diff --git a/KIPs/kip-237.md b/KIPs/kip-237.md
index 2ac68cb..0db97ee 100644
--- a/KIPs/kip-237.md
+++ b/KIPs/kip-237.md
@@ -27,14 +27,8 @@ This KIP proposes to compress the three data types to reduce the total disk size
 
 ### Compression
 A node periodically compresses storage in a background thread and removes the original data once it is confirmed safe to do so.
-Data is compressed in units called "chunks".
-An optimal chunk size should balance a high compression rate with minimal decompression overhead.
-Determining the ideal chunk capacity remains an open problem. However, our experiments suggest that a 1MB chunk size offers a favorable trade-off.
-
-![compression-size](../assets/kip-237/chunk-selecting-compression.png)
-![decompression](../assets/kip-237/chunk-selecting-decompression.png)
+Data is compressed in units called "chunks". Selecting a good chunk size is described in
 
-Currently, the chunk size is fixed at 1MB and cannot be configurable.
 
 ### Decompression
 Decompression is required when a node receives a query (header, body, or receipts).
@@ -61,10 +55,22 @@ Evaluation on a [state-migrated mainnet (as of 2024-12-02)](https://packages.kai
 | Body | 1100 | 267 | 75.7% |
 | Receipts | 1600 | 179 | 88.8% |
 | Others | ___ | ___ | - |
-| Total | ___ | ___ | __._% |
+| Total | ___ | ___ | 57.21% |
+
 It took 3 days for background compression threads to complete.
 
-### Ecosystem changes
+## Rationale
+
+### Appropriate chunk size
+An optimal chunk size should balance a high compression rate with minimal decompression overhead.
+Determining the ideal chunk capacity remains an open problem. However, our experiments suggest that a 1MB chunk size offers a favorable trade-off.
+
+![compression-size](../assets/kip-237/chunk-selecting-compression.png)
+![decompression](../assets/kip-237/chunk-selecting-decompression.png)
+
+Currently, the chunk size is fixed at 1MB and cannot be configurable.
+
+## Backwards Compatibility
 An existing database can be compressed without node shutdown.
 APIs querying the historic blocks, transactions, and receipts must work correctly because the data are not lost.