Gate saved models on solver convergence and the virial milestone
This commit is contained in:
30
experiments/check_plateau_physics_README.md
Normal file
30
experiments/check_plateau_physics_README.md
Normal file
@@ -0,0 +1,30 @@
|
||||
# Plateau endpoint milestone checker
|
||||
|
||||
Run the standalone standard-library script with one completed result folder:
|
||||
|
||||
```sh
|
||||
python3 experiments/check_plateau_physics.py /absolute/path/to/plateau-result
|
||||
```
|
||||
|
||||
It reads `metadata.txt` and `final_physical.csv` and prints standard JSON. Exit codes:
|
||||
|
||||
- **0:** `solver_and_virial_milestone=pass`: recorded strict solver convergence and all eight defined, finite virial errors at or below 1e-5.
|
||||
- **3:** those milestone checks do not all pass, including an undefined pressure integral.
|
||||
- **2:** missing or malformed required endpoint data, including an incomplete quadrature table or missing convergence outcome.
|
||||
|
||||
The eight checks comprise energy and force virials with both P(h) and P(rho), each at quadrature orders 14 and 18. An undefined check is distinguished from a finite check exceeding the threshold. Pressure validity, density/enthalpy signs, negative-density budgets, mass/J errors, and local EOS/Bernoulli measurements are reported separately. Nonfinite numbers become JSON null with a retained finite flag and NaN/infinity classification.
|
||||
|
||||
Passing this gate is **not full physical validation**. No new tolerance is assigned to mass/J or local errors. Sampled signs and extrema are not certified pointwise bounds. The script trusts the diagnostic's recorded convergence flag and physical measurements; it does not reconstruct a model, validate provenance, or independently recompute integrals. It does not inspect or modify production code.
|
||||
|
||||
Verification on the September 16 campaign:
|
||||
|
||||
| Completed case | Expected and observed exit | Distinction checked |
|
||||
| --- | ---: | --- |
|
||||
| n=1, J=0.05, no central constraint, restart 160 | 0 | Strict convergence and all virials pass |
|
||||
| Same model, restart 40 | 3 | Virials pass, strict convergence fails |
|
||||
| n=1.5, J=0, no central constraint, restart 160 | 3 | P(h) virials pass; P(rho) is undefined |
|
||||
| n=3 initial seed | 3 | P(h) virials exceed the threshold |
|
||||
|
||||
Threshold equality, one virial just above threshold, and missing required data
|
||||
were also checked. Detailed verification JSON and reports remain with the raw
|
||||
campaign artifacts in the audit workspace.
|
||||
Reference in New Issue
Block a user