brackets.
| Measure | Result |
|---|---|
| Files converted | 24 out of 24 |
| Pages represented in the BPMN | 58 out of 58 |
| Labelled shapes found | 100 % (96 – 100) |
| Connected flows found, same direction | 100 % (91.7 – 100); 3,065 out of 3,084 in all |
| Flow labels kept (“Yes”, “No”…) | 723 out of 729 |
| Annotation links kept | 89 out of 96 |
| Element types that make sense | 100 % (97.2 – 100) |
| Structural defects in the BPMN | 0 |
| Conversion time | 0.6 s median per file |
The files
- 24 real process files from client projects, drawn by hand, many of them on several pages.
- Anonymised before the test: every text in the shapes, pools and notes is replaced, the structure is untouched. We do not name the client or its sector, and the files never leave our working space.
- Compared with the original: for each file, a script lists the shapes, arrows and pages of the .vsdx and looks for each one in the BPMN produced.
How we count
- Shapes: a labelled shape is found when the BPMN holds an element with the same text. Shapes without text are not in this percentage, so they cannot inflate it.
- Flows: an arrow attached at both ends is found when the BPMN holds a flow between the same two elements, in the same direction. Arrows attached at one end only, or not attached at all, are left out of the measure.
- Pages: a page counts when at least one of its shapes is found in the BPMN. Background and empty pages do not count.
- Structural defects: references to missing elements, elements without a drawing, flows without a route, duplicate identifiers, sequence flows between pools. Any one of them can stop a BPMN tool from opening the file properly.
Where we are weaker
We also converted 20 files generated from the standard BPMN stencil. Shapes came through at 86.7 % and flows at 85.9 % (medians), every page was represented and no file had a structural defect. But only 134 of 406 flow labels and 86 of 279 annotation links were kept. We are working on it.
20 more files, drawn with a custom stencil that stores each shape’s type in a non-standard field, were refused as invalid. A fix is in progress.
What this does not measure
- Time to fix by hand after the conversion, and the import into each target tool: not measured yet.
- Your files: drawing habits differ. Test one of your own files, with no account: you see every converted page and the totals of the conversion log.
Measured on 3 November 2026 with our own benchmark tool, on the converter in production that day. We will run it again on the same files as the converter changes, and update this page.
Last updated: