diff --git a/apps/fparkan-game/src/main.rs b/apps/fparkan-game/src/main.rs
index 86891d7..cd27cda 100644
--- a/apps/fparkan-game/src/main.rs
+++ b/apps/fparkan-game/src/main.rs
@@ -473,7 +473,13 @@ fn prepare_load_progress_path(path: &std::path::Path) -> Result<(), String> {
 }
 
 fn write_load_progress(path: &std::path::Path, phase: MissionLoadPhase) -> Result<(), String> {
-    std::fs::write(path, format!("{phase:?}\n")).map_err(|err| format!("{}: {err}", path.display()))
+    use std::io::Write;
+
+    let mut file = std::fs::OpenOptions::new()
+        .append(true)
+        .open(path)
+        .map_err(|err| format!("{}: {err}", path.display()))?;
+    writeln!(file, "{phase:?}").map_err(|err| format!("{}: {err}", path.display()))
 }
 
 fn run_static_vulkan_mode(
@@ -1049,6 +1055,23 @@ mod tests {
         );
     }
 
+    #[test]
+    fn load_progress_retains_prior_milestones() -> Result<(), String> {
+        let path =
+            std::env::temp_dir().join(format!("fparkan-game-progress-{}.txt", std::process::id()));
+        prepare_load_progress_path(&path)?;
+        write_load_progress(&path, MissionLoadPhase::GraphVisualMaterials)?;
+        write_load_progress(&path, MissionLoadPhase::GraphVisualMaterialRequests(64))?;
+        let progress = std::fs::read_to_string(&path).map_err(|err| err.to_string())?;
+        std::fs::remove_file(&path).map_err(|err| err.to_string())?;
+
+        assert_eq!(
+            progress,
+            "Starting\nGraphVisualMaterials\nGraphVisualMaterialRequests(64)\n"
+        );
+        Ok(())
+    }
+
     #[test]
     fn static_preview_component_merge_offsets_indices_and_remaps_local_selectors(
     ) -> Result<(), String> {
diff --git a/docs/tomes/05-render.md b/docs/tomes/05-render.md
index c49ca66..faf85d2 100644
--- a/docs/tomes/05-render.md
+++ b/docs/tomes/05-render.md
@@ -1126,6 +1126,12 @@ termination. It does not preserve the simultaneous TEXM count, so it neither
 identifies a slow individual material nor attributes time to MAT0 parsing; the
 next profiler revision needs one cumulative snapshot of all classes.
 
+`fparkan-game --load-progress` now initializes its file with `Starting` and
+appends each later mission-load event instead of replacing the prior line. A
+bounded probe can therefore retain both MAT0 and TEXM request milestones even
+when a later event is last. The behavior is diagnostic persistence only: it
+does not alter graph traversal, resource validation, or rendering.
+
 Mission loading now raises the decoded-payload cache entry budget from 64 to
 256 while retaining its 64 MiB byte budget. This avoids premature entry-count
 eviction during resource-rich loads without making memory unbounded. The same