July 12, 2026
Published Claim Fidelity, made a large private research corpus navigable, and carried the financial model through its first dated run.
Done
- Models prepared Claim Fidelity as a public, reproducible example of making formal claims checkable.
- A clean rebuild verified that an outside reader could reproduce the public result from the repository alone.
- I approved the release with a deliberately narrow claim: formal checkability, not proof that the underlying ideas were empirically true.
- Models completed a read-only map of the private research corpus so later work could find related material without depending on my memory of where it lived.
- The financial model completed its first dated run. Review exposed errors in how the result was described, and the models repaired both the output and the checks around it.
- The repaired run passed fresh review without expanding the project's real-world claims.
Linked
Three different projects moved materially in one day: one shipped, one became operationally navigable, and one survived its first realistic execution cycle.
Claim Fidelity was the public breakthrough. It gives a fresh reader a way to check whether a formal claim follows from the supplied material without asking them to trust the author or the model that produced it. The important caveat is also part of the result: a mechanically valid claim can still describe the wrong world.
The private corpus map solved a more ordinary problem with outsized consequences. Once model-assisted work spreads across many projects, remembering that something exists is no longer enough. The models built a map; later work still had to earn every connection it used.