Metric comparability
Gramin does not collapse grammar quality into one score. A count can be mechanically comparable while its meaning still depends on how the author expressed the grammar.
| Class | Meaning | UI treatment |
|---|---|---|
| A | Cross-format comparable under the metric definition | Show in the primary comparison |
| B | Context-dependent; capability or representation affects interpretation | Show with a context badge |
| C | Format-specific or heuristic | Keep in an advanced section |
Examples include rule and alternative counts as Class A candidates, recursion and nullable analysis as context-dependent measurements, and format-specific sugar or precedence details as Class C measurements.
notApplicable is data
If a frontend cannot define a metric for its representation, the feature retains a notApplicable state and reason. The UI must not render that state as 0, because zero says the property was measured and absent.
Comparing revisions
Before comparing values, compare the representation context: frontend, capabilities, source identity, and relevant options. If that context changed, explain the context change before describing a number as an increase or decrease.
The normative definitions live in the metrics catalog and cross-format notes.