| Class | Description |
|---|---|
| CorpusRunner |
Runs pp and/or pp2 over a benchmark manifest, scores every output with
the fidelity gate on, optionally adds seeded syntax-degraded samples (pp2
only), and produces the run record the CLI writes as JSON.
|
| LayoutDegrader |
Produces layout-degraded variants of a SQL text that keep every token
(so the fidelity gate still passes) but destroy the layout in a known way.
|
| PrettyScoreMain |
PrettyScore command line.
|
| RunTools |
Operations over saved run records: finding-level diff between two runs,
baseline regression check, corpus backlog (Markdown) and blind A/B
acceptance samples.
|
| RunTools.Diff | |
| SyntaxDegrader |
Derives syntax-broken samples from a clean SQL text with a fixed seed so a
corpus of "real-looking" parse failures can be regenerated in CI without
checking the derived files in.
|
| SyntaxDegrader.Sample |
A derived sample.
|
| Enum | Description |
|---|---|
| SyntaxDegrader.Kind |