diff --git a/crates/bindings/README.md b/crates/bindings/README.md
index cfb17c1..23aa19f 100644
--- a/crates/bindings/README.md
+++ b/crates/bindings/README.md
@@ -81,6 +81,9 @@ except ValueError as why:
 | `find_citations` | [`citefind`](../citefind/README.md) |
 | `looks_like_footnote_citations` | whether a body chunk carries notes worth reading |
 | `suggest_author` | [`bibclean`](../bibclean/README.md)'s author recovery |
+| `canonical_subject` | what a closed subject scheme calls a heading |
+| `compose_subjects` | the subjects a book should carry under one |
+| `needs_subject` | whether it carries a heading the scheme knows |
 | `citation_prompt` | the question `citefind --prompt` writes |
 | `merge_citation_reply` | what `citefind --merge` does with the answer |
 
diff --git a/crates/bindings/python/catena/__init__.py b/crates/bindings/python/catena/__init__.py
index a8bbccd..6feead9 100644
--- a/crates/bindings/python/catena/__init__.py
+++ b/crates/bindings/python/catena/__init__.py
@@ -13,14 +13,17 @@ lines of JSON.
 from ._catena import (
     __version__,
     cue_windows,
+    canonical_subject,
     chunk_markdown,
     citation_prompt,
+    compose_subjects,
     count_tokens,
     find_citations,
     fuse_hits,
     identify,
     looks_like_footnote_citations,
     merge_citation_reply,
+    needs_subject,
     pack_cues,
     parse_identifier,
     parse_subtitles,
@@ -34,14 +37,17 @@ from ._catena import (
 __all__ = [
     "__version__",
     "cue_windows",
+    "canonical_subject",
     "chunk_markdown",
     "citation_prompt",
+    "compose_subjects",
     "count_tokens",
     "find_citations",
     "fuse_hits",
     "identify",
     "looks_like_footnote_citations",
     "merge_citation_reply",
+    "needs_subject",
     "pack_cues",
     "parse_identifier",
     "parse_subtitles",
diff --git a/crates/bindings/python/catena/__init__.pyi b/crates/bindings/python/catena/__init__.pyi
index b6c15aa..9d9770a 100644
--- a/crates/bindings/python/catena/__init__.pyi
+++ b/crates/bindings/python/catena/__init__.pyi
@@ -39,6 +39,17 @@ def fuse_hits(
 def identify(name: str, data: bytes) -> dict[str, Any]: ...
 def find_citations(text: str) -> list[dict[str, Any]]: ...
 def looks_like_footnote_citations(text: str) -> bool: ...
+def canonical_subject(
+    scheme: Mapping[str, Any], name: str
+) -> str | None: ...
+def compose_subjects(
+    scheme: Mapping[str, Any],
+    subjects: Iterable[str],
+    title: str | None = ...,
+) -> list[str]: ...
+def needs_subject(
+    scheme: Mapping[str, Any], existing: Iterable[str]
+) -> bool: ...
 def suggest_author(
     title: str, front_matter: str | None = ...
 ) -> dict[str, Any] | None: ...
diff --git a/crates/bindings/src/lib.rs b/crates/bindings/src/lib.rs
index ee58c30..5d4fcde 100644
--- a/crates/bindings/src/lib.rs
+++ b/crates/bindings/src/lib.rs
@@ -215,6 +215,30 @@ pub fn parse_identifier(text: &str) -> Option<idnorm::Identifier> {
     idnorm::parse(text)
 }
 
+/// What a closed subject scheme says about a set of headings.
+///
+/// The scheme is supplied rather than built in — which subjects a collection
+/// uses is the most collection-specific thing there is. These take it as the
+/// record `bibclean` reads from a file.
+pub fn canonical_subject(scheme: &bibclean::scheme::Scheme, name: &str) -> Option<String> {
+    scheme.canonical(name).map(str::to_string)
+}
+
+/// The subjects a book should carry: the chosen ones, their parents, and
+/// nothing the title rules out.
+pub fn compose_subjects(
+    scheme: &bibclean::scheme::Scheme,
+    subjects: &[String],
+    title: Option<&str>,
+) -> Vec<String> {
+    scheme.compose(subjects, title)
+}
+
+/// Whether a book has no heading the scheme recognises.
+pub fn needs_subject(scheme: &bibclean::scheme::Scheme, existing: &[String]) -> bool {
+    scheme.needs_subject(existing)
+}
+
 /// The best author a record's own title and front matter can offer.
 ///
 /// `None` when neither says anything that reads as a person's name, which is
diff --git a/crates/bindings/src/python.rs b/crates/bindings/src/python.rs
index dc97a1b..0ea49e7 100644
--- a/crates/bindings/src/python.rs
+++ b/crates/bindings/src/python.rs
@@ -220,6 +220,29 @@ fn identify<'py>(py: Python<'py>, name: &str, data: &[u8]) -> PyResult<Bound<'py
     to_py(py, &crate::identify(name, data))
 }
 
+/// The subject a heading means under a scheme, or `None`.
+#[pyfunction]
+fn canonical_subject(scheme: &Bound<'_, PyAny>, name: &str) -> PyResult<Option<String>> {
+    Ok(crate::canonical_subject(&from_py(scheme)?, name))
+}
+
+/// The subjects a book should carry under a scheme.
+#[pyfunction]
+#[pyo3(signature = (scheme, subjects, title = None))]
+fn compose_subjects(
+    scheme: &Bound<'_, PyAny>,
+    subjects: Vec<String>,
+    title: Option<&str>,
+) -> PyResult<Vec<String>> {
+    Ok(crate::compose_subjects(&from_py(scheme)?, &subjects, title))
+}
+
+/// Whether a book carries no heading the scheme recognises.
+#[pyfunction]
+fn needs_subject(scheme: &Bound<'_, PyAny>, existing: Vec<String>) -> PyResult<bool> {
+    Ok(crate::needs_subject(&from_py(scheme)?, &existing))
+}
+
 /// The best author a record's own title and front matter can offer.
 ///
 /// `None` when neither says anything that reads as a person's name.
@@ -314,6 +337,9 @@ fn _catena(module: &Bound<'_, PyModule>) -> PyResult<()> {
     module.add_function(wrap_pyfunction!(find_citations, module)?)?;
     module.add_function(wrap_pyfunction!(looks_like_footnote_citations, module)?)?;
     module.add_function(wrap_pyfunction!(suggest_author, module)?)?;
+    module.add_function(wrap_pyfunction!(needs_subject, module)?)?;
+    module.add_function(wrap_pyfunction!(compose_subjects, module)?)?;
+    module.add_function(wrap_pyfunction!(canonical_subject, module)?)?;
     module.add_function(wrap_pyfunction!(citation_prompt, module)?)?;
     module.add_function(wrap_pyfunction!(merge_citation_reply, module)?)?;
     module.add_function(wrap_pyfunction!(series_of, module)?)?;