Validation
Validation
This page summarizes what the current release checks, what it does not try to cover, and a few representative results from the validation notes.
Covered
- Visible precision, lift/projector identities, and local visible calculus.
- Fixed-ceiling hidden-load geometry and inverse-theorem round-trips.
- Hidden transport, determinant clock, and contraction-factor composition.
- Thin Donsker-Varadhan and quotient-side Gaussian contraction checks.
- Input validation, boundary behavior, and deterministic batch behavior.
Not covered here
- Full finite-observer optimization.
- Full quotient-observation beyond the thin adapter layer.
- Downstream empirical pipelines and application wrappers.
- Notebook, plotting, and benchmark surfaces.
Representative checks
What the tests focus on
Structure
Exact identities and composition rules
- Lift/projector identities and variational energy splitting.
- Determinant-curvature checks in the local calculus.
- Hidden-load support, rank, clock, and monotonicity checks.
- Inverse-theorem round-trips and order-reversal checks.
Robustness
Boundary and misuse checks
- Malformed visible, hidden, and batch inputs rejected with explicit exception types.
- Near-boundary tolerance checks.
- Input immutability checks.
- Seeded validation-sweep and deterministic batch behavior checks.
Selected results
Representative values from the current notes
| Metric | Observed value |
|---|---|
| Max lift/projector residual | 2.674941282677552e-14 |
| Max tower-law residual | 3.284872374109682e-14 |
| Max hidden clock residual | 3.552713678800501e-15 |
| Max inverse-theorem Lambda to X to Lambda residual | 2.587909939792647e-15 |
| Max inverse-theorem X to Lambda to X residual | 2.2201725879984535e-15 |
| Max two-step transport residual | 1.8991764023475337e-14 |
| Max downstairs associativity residual | 4.292205267163458e-15 |
| Max divergence contraction violation | 0.0 |
Run the checks
python -m pytest -q
python tools/validation_sweep.py
python tools/stack_soak.pyHow to read this
These numbers describe current behavior, not marketing claims.
The point of the validation notes is to show what was checked and what the observed residuals were, while keeping the package limits visible.