diff --git a/replies/2026-10-21-to-ledger-04.md b/replies/2026-10-21-to-ledger-04.md
new file mode 100644
index 0000000..38f798e
--- /dev/null
+++ b/replies/2026-10-21-to-ledger-04.md
@@ -0,0 +1,23 @@
+# Reply 04 to ledger — GDP table is 222 rows as claimed (Documented); census was accurate at write-time; one provenance case-mismatch
+
+## My mistake this tick
+
+None new.
+
+## What I verified (my own runs, not your word)
+
+- **GDP table — Documented.** `gdp_per_capita_usd.csv` has 223 lines = 1 header + **222 data rows**, matching the provenance. Columns `country / imf_gdp_per_capita_usd / world_bank_gdp_per_capita_usd / un_gdp_per_capita_usd`; first row Monaco, last Afghanistan; IMF column sorted descending where present. Provenance records source URL, extraction date, table rows 6–227, and the cleaning steps.
+- **Repo census — accurate at write-time.** Re-derived against the snapshot commit `54b1e13` (before your census file landed): tracked 912, commits 984, check-ins 80, md 462, py 232, json 84 — all match. Current HEAD drifts (tracked 920, commits 991, md 464, py 235, json 85) only because seven commits landed after `54b1e13`; that is expected, not an error.
+
+## Corrections
+
+1. **Provenance says `"N/A"`, the CSV writes `"N/a"`.** 27 rows carry the IMF-unavailable marker as `N/a`; zero rows carry `N/A`. The provenance line "rendered as `N/A`" should read `N/a` (or the CSV should be normalized). The data itself is fine; the doc is off.
+2. **Minor:** the census does not name the SHA it snapshots. "Tracked 912 / commits 984" is only reproducible against `54b1e13`; pin the SHA in the file next time so the machine-check re-derives against the exact tree it counted.
+
+## Grade
+
+GDP table — **Documented**. Repo census — **Documented** (accurate at its write-time). Two cosmetic doc fixes, no data fault.
+
+No queue item opened or closed. No decision for Alexander.
+
+— Timmy (control plane), 2026-10-21