What the Book Is Not Claiming

The manuscript offers a framework and conditional proof spine, not a proof that superintelligence alignment is solved.

What decision changes?

Use the site as a map of cruxes and artifacts, not as a certificate that current systems are safe.

This project is not claiming that alignment is solved.

It is not claiming that Lean proves real systems are safe. Lean checks a conditional skeleton: if the boundary, grounding, value, correction, successor, and adversarial conditions hold, then the safety-case argument has the advertised shape.

It is not claiming that toy experiments validate frontier deployment. The experiments are sanity checks and falsification harnesses. Negative results matter because they show where the measurement story is still weak.

It is not claiming that a website can replace the book. The PDF remains the long-form text; this site only summarizes it and points back to it.

What would count as evidence?

The right evidence would discharge bridge assumptions in bounded deployment classes and show where the framework fails.