diff --git a/crates/arbor-watcher/examples/centrality_dist.rs b/crates/arbor-watcher/examples/centrality_dist.rs
new file mode 100644
index 0000000..9969eb4
--- /dev/null
+++ b/crates/arbor-watcher/examples/centrality_dist.rs
@@ -0,0 +1,55 @@
+//! Compares the old max-normalized centrality distribution against percentile
+//! rank, to calibrate risk thresholds against evidence rather than intuition.
+//!
+//! Usage: cargo run -p arbor-watcher --example centrality_dist -- <dir>
+
+use arbor_graph::compute_centrality;
+use arbor_watcher::{index_directory, IndexOptions};
+use std::path::Path;
+
+fn main() {
+    let dir = std::env::args().nth(1).unwrap_or_else(|| ".".to_string());
+    let result = index_directory(Path::new(&dir), IndexOptions::default()).expect("index");
+    let graph = result.graph;
+
+    let scores = compute_centrality(&graph, 20, 0.85);
+    let nodes: Vec<_> = graph.node_indexes().collect();
+    let n = nodes.len();
+    if n == 0 {
+        println!("empty graph");
+        return;
+    }
+
+    // Raw PageRank, then the old max-normalization: score / max.
+    let raw: Vec<f64> = nodes.iter().map(|&i| scores.get_raw(i)).collect();
+    let max = raw.iter().cloned().fold(0.0f64, f64::max);
+    let max_norm: Vec<f64> = raw.iter().map(|r| if max > 0.0 { r / max } else { 0.0 }).collect();
+    let pct: Vec<f64> = nodes.iter().map(|&i| scores.get(i)).collect();
+
+    let frac_above = |v: &[f64], t: f64| v.iter().filter(|x| **x > t).count() as f64 / n as f64;
+
+    println!("\n{dir}");
+    println!("  nodes {n}");
+    println!("\n  threshold   max-normalized   percentile");
+    for t in [0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95, 0.98] {
+        println!(
+            "  > {t:<9.2} {:>12.2}%  {:>10.2}%",
+            frac_above(&max_norm, t) * 100.0,
+            frac_above(&pct, t) * 100.0
+        );
+    }
+
+    // What percentile selects the same population the old 0.7 / 0.4 bars did?
+    let target_high = frac_above(&max_norm, 0.7);
+    let target_med = frac_above(&max_norm, 0.4);
+    println!(
+        "\n  old HIGH bar (max-norm > 0.70) selected {:.2}% of nodes -> percentile {:.3}",
+        target_high * 100.0,
+        1.0 - target_high
+    );
+    println!(
+        "  old MED  bar (max-norm > 0.40) selected {:.2}% of nodes -> percentile {:.3}",
+        target_med * 100.0,
+        1.0 - target_med
+    );
+}