Poor road design is mainly a political problem. We know how to make good roads, on the technical side it is a "solved problem", but we lack the political will to do it. So setting the proper framing did solve the root problem
Then target leadership. They're the ones responsible for building better roads (which is the whole advertised purpose of this project!).
Writing an NLP bot that rewords articles to blame the driver 100% of the time is nothing but an exercise in getting a nice revenge dopamine rush. But drivers aren't the ones building roads.
Leadership (politicians) respond to public sentiment and attitudes held by their constituents. They're not going to build safer streets, despite the benefits and means to do so, without the political will of citizens. Shifting the way the public perceives fatal car crashes would compel more people to bring up the issue with their local leaders and gather the momentum needed for such changes.
> But the solution that comes out of that framing is to somehow shame drivers into being better, as opposed to making roads better.
This tool actually includes a recommendation called "framing" which asks writers to incorporate sentences specifically to highlight the systemic nature of the problem as opposed to blaming individual drivers. Click on one of the examples to see, explained at https://visionzeroreporting.com/issues#framing:
Article lacks thematic framing. Readers who encounter episodic frames tend to hold individuals responsible for negative outcomes and put less pressure on public leaders to make changes. [Learn more about framing]
"We need to limit speed and the amount of interactions between cars and pedestrians" doesn't sound to me like relying on shaming (which never works). Rather, it evokes like building e.g. walkways and bike paths that are physically separated from roads with cars. That is much more effective.