XYPHERS THEORY OF AIM REPRODUCTION

The verifier needs only a Rust toolchain. It has no external dependencies.
In a release build the test takes a few seconds and the run about fifteen
minutes.

From source/xypher-theory-of-aim-proof/:

  cargo test --release --locked
  cargo run --release --locked --quiet

The report is 104 lines: the gates G01-G12, the controls C01-C09 with where
each failed, one line per world and agent with the exact edge E, the frozen
prediction E_th and their ratio, then the five hypothesis verdicts, and it
ends:

  OVERALL PASS

To compare the whole report with the recorded run:

  cargo run --release --locked --quiet | diff ../../expected-report.txt -

No output from diff means your run reproduced the recorded report exactly.
Its SHA-256 is

  76685af36d4a0dbde32ef41d1734ca7249a429bbe2556e6fe8e9de00bb5816dc

The independent check needs only Python 3 (with python-flint it takes about
fifteen seconds; with the standard library alone, about ninety minutes). From
this directory:

  python3 independent_check.py expected-report.txt theory/predictions_cef5.txt

It prints one line per world and agent, then

  H and J agree with the frozen report in 76 of 76 steady states

and the five verdicts, as in expected-independent-check.txt.

The theory itself needs Python 3 with numpy. From theory/:

  python3 predictions_cef5.py

reproduces predictions_cef5.txt.

Reproduction boundary
  This is an exact replay of a frozen finite construction. A passing replay
  reproduces the recorded result; the scope is the one stated in README.txt.