diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index 16739462..9cf1d3f7 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -122,6 +122,11 @@ this package adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 
 ## [Unreleased]
 
+- Let `/Prev` and `/XRefStm` be ordinary PDF dictionary keys: they mark an unsupported
+  incremental update only as trailer keys, so an outline item's `/Prev` no longer makes the
+  whole document unreadable (#427).
+- 允许 `/Prev` 与 `/XRefStm` 作为普通 PDF 字典键：仅当它们出现在 trailer 中才表示不支持的增量更新，大纲项里的 `/Prev` 不再使整份文档被判不可读（#427）。
+
 - Add current-scope `ops_status` diagnostics and public `diagnose(ctx, agent)` inspection; report integration, host/runtime, presentation and external provider readiness without executing or activating capabilities (#440).
 - 新增当前作用域 `ops_status` 与公开 `diagnose(ctx, agent)`，区分接入、宿主/运行时、展示及外部提供者状态，不执行或激活能力（#440）。
 - Expose opt-in Native capability providers with original names/contracts, explicit preset allow/startup/disable controls, per-instance selection invalidation and provider-owned authorization; reuse scoped discovery publication without requiring the OPS worker for external-only sessions (#440).
diff --git a/docs/native-kernel.md b/docs/native-kernel.md
index 35bd580a..4200fdff 100644
--- a/docs/native-kernel.md
+++ b/docs/native-kernel.md
@@ -246,7 +246,10 @@ xref table, an exact final startxref/EOF footer, unique dictionary keys and obje
 identities, and directly declared stream lengths. Every active xref offset must point
 exactly at its object header. Objects cannot overlap; gaps must contain only whitespace
 or comments. This deliberately rejects incremental updates, encryption, object/xref
-streams, recovery scans and ambiguous duplicate data, including decoded forbidden names.
+streams, recovery scans and ambiguous duplicate data. `/Encrypt`, `/ObjStm` and `/XRef`
+are refused wherever they appear, decoded escapes included, while `/Prev` and `/XRefStm` are
+refused only as trailer keys — the one place they denote an earlier xref section — so a
+dictionary that merely names them, such as an outline item's previous sibling, stays readable.
 
 Bounds are checked before indexed allocation: 16 MiB per input, 10,000 xref entries,
 100,000 metadata values, depth 64, 4,096 keys per dictionary, 256 bytes per name/token,
diff --git "a/docs/\345\216\237\347\224\237\345\206\205\346\240\270.md" "b/docs/\345\216\237\347\224\237\345\206\205\346\240\270.md"
index ad24d94b..2d8a0006 100644
--- "a/docs/\345\216\237\347\224\237\345\206\205\346\240\270.md"
+++ "b/docs/\345\216\237\347\224\237\345\206\205\346\240\270.md"
@@ -196,8 +196,10 @@ FILE_CHANGED 始终使所请求页不完整。引用页生成的 source 类型
 内部 PDF 容器解析器仅接受受限的 PDF 1.0–1.7 子集：单个传统 xref 表、
 准确的末尾 startxref/EOF、唯一字典键及对象标识，以及直接整数形式的流长度。
 每个有效 xref 偏移必须准确指向对象头；对象不可重叠，间隙只能包含空白或注释。
-增量更新、加密、对象流/xref 流、恢复扫描及重复歧义数据均被拒绝，名称转义后
-仍会检查不支持的结构名称。
+增量更新、加密、对象流/xref 流、恢复扫描及重复歧义数据均被拒绝。
+`/Encrypt`、`/ObjStm` 与 `/XRef` 无论出现在哪里都会被拒绝，名称转义后也不例外；
+`/Prev` 与 `/XRefStm` 仅作为 trailer 键时才表示较早的 xref 区段并被拒绝，
+因此仅使用这些名称的普通字典（如大纲项的前一同级引用）仍可读取。
 
 在索引分配前实施边界：输入 16 MiB、xref 条目 10,000、元数据值 100,000、
 嵌套深度 64、每字典 4,096 个键、名称/词元 256 字节、字符串及编码/解码流
diff --git a/native/crates/ops-documents/src/pdf_container.rs b/native/crates/ops-documents/src/pdf_container.rs
index 3936e2c3..d5a2debb 100644
--- a/native/crates/ops-documents/src/pdf_container.rs
+++ b/native/crates/ops-documents/src/pdf_container.rs
@@ -193,6 +193,16 @@ pub fn parse(data: &[u8]) -> Result<Document<'_>> {
     }
     let trailer = lexer.value(0)?;
     let trailer = trailer.dict()?;
+    // Only a trailer key means "this file has an earlier xref section"; the same
+    // names inside an object dictionary are ordinary keys (outline items carry
+    // `/Prev` for their previous sibling).
+    for key in [b"Prev".as_slice(), b"XRefStm"] {
+        if trailer.contains_key(key) {
+            return Err(unsupported(
+                "Encrypted, incremental, object-stream and xref-stream PDFs are unsupported",
+            ));
+        }
+    }
     let size = get(trailer, b"Size")?.integer()?;
     if size == 0
         || size > MAX_OBJECTS
diff --git a/native/crates/ops-documents/src/pdf_container_tests.rs b/native/crates/ops-documents/src/pdf_container_tests.rs
index 706d3a13..81d5f979 100644
--- a/native/crates/ops-documents/src/pdf_container_tests.rs
+++ b/native/crates/ops-documents/src/pdf_container_tests.rs
@@ -75,7 +75,7 @@ fn duplicate_keys_and_escaped_unsupported_names_are_rejected() {
     ] {
         rejected(&fixture(&[value]), "DOCUMENT_PARSE_ERROR");
     }
-    for name in ["Encrypt", "Pr#65v", "XRefStm", "ObjStm", "XRef"] {
+    for name in ["Encrypt", "Enc#72ypt", "ObjStm", "XRef"] {
         rejected(
             &fixture(&[format!("<< /{name} null >>").as_bytes()]),
             "UNSUPPORTED_DOCUMENT",
@@ -83,6 +83,32 @@ fn duplicate_keys_and_escaped_unsupported_names_are_rejected() {
     }
 }
 #[test]
+fn outline_prev_is_an_ordinary_key_while_a_trailer_prev_stays_unsupported() {
+    // An outline item names its previous sibling with /Prev, which is not an
+    // incremental-update marker; refusing the token made the whole document
+    // unreadable (#427).
+    let pdf = fixture(&[
+        b"<< /Type /Catalog /Pages 2 0 R /Outlines 3 0 R >>",
+        b"<< /Type /Pages /Kids [] /Count 0 >>",
+        b"<< /Type /Outlines /First 4 0 R /Last 5 0 R /Count 2 >>",
+        b"<< /Title (First) /Parent 3 0 R /Next 5 0 R >>",
+        b"<< /Title (Second) /Parent 3 0 R /Prev 4 0 R /XRefStm 4 0 R >>",
+    ]);
+    let document = parse(&pdf).unwrap();
+    let second = document.object((5, 0)).unwrap().value.dict().unwrap();
+    assert!(matches!(get(second, b"Prev").unwrap(), Value::Ref(4, 0)));
+    // The same names still refuse a file that really has an earlier xref section.
+    for key in [b"/Prev".as_slice(), b"/Pr#65v", b"/XRefStm"] {
+        let mut trailer = b"/Root 1 0 R ".to_vec();
+        trailer.extend_from_slice(key);
+        trailer.extend_from_slice(b" 9 0 R >>");
+        rejected(
+            &replace(&pdf, b"/Root 1 0 R >>", &trailer),
+            "UNSUPPORTED_DOCUMENT",
+        );
+    }
+}
+#[test]
 fn offset_alias_overlap_and_unindexed_data_are_never_recovered() {
     let pdf = fixture(&[b"<<>>", b"(value)"]);
     let document = parse(&pdf).unwrap();
diff --git a/native/crates/ops-documents/src/pdf_syntax.rs b/native/crates/ops-documents/src/pdf_syntax.rs
index d540cb18..01438da5 100644
--- a/native/crates/ops-documents/src/pdf_syntax.rs
+++ b/native/crates/ops-documents/src/pdf_syntax.rs
@@ -187,6 +187,13 @@ impl<'a> Lexer<'a> {
         }
         Err(bad("Unclosed PDF literal string"))
     }
+    /// A name is a token, not a policy decision.
+    ///
+    /// `/Prev` and `/XRefStm` only describe an incremental update while they are
+    /// trailer keys, and an outline item legitimately carries `/Prev` as its
+    /// previous sibling. Those two are therefore refused where a trailer is read
+    /// (`pdf_container::parse`), not here; `/Encrypt`, `/ObjStm` and `/XRef`
+    /// stay refused at the token because no supported structure can hold them.
     fn name(&mut self) -> Result<Vec<u8>> {
         self.at += 1;
         let mut name = Vec::new();
@@ -213,10 +220,7 @@ impl<'a> Lexer<'a> {
                 return Err(limit("PDF name exceeds 256 bytes"));
             }
         }
-        if matches!(
-            name.as_slice(),
-            b"Encrypt" | b"Prev" | b"XRefStm" | b"ObjStm" | b"XRef"
-        ) {
+        if matches!(name.as_slice(), b"Encrypt" | b"ObjStm" | b"XRef") {
             return Err(unsupported(
                 "Encrypted, incremental, object-stream and xref-stream PDFs are unsupported",
             ));