review classifier: tests-with-behaviour asks for tests of the logic, not of every line of glue

.hyperhive/review-classifier.yml

@@ -33,14 +33,21 @@ classifications:
     threshold: 75
   - name: tests-with-behaviour
     description: >-
-      Every change to Rust or application code behaviour comes with a new or
-      updated test that exercises it. A diff whose changed files are all test
-      code (a fixture under `nix/module-eval/`, `nix/checks.nix`, a crate's
-      `tests/` directory, or Rust inside `#[cfg(test)]` or a `#[test]`
-      function) changes no behaviour: the question is not applicable. A NixOS
-      module assertion, a Grafana/alerting rule, or the literal text of a log
-      or tracing message is not held to this either. A PR that changes
-      behaviour and tests together matches.
+      The decision-making logic a change adds or alters in Rust or
+      application code has a new or updated test that exercises it: pure
+      functions, state transitions, request and response handling, parsing,
+      validation. A new code path with real logic and no test does not
+      match, and neither does a bug fix with no regression test. Thin glue
+      needs no test of its own: UI component wiring, JSX or markup, a
+      routing table, registering an event listener, re-exporting, or
+      calling a function that is already tested. A diff whose changed files
+      are all test code (a fixture under `nix/module-eval/`,
+      `nix/checks.nix`, a crate's `tests/` directory, or Rust inside
+      `#[cfg(test)]` or a `#[test]` function) changes no behaviour: the
+      question is not applicable. So is a diff that changes only
+      documentation. A NixOS module assertion, a Grafana/alerting rule, or
+      the literal text of a log or tracing message is not held to this
+      either. A PR that changes logic and tests together matches.
     polarity: positive
     threshold: 65
   - name: docs-updated