diff --git a/checkins/2026-10-19-mercury.md b/checkins/2026-10-19-mercury.md
new file mode 100644
index 0000000..510955a
--- /dev/null
+++ b/checkins/2026-10-19-mercury.md
@@ -0,0 +1,18 @@
+# 2026-10-19 — mercury
+
+**Did:**
+- **Live census:** `curl -s 'https://alexanderwhitestone.com/stackchain/blocks/' | grep -o '<li id=' | wc -l` → **121** (rc 0). Badge breakdown: `1 badge corroborated`, `115 badge documented`, `5 badge testimony` (rc 0). No `unrecovered` badge present.
+- **R5-b State.db pair (pre-window):** `stat -c '%s | mtime=%y | ctime=%z | %n' /root/.hermes/state.db /root/.hermes/state.db-wal` → **4,159,479,808 B** (mtime 11:59:02, ctime 11:59:56) and **67,108,864 B** (mtime 11:59:46, ctime 12:00:18) (rc 0).
+- **R5-b Header bytes:** `python3 -c "f=open('/root/.hermes/state.db','rb');h=f.read(100);f.close();print('write_version',h[18],'read_version',h[19])"` → **write_version 2 read_version 2** (rc 0).
+- **R5-b Window:** `sleep 300` pair timed out after 300s with **no output**. The window measurement failed to complete.
+- **R5-b State.db pair (post-window):** `stat -c '%s | mtime=%y | ctime=%z | %n' /root/.hermes/state.db /root/.hermes/state.db-wal` → **4,160,958,464 B** (mtime 12:04:43, ctime 12:05:00) and **67,108,864 B** (mtime 12:05:11, ctime 12:05:19) (rc 0).
+- **R5-a Digests (fresh):** `head -c "$N" /root/.hermes/state.db | sha256sum` → **`b5fedbfec30513781241ed0205309e31d115be035223358e7e2417bab539f96b`** (prefix 4,160,958,464 B, rc 0). `head -c "$N" /root/.hermes/state.db-wal | sha256sum` → **`c5b6071236f297faa390aa8e67df7d64e6a99cc041bc3ddada051d8eb5c13b3c`** (prefix 67,108,864 B, rc 0).
+- **R5-a Digests (historical):** `sha256sum /root/.hermes/state.db and /root/.hermes/state.db-wal, plus the two digest domains` → **rc 127** ("No such file"). The command was malformed (shell parsed the string as a single filename). The historical digests (`41c8be74…`, `4a2d811b…`) remain **Unrecovered by construction** as stated in reply 09; this run does not change that grade.
+
+**Blocked:** The 300s window measurement timed out with no output. I have pre/post stats showing state.db grew **1,478,656 B** (~1.4 MiB) between 11:59:02 and 12:04:43, but the exact window duration is unverified because the `sleep 300` command did not complete. I will not divide by an unverified time to manufacture a rate.
+
+**Next:** Re-run the window measurement with a shorter sleep (e.g., 60s) to get a clean pre/post pair with a verified duration.
+
+**Ask:** None.
+
+Standing census unchanged from 2026-10-18: 121 = 1/115/5/0.