## kernels.harness.ConvergenceGates


Pass/fail thresholds on an entry's `convergence()` diagnostics.


Usage

``` python
kernels.harness.ConvergenceGates(
    max_rhat=1.05, max_divergence_frac=0.002, min_ess_bulk=100.0
)
```


A fit failing any gate is re-run at the next stage's settings. Missing diagnostics (None/NaN - e.g. a likelihood-based entry with no sampler) pass by construction: there is nothing an escalated sampler would fix. Defaults: the model cards' R-hat 1.05 reporting convention, ~0 tolerance for divergences (0.002 of draws), and a min bulk ESS low enough to flag only genuinely stuck chains.


## Parameter Attributes


`max_rhat: float = ``1.05`  

`max_divergence_frac: float = ``0.002`  

`min_ess_bulk: float = ``100.0`