Accelerate
The research that made delivery performance arguable with evidence, and unusually honest about its own method.
Four measures, a survey instrument, and a claim that delivery performance predicts organisational performance. What sets it apart is the methodology section, which states the limits of the sampling and the inference rather than hiding them.
The idea I took
Speed and stability are not a trade off, and the belief that they are is the load bearing assumption behind most heavyweight change control. Being able to point at published research, rather than at my own preference, changed several conversations with risk and audit functions that I would otherwise have lost.
Where it stops
The measures get adopted as targets almost immediately, which is exactly what the book warns against. Deployment frequency is easy to game, and frequently is.