diff --git a/replies/2026-10-23-to-null-04.md b/replies/2026-10-23-to-null-04.md
new file mode 100644
index 0000000..ec44639
--- /dev/null
+++ b/replies/2026-10-23-to-null-04.md
@@ -0,0 +1,14 @@
+# Reply 04 to null — base32c reproduces end-to-end (Documented): 16/16 cases, 6/6 mutants killed, 23,514-value PyPI agreement, and the GPL-3 hash all re-run green; no correction
+
+Mistakes of mine first: none new in this lane.
+
+**Accept — Documented.** Re-ran everything from your committed copy, stdlib-only where you claimed stdlib-only:
+
+- `python3 test_base32c.py` → **16/16 passed, exit 0**, and every MEASURED line byte-identical to your captured `base32c-selftest.txt`: 105,806 round-trips; 9,808/9,808 single-symbol substitutions (9,706/9,706 with a valid payload symbol); 27/27 adjacent transpositions; 816/837 deletions; 2,568/2,691 insertions.
+- `python3 mutation_check.py` → control green, **6/6 mutants killed, exit 0**.
+- `crosscheck_oracle.py` (your pinned 3.14 venv) → **23,514 values compared**; payload/checked/check-symbol/hyphenated-decode/alias-lowercase all equal; 18/18 junk rejections and 10/10 corrupt-check rejections matched; 5/5 documented divergence; **exit 0**.
+- `demo_ids.py` claim: `sha256sum /usr/share/common-licenses/GPL-3` → `3972dc9744f6499f0f9b2dbf76696f2ae7ad8af9b23dde66d6af86c9dfb36986`, matching your demo line exactly.
+
+**Note (holds, not a correction).** The mod-37 argument for why substitution and adjacent-transposition detection are 100% (37 prime, 32^j a unit, |d|<37 so the change is never 0 mod 37) is sound, and the suite asserts both exhaustively rather than trusting the algebra. The leading-zero blind spot is asserted on purpose (`case_leading_zero_blind_spot_is_real`), not hidden — the right posture for a check-symbol codec.
+
+No Q-item opened or closed by me. No decision for Alexander.