diff --git a/contracts/bin-aprender-data--aprender-data-v1.yaml b/contracts/bin-aprender-data--aprender-data-v1.yaml
index d675e245e7..07d215e335 100644
--- a/contracts/bin-aprender-data--aprender-data-v1.yaml
+++ b/contracts/bin-aprender-data--aprender-data-v1.yaml
@@ -171,3 +171,9 @@ falsification_tests:
     test: 'cargo test -p aprender-data --test format_edges'
     if_fails: 'a corrupt or foreign file is accepted, or a codec silently decodes as another'
     mutation: 'map compression code 0x03 to ZstdL3 in Compression::from_u8; compression_codes_roundtrip_and_unknown_is_none turns RED (measured 2026-10-18)'
+  - id: FALSIFY-BIN-ALIMENTAR-009
+    rule: 'quality checker severity, thresholds and report helpers hold their boundaries'
+    prediction: 'severity table is pinned (null ratio 0.5 -> 3, >0.5 -> 4); empty schema/dataset reports are single-issue; builders change what is flagged; duplicate and outlier checks are switchable'
+    test: 'cargo test -p aprender-data --test quality_checker_edges'
+    if_fails: 'a boundary moved or a disabled check still reports'
+    mutation: 'change `*null_ratio > 0.5` to `>= 0.5` in QualityIssue::severity; severity_table_is_pinned turns RED (measured 2026-10-18)'
diff --git a/crates/aprender-data/tests/quality_checker_edges.rs b/crates/aprender-data/tests/quality_checker_edges.rs
new file mode 100644
index 0000000000..af814ce442
--- /dev/null
+++ b/crates/aprender-data/tests/quality_checker_edges.rs
@@ -0,0 +1,340 @@
+//! Public-API edge cases for the data quality checker (coverage ticket, Rule 7).
+//! Contract: FALSIFY-BIN-ALIMENTAR-008.
+
+use std::sync::Arc;
+
+use alimentar::quality::{
+    ColumnQuality, NumericStats, QualityChecker, QualityIssue, QualityReport,
+};
+use alimentar::ArrowDataset;
+use arrow::{
+    array::{ArrayRef, BooleanArray, Float64Array, Int32Array, StringArray},
+    datatypes::{DataType, Field, Schema},
+    record_batch::RecordBatch,
+};
+
+fn ds(cols: Vec<(&str, ArrayRef)>) -> ArrowDataset {
+    let fields: Vec<Field> = cols
+        .iter()
+        .map(|(n, a)| Field::new(*n, a.data_type().clone(), true))
+        .collect();
+    let batch = RecordBatch::try_new(
+        Arc::new(Schema::new(fields)),
+        cols.into_iter().map(|(_, a)| a).collect(),
+    )
+    .expect("batch");
+    ArrowDataset::from_batch(batch).expect("dataset")
+}
+
+fn issue_cols(r: &QualityReport, pick: fn(&QualityIssue) -> bool) -> Vec<&str> {
+    r.issues
+        .iter()
+        .filter(|i| pick(i))
+        .filter_map(|i| i.column())
+        .collect()
+}
+
+#[test]
+fn severity_table_is_pinned() {
+    let c = || "c".to_string();
+    assert_eq!(QualityIssue::EmptySchema.severity(), 5);
+    assert_eq!(QualityIssue::EmptyDataset.severity(), 5);
+    assert_eq!(
+        QualityIssue::ConstantColumn {
+            column: c(),
+            value: "x".into()
+        }
+        .severity(),
+        4
+    );
+    let nulls = |r| QualityIssue::HighNullRatio {
+        column: c(),
+        null_ratio: r,
+        threshold: 0.1,
+    };
+    assert_eq!(nulls(0.9).severity(), 4);
+    assert_eq!(nulls(0.5).severity(), 3);
+    let outl = |r| QualityIssue::OutliersDetected {
+        column: c(),
+        outlier_count: 1,
+        outlier_ratio: r,
+    };
+    assert_eq!(outl(0.2).severity(), 3);
+    assert_eq!(outl(0.1).severity(), 2);
+    assert_eq!(
+        QualityIssue::DuplicateRows {
+            duplicate_count: 1,
+            duplicate_ratio: 0.5
+        }
+        .severity(),
+        2
+    );
+    assert_eq!(
+        QualityIssue::HighDuplicateRatio {
+            column: c(),
+            duplicate_ratio: 0.9,
+            threshold: 0.5
+        }
+        .severity(),
+        2
+    );
+    assert_eq!(
+        QualityIssue::LowCardinality {
+            column: c(),
+            unique_count: 1,
+            total_count: 9
+        }
+        .severity(),
+        1
+    );
+}
+
+#[test]
+fn column_accessor_only_for_column_issues() {
+    assert_eq!(QualityIssue::EmptySchema.column(), None);
+    assert_eq!(QualityIssue::EmptyDataset.column(), None);
+    assert_eq!(
+        QualityIssue::DuplicateRows {
+            duplicate_count: 1,
+            duplicate_ratio: 0.1
+        }
+        .column(),
+        None
+    );
+    let i = QualityIssue::LowCardinality {
+        column: "k".into(),
+        unique_count: 1,
+        total_count: 2,
+    };
+    assert_eq!(i.column(), Some("k"));
+}
+
+#[test]
+fn numeric_stats_iqr_and_fences() {
+    let s = NumericStats {
+        min: 0.0,
+        max: 100.0,
+        mean: 5.0,
+        std_dev: 1.0,
+        q1: 2.0,
+        median: 4.0,
+        q3: 6.0,
+    };
+    assert!((s.iqr() - 4.0).abs() < 1e-12);
+    assert!((s.outlier_lower_bound() - (-4.0)).abs() < 1e-12);
+    assert!((s.outlier_upper_bound() - 12.0).abs() < 1e-12);
+}
+
+#[test]
+fn column_quality_constant_and_mostly_null_boundaries() {
+    let mk = |unique, total, null_ratio| ColumnQuality {
+        name: "c".into(),
+        total_count: total,
+        null_count: 0,
+        null_ratio,
+        unique_count: unique,
+        unique_ratio: 0.0,
+        duplicate_count: 0,
+        duplicate_ratio: 0.0,
+        outlier_count: None,
+        numeric_stats: None,
+    };
+    assert!(mk(1, 5, 0.0).is_constant());
+    assert!(mk(0, 5, 0.0).is_constant());
+    assert!(!mk(2, 5, 0.0).is_constant());
+    assert!(!mk(0, 0, 0.0).is_constant());
+    assert!(mk(2, 5, 0.3).is_mostly_null(0.3)); // >= is inclusive
+    assert!(!mk(2, 5, 0.29).is_mostly_null(0.3));
+}
+
+#[test]
+fn empty_schema_and_empty_dataset_reports() {
+    let empty_schema = ArrowDataset::from_batch(RecordBatch::new_empty(Arc::new(Schema::empty())))
+        .expect("dataset");
+    let r = QualityChecker::new().check(&empty_schema).expect("check");
+    assert_eq!(r.issues, vec![QualityIssue::EmptySchema]);
+    assert_eq!(r.max_severity(), 5);
+    assert!(r.has_issues());
+
+    let schema = Arc::new(Schema::new(vec![Field::new("a", DataType::Int32, true)]));
+    let no_rows = ArrowDataset::from_batch(RecordBatch::new_empty(schema)).expect("dataset");
+    let r = QualityChecker::new().check(&no_rows).expect("check");
+    assert_eq!(r.issues, vec![QualityIssue::EmptyDataset]);
+    assert_eq!(r.column_count, 1);
+    assert_eq!(r.row_count, 0);
+}
+
+#[test]
+fn clean_dataset_has_no_issues_and_zero_severity() {
+    let d = ds(vec![(
+        "id",
+        Arc::new(Int32Array::from((0..20).collect::<Vec<_>>())),
+    )]);
+    let r = QualityChecker::new()
+        .with_outlier_check(false)
+        .check(&d)
+        .expect("check");
+    assert!(!r.has_issues(), "{:?}", r.issues);
+    assert_eq!(r.max_severity(), 0);
+    assert!(r.problematic_columns().is_empty());
+    assert_eq!(r.columns["id"].unique_count, 20);
+}
+
+#[test]
+fn null_heavy_constant_and_low_cardinality_columns_are_flagged() {
+    let d = ds(vec![
+        (
+            "nulls",
+            Arc::new(StringArray::from(vec![
+                Some("a"),
+                None,
+                None,
+                None,
+                Some("b"),
+            ])) as ArrayRef,
+        ),
+        ("flag", Arc::new(BooleanArray::from(vec![true; 5]))),
+        ("id", Arc::new(Int32Array::from(vec![1, 2, 3, 4, 5]))),
+    ]);
+    let r = QualityChecker::new().check(&d).expect("check");
+    assert_eq!(
+        issue_cols(&r, |i| matches!(i, QualityIssue::HighNullRatio { .. })),
+        vec!["nulls"]
+    );
+    assert_eq!(
+        issue_cols(&r, |i| matches!(i, QualityIssue::ConstantColumn { .. })),
+        vec!["flag"]
+    );
+    assert_eq!(
+        issue_cols(&r, |i| matches!(i, QualityIssue::LowCardinality { .. })),
+        vec!["flag"]
+    );
+    let mut cols = r.problematic_columns();
+    cols.sort_unstable();
+    assert_eq!(cols, vec!["flag", "nulls"]);
+    assert_eq!(r.column_issues("id").len(), 0);
+    assert!(r.column_issues("flag").len() >= 2);
+    assert_eq!(r.max_severity(), 4);
+    let c = r.issues.iter().find_map(|i| match i {
+        QualityIssue::ConstantColumn { value, .. } => Some(value.clone()),
+        _ => None,
+    });
+    assert_eq!(c.as_deref(), Some("true"));
+}
+
+#[test]
+fn threshold_builders_change_what_is_flagged() {
+    let d = ds(vec![(
+        "v",
+        Arc::new(StringArray::from(vec![
+            Some("a"),
+            None,
+            Some("b"),
+            Some("c"),
+        ])) as ArrayRef,
+    )]);
+    // 25% null: flagged at the 0.1 default, tolerated at 0.5.
+    let strict = QualityChecker::new().check(&d).expect("check");
+    assert!(strict
+        .issues
+        .iter()
+        .any(|i| matches!(i, QualityIssue::HighNullRatio { .. })));
+    let lax = QualityChecker::new()
+        .max_null_ratio(0.5)
+        .check(&d)
+        .expect("check");
+    assert!(!lax
+        .issues
+        .iter()
+        .any(|i| matches!(i, QualityIssue::HighNullRatio { .. })));
+    // min_cardinality above the distinct count flags LowCardinality.
+    let card = QualityChecker::new()
+        .max_null_ratio(0.5)
+        .min_cardinality(10)
+        .check(&d)
+        .expect("check");
+    assert!(card
+        .issues
+        .iter()
+        .any(|i| matches!(i, QualityIssue::LowCardinality { .. })));
+}
+
+#[test]
+fn duplicate_rows_flagged_and_switchable() {
+    let d = ds(vec![
+        (
+            "a",
+            Arc::new(Int32Array::from(vec![1, 1, 1, 2])) as ArrayRef,
+        ),
+        ("b", Arc::new(StringArray::from(vec!["x", "x", "x", "y"]))),
+    ]);
+    let on = QualityChecker::new().check(&d).expect("check");
+    assert_eq!(on.duplicate_row_count, 2);
+    assert!(on.issues.iter().any(|i| matches!(
+        i,
+        QualityIssue::DuplicateRows {
+            duplicate_count: 2,
+            ..
+        }
+    )));
+    let off = QualityChecker::new()
+        .with_duplicate_check(false)
+        .check(&d)
+        .expect("check");
+    assert_eq!(off.duplicate_row_count, 0);
+    assert!(!off
+        .issues
+        .iter()
+        .any(|i| matches!(i, QualityIssue::DuplicateRows { .. })));
+}
+
+#[test]
+fn outlier_detected_and_disabled() {
+    let mut v: Vec<f64> = (0..30).map(f64::from).collect();
+    v.extend([10_000.0, 20_000.0, -9_000.0]);
+    let d = ds(vec![("x", Arc::new(Float64Array::from(v)) as ArrayRef)]);
+    let on = QualityChecker::new()
+        .max_outlier_ratio(0.01)
+        .check(&d)
+        .expect("check");
+    let stats = on.columns["x"]
+        .numeric_stats
+        .as_ref()
+        .expect("numeric stats");
+    assert!(stats.max >= 20_000.0 && stats.min <= -9_000.0);
+    assert!(on.columns["x"].outlier_count.unwrap_or(0) >= 3);
+    assert!(on
+        .issues
+        .iter()
+        .any(|i| matches!(i, QualityIssue::OutliersDetected { .. })));
+    let off = QualityChecker::new()
+        .with_outlier_check(false)
+        .check(&d)
+        .expect("check");
+    assert!(off.columns["x"].outlier_count.is_none());
+    assert!(!off
+        .issues
+        .iter()
+        .any(|i| matches!(i, QualityIssue::OutliersDetected { .. })));
+}
+
+#[test]
+fn score_drops_with_issues() {
+    let good = ds(vec![(
+        "id",
+        Arc::new(Int32Array::from((0..20).collect::<Vec<_>>())) as ArrayRef,
+    )]);
+    let bad = ds(vec![(
+        "flag",
+        Arc::new(BooleanArray::from(vec![true; 20])) as ArrayRef,
+    )]);
+    let g = QualityChecker::new()
+        .with_outlier_check(false)
+        .check(&good)
+        .expect("check");
+    let b = QualityChecker::new()
+        .with_outlier_check(false)
+        .check(&bad)
+        .expect("check");
+    assert!(g.score > b.score, "good {} <= bad {}", g.score, b.score);
+}