diff --git a/README.md b/README.md
index ec6f3c0..f98ec8f 100644
--- a/README.md
+++ b/README.md
@@ -127,8 +127,9 @@ Zero dependencies beyond `rich`. No API keys, no network: it reads local files.
 **Timestamps without an offset are treated as UTC.** Explicit offsets are respected when sorting
 runs and calculating durations. Runs with missing or invalid timestamps sort before dated runs.
 
-**Two passes over the transcript, not one.** Results can appear before every use has been seen in
-unusual orderings. A 34MB file is cheap to scan twice compared to getting the pairing subtly wrong.
+**One pass over the transcript, with two maps.** `tool_use` and `tool_result` blocks are collected
+into separate maps and joined by `tool_use_id` afterwards. This keeps pairing correct when records
+appear in unusual orderings.
 
 **A torn final line is skipped, not fatal.** A live session being appended to yields half-written
 JSON. Refusing to parse would mean you cannot analyse a run until it is over, which is exactly
diff --git a/agentrace/parse.py b/agentrace/parse.py
index 56990a9..261f2f0 100644
--- a/agentrace/parse.py
+++ b/agentrace/parse.py
@@ -116,8 +116,8 @@ def _records(path: Path) -> Iterator[dict]:
 def parse_session(path: Path) -> Session:
     """Pull every Agent delegation out of one transcript.
 
-    Two passes over the file rather than one: results can appear before we have seen every use in
-    weird orderings, and a 34MB file is cheap to scan twice compared to getting this subtly wrong.
+    One pass collects tool_use and tool_result blocks into separate maps, then joins them by
+    tool_use_id. Buffering both sides keeps pairing correct when records arrive in a weird order.
     """
     uses: dict[str, dict] = {}
     results: dict[str, dict] = {}