July 10, 2026
Started an executable financial-systems model to test whether complicated obligations could be changed without hand-waving away the consequences.
Done
- Models turned the initial research frame into working software rather than another descriptive memo.
- The first bounded example could represent a situation, apply a limited change, and reject inputs it could not interpret safely.
- Review found places where an apparently valid result could conceal an unsupported assumption; the models repaired those paths.
- A working result now required independent review before it could become accepted source; running successfully was no longer enough.
- I kept the result bounded to a controlled model. It was not evidence about real households and was not financial advice.
The purpose of the project was simple to state and difficult to implement: make the consequences of a proposed financial change explicit enough that they could be checked.
The first working slice made that question concrete. It also exposed the usual model shortcut—quietly filling an undefined gap with something reasonable. Here, undefined cases had to remain undefined or fail. That made the software less impressive in a demo and much more useful as a research instrument.