diff --git a/README.md b/README.md
index c965dca..c947e85 100644
--- a/README.md
+++ b/README.md
@@ -306,6 +306,8 @@ ccRewind/
 
 考古裡長出了追溯。你以為自己在翻歷史，其實你在追一條線：這個 bug 是哪段對話埋的？那個設計轉彎是在哪裡決定的？沒人規劃過這件事。是用出來的。
 
+然後那條線斷了。Session JSONL 預設 30 天就清掉，`--resume` 失效，session ID 指向的對話不見了。但 ccRewind 的 SQLite 還在——用 Python 內建的 `sqlite3` 查 `~/.ccrewind/index.db`，摘要、標籤、檔案歷史都撈得回來，貼進新的 session 就接上了。追溯又活了。這也不是規劃出來的。
+
 至於什麼內容值得留下來、我們又該怎麼把它收起來，這條線繼續走下去，長出了 ccRecall，成為 ccFamily 的一環。那又是另外一個故事了。
 
 ### Non-goals
diff --git a/README_EN.md b/README_EN.md
index 12645fe..02df412 100644
--- a/README_EN.md
+++ b/README_EN.md
@@ -305,6 +305,8 @@ I'd been browsing old sessions to revisit past design decisions — plain archae
 
 Archaeology turned into traceability. You think you're browsing history; really you're chasing a decision trail — which conversation buried this bug, where that design pivot happened. Nobody planned that workflow. It emerged from use.
 
+Then the trail went cold. Session JSONL files expire after 30 days by default — `--resume` breaks, the session ID points to nothing. But ccRewind's SQLite is still there. Query `~/.ccrewind/index.db` with Python's built-in `sqlite3` — summaries, tags, file history, all retrievable. Paste it into a new session and the thread picks up again. Traceability, revived. This too wasn't planned.
+
 As for which conversations are worth keeping, and how to file them away — that question outgrew this scope and became another tool: ccRecall, the second piece of ccFamily. But that's a story for another day.
 
 ### Non-goals