diff --git a/crates/fgumi-metrics/src/dedup.rs b/crates/fgumi-metrics/src/dedup.rs
index 3b52d4448..853dd3715 100644
--- a/crates/fgumi-metrics/src/dedup.rs
+++ b/crates/fgumi-metrics/src/dedup.rs
@@ -353,9 +353,14 @@ impl crate::ProcessingMetrics for DeduplicationMetrics {
 }
 
 /// Serializable row for the `--duplication-ladder` sampled duplication
-/// saturation curve: "after `templates_seen` templates processed (in
-/// coordinate order) for this library, what cumulative fraction were
-/// duplicates".
+/// ladder: "after `templates_seen` templates processed (in coordinate order)
+/// for this library, what cumulative fraction were duplicates".
+///
+/// This is not a saturation curve. A template's duplicate status is decided
+/// by the other templates at its own position, so the fraction after N
+/// templates is the duplicate rate of the genome covered so far, not the rate
+/// a library sequenced to N templates would show. Changes along the ladder
+/// reflect regions with different duplicate rates.
 ///
 /// One row per (library, snapshot) — a library gets a row each time its
 /// cumulative `templates_seen` crosses a multiple of `--ladder-interval`,
@@ -376,10 +381,9 @@ pub struct DuplicationLadderMetrics {
     /// `templates_seen` on the first snapshot.
     pub window_templates: u64,
     /// Marginal duplicate fraction over just this window's templates
-    /// (`window_duplicate_templates / window_templates`). Often the more legible
-    /// view of the saturation curve than the cumulative `duplicate_fraction`,
-    /// since it isolates each depth band instead of averaging over all prior
-    /// ones. Mirrors dupblaster's per-window complexity columns.
+    /// (`window_duplicate_templates / window_templates`). Often more legible than
+    /// the cumulative `duplicate_fraction`, since it isolates each window
+    /// instead of averaging over all prior ones. Mirrors dupblaster's per-window complexity columns.
     #[serde(with = "crate::float")]
     pub window_duplicate_fraction: f64,
 }
diff --git a/src/lib/commands/dedup.rs b/src/lib/commands/dedup.rs
index fd2852a95..489057dc3 100644
--- a/src/lib/commands/dedup.rs
+++ b/src/lib/commands/dedup.rs
@@ -256,7 +256,7 @@ pub(crate) struct CollectedDedupCounts {
 }
 
 //////////////////////////////////////////////////////////////////////////////
-// Duplication saturation ladder (--duplication-ladder)
+// Duplication ladder (--duplication-ladder)
 //////////////////////////////////////////////////////////////////////////////
 
 /// Per-library cumulative counters and emitted snapshot rows backing
@@ -264,8 +264,8 @@ pub(crate) struct CollectedDedupCounts {
 ///
 /// # Ordering
 ///
-/// A saturation curve plots "after N templates processed, in coordinate
-/// order, what cumulative fraction were duplicates" — so [`Self::record`]
+/// The ladder records "after N templates processed, in coordinate order, what
+/// cumulative fraction were duplicates" — so [`Self::record`]
 /// MUST be called in strict serial/coordinate order, one call per position
 /// group. It is wired into the chain's `MiAssignDedup` step (see
 /// `pipeline::chains::commands::dedup::build_mi_assign_step`), which the
@@ -291,7 +291,7 @@ pub(crate) struct DuplicationLadderRecorder {
     rows: Vec<(u16, u64, u64, u64, u64)>,
 }
 
-/// Running cumulative state for one library's saturation ladder.
+/// Running cumulative state for one library's duplication ladder.
 #[derive(Default)]
 struct LadderLibraryState {
     /// Cumulative templates seen so far for this library.
@@ -1221,10 +1221,12 @@ pub struct MarkDuplicates {
     #[arg(short = 'H', long = "family-size-histogram")]
     pub family_size_histogram: Option<PathBuf>,
 
-    /// Path to write the sampled duplication saturation ladder: per-library
-    /// cumulative duplicate fraction vs. templates seen (in coordinate
-    /// order), snapshotted every `--ladder-interval` templates. Off by
-    /// default (no recorder is built, so no added work).
+    /// Path to write the sampled duplication ladder: per-library cumulative
+    /// duplicate fraction vs. templates seen (in coordinate order),
+    /// snapshotted every `--ladder-interval` templates. Because templates are
+    /// counted in coordinate order, this shows how the duplicate rate varies
+    /// along the genome; it is not a saturation curve. Off by default (no
+    /// recorder is built, so no added work).
     #[arg(long = "duplication-ladder")]
     pub duplication_ladder: Option<PathBuf>,
 
diff --git a/src/lib/pipeline/chains/builder.rs b/src/lib/pipeline/chains/builder.rs
index b22c1064b..5151aa163 100644
--- a/src/lib/pipeline/chains/builder.rs
+++ b/src/lib/pipeline/chains/builder.rs
@@ -6027,7 +6027,7 @@ impl<'a> ChainBuilder<'a> {
             dedup.include_unmapped,
             accumulators_for_process,
         );
-        // Duplication-saturation ladder recorder (`--duplication-ladder`).
+        // Duplication ladder recorder (`--duplication-ladder`).
         // `Some` only when the flag is set; shared (via `Arc`) between the serial
         // MI-assign step, which accumulates it in coordinate order, and the
         // finalize hook, which writes it. `None` means the MI-assign step does
diff --git a/src/lib/pipeline/chains/commands/dedup.rs b/src/lib/pipeline/chains/commands/dedup.rs
index 66270aa42..39acc5567 100644
--- a/src/lib/pipeline/chains/commands/dedup.rs
+++ b/src/lib/pipeline/chains/commands/dedup.rs
@@ -148,7 +148,7 @@ impl FinalizeHook for DedupFinalizeHook {
             write_family_size_histogram(&final_family_sizes, path)?;
         }
 
-        // Write the duplication-saturation ladder if requested. Accessed through
+        // Write the duplication ladder if requested. Accessed through
         // the lock (not `Arc::try_unwrap`): the serial MI-assign step holds a
         // clone of this `Arc` that may still be alive here.
         if let Some((path, recorder)) = duplication_ladder {
@@ -268,7 +268,7 @@ pub(crate) fn build_process_step(
 /// monotonically increasing MI offsets to each batch.
 ///
 /// When `ladder_recorder` is `Some`, this step also records the
-/// `--duplication-ladder` saturation curve — per position group, in this
+/// `--duplication-ladder` — per position group, in this
 /// serial/coordinate-order seam, **not** in the parallel serialize step.
 /// `MiAssign` is `Serial` + `ByItemOrdinal`, so batches reach this closure in
 /// input-record order and the groups within a batch are in coordinate order —