diff --git a/CHANGELOG.md b/CHANGELOG.md
index 5cde8de..2ba5047 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -161,6 +161,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/).
 - XLSX packages missing `_rels/.rels` no longer fail inside calamine.
 - Release builds use `panic = "unwind"`, so a parser panic becomes an error instead of aborting the
   CLI or the host Python/Node process.
+- Text that only looks like a block marker is no longer escaped: `4.6 out of 5` stays `4.6`, not
+  `4\.6`; same for `#tag` and `+1`. Markers followed by whitespace (`1. `, `# `, `+ `) are still
+  escaped.
 - Exclude selectors now apply to `<li>`, `<tr>`, `<th>` and `<td>`; they were ignored there.
 - Indented HTML (most real web pages) now keeps blank lines between blocks; previously
   headings, paragraphs, lists and tables after the first ran together with a leading space
diff --git a/crates/markkit/src/render/escape.rs b/crates/markkit/src/render/escape.rs
index aca047c..3621b59 100644
--- a/crates/markkit/src/render/escape.rs
+++ b/crates/markkit/src/render/escape.rs
@@ -42,17 +42,33 @@ pub(crate) fn push_text(buf: &mut String, s: &str, pos: Position) -> bool {
                 continue;
             }
             line_start = false;
+            // A marker only opens a block when followed by whitespace. At the end of
+            // the run the next node may supply it, so escape there too.
+            let opens = |at: usize| chars.get(at).is_none_or(|n| *n == ' ' || *n == '\t');
             match c {
-                '#' | '>' | '-' | '+' | '=' => {
+                // `-` (thematic break, setext), `=` (setext), `>` (quote): always.
+                '>' | '-' | '=' => {
                     buf.push('\\');
                     buf.push(c);
                     i += 1;
                     continue;
                 }
+                '#' | '+' => {
+                    let run = chars[i..].iter().take_while(|d| **d == c).count();
+                    if c == '#' && opens(i + run) || c == '+' && opens(i + 1) {
+                        buf.push('\\');
+                        buf.push(c);
+                        i += 1;
+                        continue;
+                    }
+                }
                 '0'..='9' => {
-                    // Ordered list marker: up to 9 digits then `.` or `)`.
+                    // Ordered list marker: up to 9 digits, `.` or `)`, then whitespace.
                     let digits = chars[i..].iter().take_while(|d| d.is_ascii_digit()).count();
-                    if digits <= 9 && matches!(chars.get(i + digits), Some('.') | Some(')')) {
+                    if digits <= 9
+                        && matches!(chars.get(i + digits), Some('.') | Some(')'))
+                        && opens(i + digits + 1)
+                    {
                         buf.extend(&chars[i..i + digits]);
                         buf.push('\\');
                         buf.push(chars[i + digits]);
diff --git a/crates/markkit/tests/output_safety.rs b/crates/markkit/tests/output_safety.rs
index 84f485c..7c712fd 100644
--- a/crates/markkit/tests/output_safety.rs
+++ b/crates/markkit/tests/output_safety.rs
@@ -124,6 +124,42 @@ fn line_start_block_markers_are_escaped() {
     }
 }
 
+#[test]
+fn line_start_marker_lookalikes_render_verbatim() {
+    // Not block markers: no whitespace after `#`, `+`, or the ordered `.`/`)`.
+    for s in [
+        "4.6 out of 5 stars",
+        "3.14 is pi",
+        "1)x",
+        "#hashtag",
+        "##tag",
+        "+1 for this",
+        "2024.10.27 release",
+    ] {
+        let md = render(vec![para(vec![text(s)])]);
+        assert_eq!(md, format!("{s}\n"), "{s:?} over-escaped");
+        assert_eq!(block_tags(&md), vec!["Paragraph"], "{s:?} -> {md:?}");
+    }
+}
+
+#[test]
+fn line_start_marker_at_run_end_stays_escaped() {
+    // The next inline node may start with whitespace and complete the marker.
+    for (a, b) in [
+        ("1.", " x"),
+        ("#", " h"),
+        ("##", " h"),
+        ("+", " li"),
+        ("7)", "\tx"),
+    ] {
+        let md = render(vec![para(vec![text(a), text(b)])]);
+        assert_eq!(block_tags(&md), vec!["Paragraph"], "{a:?}+{b:?} -> {md:?}");
+        let md = render(vec![para(vec![text(a)])]);
+        assert_eq!(block_tags(&md), vec!["Paragraph"], "{a:?} -> {md:?}");
+        assert_eq!(visible_text(&md), a, "{a:?} -> {md:?}");
+    }
+}
+
 #[test]
 fn line_start_after_hard_break_is_escaped() {
     let md = render(vec![para(vec![
diff --git a/knowledge/log.md b/knowledge/log.md
index 58ade41..1f1eb5e 100644
--- a/knowledge/log.md
+++ b/knowledge/log.md
@@ -3,6 +3,7 @@
 ## 2026-10-27
 
 * **XLSB support (#11)**: Excel binary workbooks are read by a new bounds-checked BIFF12 reader (`spreadsheet/xlsb.rs`) instead of being rejected; calamine's XLSB reader was not used because it panics on malformed records. Same caps, hidden-sheet and macro rules as XLSX. See [Format Support](formats/format-support.md) and [Threat Model](security/threat-model.md) T6c.
+* **Over-escaping fix**: line-start `#`, `+` and ordered markers are escaped only when followed by whitespace or the end of the text run, so `4.6 out of 5` no longer renders `4\.6` (eval gap). `> - =` stay always escaped. See [Threat Model](security/threat-model.md) T1, [Evaluation Harness](operations/eval.md).
 * **MCP server**: `markkit mcp` serves `MarkkitTool` over MCP stdio (initialize, ping, tools/list, tools/call); line size capped from `--max-size`, tool failures as `isError` results. New [MCP Server](integrations/mcp-server.md) concept; MCP messages row in the [Threat Model](security/threat-model.md).
 * **Spreadsheet follow-ups (#11)**: single-row sheets render as data under column letters; XLSX/ODS/XLS titles come from document properties; XLSB content type no longer maps to XLSX; XLSX without `_rels/.rels` parses; trailing-newline output documented as intended; release profile switched to `panic = "unwind"` so `catch_unwind` is effective in markkit's own binaries. See [Format Support](formats/format-support.md) and [Threat Model](security/threat-model.md) T6c.
 * **Plain text and Markdown input**: text is always on and goes through the IR; Markdown (`markdown` feature) is re-parsed with pulldown-cmark rather than passed through, so raw HTML never reaches output. Neither is sniffed from bytes. Shares the call's `Limits` budget (timeout, nodes, nesting depth). New threat T7 and invariant S18. See [Format Support](formats/format-support.md) and [Threat Model](security/threat-model.md).
diff --git a/knowledge/operations/eval.md b/knowledge/operations/eval.md
index 77352bb..6d0e214 100644
--- a/knowledge/operations/eval.md
+++ b/knowledge/operations/eval.md
@@ -152,7 +152,7 @@ collapsed to one column and dropped every `<td>`, fixed by IR v2 row-header
   question title and body are dropped (forum 81% default vs 62%).
 - **`main_content` on the wiki page** keeps the TOC, `[edit]` links and
   categories.
-- **Over-escaping** (since #23): a paragraph starting `4.6 out of 5` renders
+- **Over-escaping** (since #23, fixed 2026-10-27): a paragraph starting `4.6 out of 5` renders
   `4\.6 ...`; `4.6` is not an ordered-list marker (needs `4.` + space), so the
   backslash is noise for text consumers (commerce, both modes; before #23
   main_content passed the page).
diff --git a/knowledge/security/threat-model.md b/knowledge/security/threat-model.md
index 2eaa002..56aa6d3 100644
--- a/knowledge/security/threat-model.md
+++ b/knowledge/security/threat-model.md
@@ -41,7 +41,8 @@ that renders as headings, links, code fences or tables).
 - Output is markdown text with no HTML pass-through, enforced at render time by
   `render/escape.rs` for every format (the IR keeps raw text):
   - Text escapes `` ` * [ ] < ~ | ``, `_` at word boundaries, entity-like `&`, a
-    trailing `\` or `!`, and block markers (`# > - + =`, `1.`/`1)`) at line start.
+    trailing `\` or `!`, and block markers at line start: `> - =` always, `#` runs,
+    `+` and `1.`/`1)` only before whitespace or the end of the text run.
     Newlines inside text become spaces.
   - Link schemes allowlisted (`http`, `https`, `mailto`, `tel`, `ftp`, relative);
     image schemes `http`, `https`, relative and raster `data:image/*`. Others are