docs: a real example on recompute

benchmarks/benchmark_no_recompute.py

@@ -61,12 +61,19 @@ def main():
     t_recompute = bench_once(index_path_recompute, recompute=True)
     t_norecompute = bench_once(index_path_norecompute, recompute=False)
 
-    size_recompute = sum(
+    # Compute sizes only for files belonging to each index prefix
-        f.stat().st_size for f in Path(index_path_recompute).parent.iterdir() if f.is_file()
+    def _size_for(prefix: str) -> int:
-    )
+        p = Path(prefix)
-    size_norecompute = sum(
+        base = p.parent
-        f.stat().st_size for f in Path(index_path_norecompute).parent.iterdir() if f.is_file()
+        stem = p.stem  # e.g., 'recompute.leann'
-    )
+        total = 0
+        for f in base.iterdir():
+            if f.is_file() and f.name.startswith(stem):
+                total += f.stat().st_size
+        return total
+
+    size_recompute = _size_for(index_path_recompute)
+    size_norecompute = _size_for(index_path_norecompute)
 
     print("Benchmark results (HNSW):")
     print(

docs/configuration-guide.md

@@ -360,6 +360,15 @@ Trade-offs:
 - Significantly higher storage (10–100× vs selective recomputation)
 - Slightly larger memory footprint during build and search
 
+Real-world quick benchmark (HNSW, 5k texts; script `benchmarks/benchmark_no_recompute.py`):
+
+```text
+recompute=True:  ~6.58s; size ~1.1MB
+recompute=False: ~0.10s; size ~16.6MB
+
+Conclusion: no-recompute is much faster but uses more storage; recompute is smaller but has higher first-hop latency.
+```
+
 
 ## Further Reading