diff --git a/firmware/acid-firmware/src/db/mod.rs b/firmware/acid-firmware/src/db/mod.rs
index 47d88d6..7b433c9 100644
--- a/firmware/acid-firmware/src/db/mod.rs
+++ b/firmware/acid-firmware/src/db/mod.rs
@@ -146,31 +146,34 @@ type DbPathSegment<'a> = Cow<'a, str>;
 type DbPathBuf<'a> = Vec<DbPathSegment<'a>>;
 type DbPath<'a> = [DbPathSegment<'a>];
 
-struct DbKey(Vec<u8>);
+struct DbKey {
+    /// Segments separated by `0x00`, with the whole thing suffixed with `[0x00, 0xFF]`.
+    bytes: Vec<u8>,
+}
 
 impl Deref for DbKey {
     type Target = [u8];
 
     fn deref(&self) -> &Self::Target {
-        &self.0[0..(self.0.len() - 2)]
+        &self.bytes[0..(self.bytes.len() - 2)]
     }
 }
 
 impl DbKey {
     fn from_raw(mut key: Vec<u8>) -> Self {
-        key.extend_from_slice(&[0, 1]);
-        Self(key)
+        key.extend_from_slice(&[0x00, 0xFF]);
+        Self { bytes: key }
     }
 
     fn new<'a>(path: impl IntoIterator<Item = DbPathSegment<'a>>) -> Self {
         // Null bytes are not allowed within path segments, and will cause a panic.
-        // The segments are separated by `[0, 0]`.
-        // Two null-bytes are used to allow for easy range lookups by suffixing the key with:
-        // `[0]..[0, 1]`
-        // To avoid reallocations, we always suffix the key with `[0, 1]`.
-        // Then, a specific key can be looked up using by omitting the last two bytes.
-        // By omitting one byte, you get the start of the range of all paths within this path.
-        // By not omitting any bytes, you get the end of that range.
+        // Bytes of `0xFF` cannot appear in valid UTF-8.
+        // The byte vector stored in a `DbKey` is formed by interspersing the segments with `0x00`, and suffixing the key with `[0x00, 0xFF]`.
+        // This lets us represent three significant keys with a single allocation:
+        // * The key for querying the path itself: `bytes[..bytes.len() - 2]`, e.g. `b"first\x00second"`
+        // * The keys for range-querying the path's children:
+        //   * Start (inclusive): `bytes[..bytes.len() - 1]`, e.g. `b"first\x00second\x00"`
+        //   * End (exclusive): `bytes[..]`, e.g. `b"first\x00second\x00\xFF"`
 
         let mut bytes = Vec::new();
 
@@ -179,22 +182,23 @@ impl DbKey {
                 !segment.as_bytes().contains(&0x00),
                 "A path segment must not contain null bytes."
             );
+            // No need to check for `0xFF` bytes in UTF-8 strings.
 
             bytes.extend_from_slice(segment.as_bytes());
-            bytes.extend_from_slice(&[0, 0]);
+            bytes.push(0x00);
         }
 
         if let Some(last_byte) = bytes.last_mut() {
-            *last_byte = 1;
+            bytes.push(0xFF);
         } else {
             panic!("An empty path is not a valid path.");
         }
 
-        DbKey(bytes)
+        DbKey { bytes }
     }
 
     fn range_of_children(&self) -> Range<&[u8]> {
-        (&self.0[0..(self.0.len() - 1)])..(&self.0[..])
+        (&self.bytes[0..(self.bytes.len() - 1)])..(&self.bytes[..])
     }
 
     fn segments(&self) -> impl Iterator<Item = DbPathSegment<'_>> {
@@ -209,7 +213,7 @@ impl DbKey {
                 if let Some(end_index) = self.rest.iter().position(|byte| *byte == 0) {
                     let segment = &self.rest[..end_index];
                     let segment = str::from_utf8(segment).unwrap();
-                    self.rest = &self.rest[end_index + 2..];
+                    self.rest = &self.rest[end_index + 1..];
 
                     Some(Cow::Borrowed(segment))
                 } else {
@@ -219,7 +223,7 @@ impl DbKey {
         }
 
         SegmentIterator {
-            rest: self.0.as_slice(),
+            rest: self.bytes.as_slice(),
         }
     }
 }