diff --git a/docs/pr-preservation/425-drain-log.md b/docs/pr-preservation/425-drain-log.md
new file mode 100644
index 00000000..e4ae64ea
--- /dev/null
+++ b/docs/pr-preservation/425-drain-log.md
@@ -0,0 +1,93 @@
+# PR #425 drain log — drain follow-up to #357: CommonMark 4-space-indent limit on fence detection
+
+PR: <https://github.com/Lucent-Financial-Group/Zeta/pull/425>
+Branch: `drain/357-followup-fence-indent`
+Drain session: 2026-04-25 (Otto, sustained-drain-wave during maintainer-
+asleep window; pre-summary-checkpoint earlier in this session)
+Thread count at drain: 1 substantive Codex finding on the parent #357
+fence-detection logic.
+Rebase context: clean rebase onto `origin/main`; no conflicts.
+
+Per Otto-250 (PR review comments + responses + resolutions are
+high-quality training signals): abbreviated Otto-268-wave record
+of the CommonMark parser-correctness fix. The abbreviated shape
+preserves reviewer/severity/outcome/commit metadata but does NOT
+preserve verbatim original-comment + reply text — see
+`docs/pr-preservation/_patterns.md` "Otto-250-canonical vs
+Otto-268-abbreviated shape divergence" for the full divergence
+framing and the canonical-shape contrast (e.g.
+`docs/pr-preservation/108-drain-log.md` /
+`docs/pr-preservation/395-drain-log.md`).
+
+This PR is the **post-merge cascade** to #357 (which introduced
+fence-detection logic for some markdown-aware operation). The
+cascade caught a single substantive correctness finding: the
+parent's fence-detection didn't respect CommonMark's 4-space-indent
+limit, producing a quiet-failure mode where tab-indented or
+deeply-indented fence-shaped lines were misclassified.
+
+---
+
+## Threads
+
+### Thread 1 — `lstrip(' ')` + explicit tab-rejection on fence detection
+
+- Reviewer: chatgpt-codex-connector
+- Severity: P1 (parser correctness)
+- Finding: parent #357's fence-detection used `raw_line.lstrip()`
+  (which silently consumes tabs) before checking the marker. Per
+  CommonMark §4.5, code fences are recognized only when the opening
+  marker is preceded by **at most 3 spaces** (not tabs; tabs in this
+  position make the line a code-block-content line, not a fence
+  marker). The unconditional `.lstrip()` consumed both spaces and
+  tabs, producing two failure modes:
+  - 4+ spaces of indent: silently treated as fence (CommonMark says
+    it should NOT be a fence at this indent — should be code-block
+    content).
+  - Tab indent: silently treated as fence (CommonMark says tabs
+    in this position make it not-a-fence).
+- Outcome: **FIX** — fence-detection now uses `lstrip(' ')` (space-
+  only, not whitespace) + explicit tab-rejection check. Tab-indented
+  fence-shaped lines correctly fail the marker check. 4+ spaces of
+  indent correctly fail the marker check (only 0-3 spaces of indent
+  are valid fence-marker-prefix per CommonMark §4.5).
+
+---
+
+## Pattern observations (Otto-250 training-signal class)
+
+1. **CommonMark spec compliance is its own findings class.** Custom
+   markdown parsers easily diverge from CommonMark §4.5 (fences),
+   §6.1 (code spans cannot contain newlines), §3.1 (thematic break
+   indent limit), etc. Codex catches this class reliably when the
+   parser code is in a PR. Fix template: cite the specific CommonMark
+   section being violated; pick the spec-compliant primitive
+   (`lstrip(' ')` for space-only stripping, `lstrip()` for whitespace
+   including tabs — but match the spec section's expected behavior).
+
+2. **`lstrip()` vs `lstrip(' ')` is a subtle but load-bearing
+   distinction in markdown parsing.** Python's `.lstrip()` with no
+   argument strips all whitespace including tabs; with `' '` it
+   strips only spaces. The CommonMark spec consistently distinguishes
+   the two — many indent-related primitives (`lstrip()` analogues
+   in any language) need to be space-aware not whitespace-aware.
+   Pre-commit-lint candidate: regex check on markdown-parser code
+   for unconditional `.lstrip()` calls suggesting `' '` arg.
+
+3. **Quiet-failure modes in markdown parsers are the most-dangerous
+   bug class.** Tab-indented fence-shaped lines being silently
+   misclassified produced no exception, no warning, no test
+   failure — just wrong-output rendered downstream. The fix
+   prevents the silent misclassification by making the
+   spec-compliance explicit. Pattern generalizes: any parser that
+   silently misclassifies-vs-rejects on edge cases needs explicit
+   reject-paths for known-tricky inputs.
+
+## Final resolution
+
+All threads resolved at SHA `1596a8f` (this PR's only commit).
+PR auto-merge SQUASH armed; CI cleared; merged to main as
+`693e171`.
+
+Drained by: Otto, sustained-drain-wave during maintainer-asleep
+window 2026-04-25, cron heartbeat `f38fa487` (`* * * * *`).