fix: emit number_of_cases as a JSON integer instead of source text

CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to this project are documented in this file.
 
+## Unreleased
+
+### Fixed
+- **`number_of_cases` is now emitted as a JSON integer instead of source text.** The slot is typed `int` on biolink-model's `EntityToDiseaseAssociation` / `EntityToPhenotypicFeatureAssociation`, but it was absent from `lib.numeric_columns`' exact set — since 15.1's `STUDY_SIZE_EXEMPT_PATTERN` (#119) stopped it being *renamed* onto `study_size`, nothing cast its *values*, so the raw TSV cell (`"1"`) shipped on the edge NDJSON. Pydantic's lax validation coerced the string back to int inside `validate_kgx`, so the record still validated, and the Rust `uuid_on_collision: merge` recompute already wrote a real int — leaving the shipped graph type-inconsistent edge to edge. `number_of_cases` now rides the same `clean_numeric` / `format_numeric` machinery as `study_size`: `numeric_slot_kind` reads the `int` typing off the installed model, fractional / negative / non-numeric counts become null, and the release-mode `drop_low_number_of_cases` filter is untouched (it already cast inline).
+
 ## 16.6.1 - 2026-09-04
 
 ### Performance

src/tablassert/lib.py

@@ -698,27 +698,27 @@ def math_op(lf: pl.LazyFrame, col: str, func: Functions, args: list[Literal[Toke
 def numeric_columns(names: list[str]) -> list[str]:
     """Return column names that should be coerced and formatted as numbers.
 
-    P-value columns by substring plus the exact ``effect_size`` and
-    ``study_size`` fields. The old ``sample_size`` / ``relationship_strength`` /
-    ``supporting_study_size`` names are absent on purpose: the column coercions
-    rename them to ``study_size`` / ``effect_size`` before ``clean_numeric`` /
-    ``format_numeric`` run.
+    P-value columns by substring plus the exact ``effect_size``,
+    ``number_of_cases`` and ``study_size`` fields. The old ``sample_size`` /
+    ``relationship_strength`` / ``supporting_study_size`` names are absent on
+    purpose: the column coercions rename them to ``study_size`` / ``effect_size``
+    before ``clean_numeric`` / ``format_numeric`` run.
 
     Args:
         names: Schema column names to filter.
 
     Returns:
         Subset of ``names`` destined for numeric coercion/formatting.
     """
-    # P-value columns by substring plus exact effect-size and study-size fields.
-    exact: set[str] = {"effect_size", "study_size"}
+    # P-value columns by substring plus exact effect-size, case-count and study-size fields.
+    exact: set[str] = {"effect_size", "number_of_cases", "study_size"}
     return [c for c in names if ("p_value" in c.lower()) or (c in exact)]
 
 
 def clean_numeric(lf: pl.LazyFrame) -> pl.LazyFrame:
     """Coerce numeric annotation columns to Float64, dropping non-numeric values to null.
 
-    Only touches p-value, effect-size and study-size columns.
+    Only touches p-value, effect-size, number-of-cases and study-size columns.
 
     Args:
         lf: Source LazyFrame.
@@ -746,7 +746,8 @@ def format_numeric(lf: pl.LazyFrame) -> pl.LazyFrame:
 
     Remaining numeric columns are emitted as real JSON numbers when the installed
     biolink model types them ``int`` / ``float`` -- ``effect_size`` (``float``,
-    biolink-model#1774) and ``study_size`` (``int`` on the inlined ``Study``,
+    biolink-model#1774), ``number_of_cases`` (``int`` on the disease/phenotype
+    ``Association`` classes) and ``study_size`` (``int`` on the inlined ``Study``,
     biolink-model#1770) -- and keep the controlled decimal general string notation
     (``{:.4g}``) only while a slot stays untyped, since such values end up in
     human-readable text. Null values stay null.

tests/test_lib.py

@@ -1301,10 +1301,11 @@ def test_numeric_columns_matches_p_value_substring() -> None:
 
 
 def test_numeric_columns_matches_exact_names() -> None:
-    """numeric_columns matches exact effect size and study size names."""
-    names: list[str] = ["effect_size", "study_size", "cohort", "sample_size", "relationship_strength"]
+    """numeric_columns matches exact effect size, number-of-cases and study size names."""
+    names: list[str] = ["effect_size", "number_of_cases", "study_size", "cohort", "sample_size", "relationship_strength"]
     result: list[str] = numeric_columns(names)
     assert "effect_size" in result
+    assert "number_of_cases" in result
     assert "study_size" in result
     assert "cohort" not in result
     # Old names are superseded: coercion renames them before clean_numeric/format_numeric run.
@@ -1406,6 +1407,20 @@ def test_format_numeric_nulls_invalid_study_counts() -> None:
     assert result.schema["study_size"] == pl.Int64
 
 
+def test_format_numeric_emits_number_of_cases_as_int() -> None:
+    """number_of_cases leaves the pipeline as a real JSON int, not a TSV-text string.
+
+    biolink-model 4.4.4 types ``number_of_cases`` ``int`` on the disease/phenotype
+    ``Association`` classes, and the Rust ``uuid_on_collision: merge`` recompute already
+    writes an int union length, so first-wins edges must not ship the raw source string.
+    Fractional, negative, non-finite and non-numeric counts become null, like study counts.
+    """
+    lf: pl.LazyFrame = pl.DataFrame({"number_of_cases": ["1", "25", "0.42", "-3", "abc", None]}).lazy()
+    result: pl.DataFrame = format_numeric(clean_numeric(lf)).collect()
+    assert result["number_of_cases"].to_list() == [1, 25, None, None, None, None]
+    assert result.schema["number_of_cases"] == pl.Int64
+
+
 def test_format_numeric_preserves_nulls() -> None:
     """format_numeric preserves nulls as null."""
     lf: pl.LazyFrame = pl.DataFrame({"p_value": ["1e-8", "N/A", "0.05"]}).lazy()