diff --git a/src/reviewer/cli.py b/src/reviewer/cli.py
index f4b49ad..d90f4ac 100644
--- a/src/reviewer/cli.py
+++ b/src/reviewer/cli.py
@@ -188,6 +188,7 @@ def _build_paper_json(
             "explanation": c.explanation,
             "comment_type": c.comment_type,
             "paragraph_index": c.paragraph_index,
+            "severity": c.severity,
         })
 
     model_short = _model_short_name(result.model) if result.model else ""
diff --git a/src/reviewer/models.py b/src/reviewer/models.py
index 671bbc0..b00e9bb 100644
--- a/src/reviewer/models.py
+++ b/src/reviewer/models.py
@@ -2,6 +2,8 @@
 
 from dataclasses import dataclass, field
 
+SEVERITY_VALUES = ("minor", "moderate", "major")
+
 
 @dataclass
 class Comment:
@@ -11,6 +13,7 @@ class Comment:
     explanation: str    # reviewer's explanation
     comment_type: str   # "technical" or "logical"
     paragraph_index: int | None = None  # 0-based index in split paragraphs
+    severity: str | None = None         # one of SEVERITY_VALUES, or None when unknown
 
     def to_dict(self) -> dict:
         d = {
@@ -21,6 +24,8 @@ def to_dict(self) -> dict:
         }
         if self.paragraph_index is not None:
             d["paragraph_index"] = self.paragraph_index
+        if self.severity:
+            d["severity"] = self.severity
         return d
 
 
diff --git a/src/reviewer/prompts.py b/src/reviewer/prompts.py
index ea7d7d7..d46425d 100644
--- a/src/reviewer/prompts.py
+++ b/src/reviewer/prompts.py
@@ -48,6 +48,12 @@
 DO_NOT_FLAG_CHUNKED = DO_NOT_FLAG_BASE.rstrip() + """
 - Incomplete text at passage boundaries"""
 
+SEVERITY_RUBRIC = """\
+For each issue, assign a "severity" tier:
+- "major": undermines a key claim, methodology, or comparison; affects conclusions
+- "moderate": real error or gap, but localized and fixable
+- "minor": framing concern, mild overclaim, or ambiguity resolvable from context"""
+
 DO_NOT_FLAG_PROGRESSIVE = DO_NOT_FLAG_CHUNKED.rstrip() + """
 - Notation not yet in the summary — it may be introduced later"""
 
@@ -57,6 +63,7 @@
 - "quote": the exact verbatim text (preserving LaTeX)
 - "explanation": deep reasoning — what you initially thought, whether context resolves it, and what specifically remains problematic
 - "type": "technical" or "logical"
+- "severity": "minor", "moderate", or "major"
 """
 
 JSON_OBJECT_OUTPUT = """\
@@ -99,6 +106,8 @@
 
 {DO_NOT_FLAG_CHUNKED}
 
+{SEVERITY_RUBRIC}
+
 {JSON_ARRAY_OUTPUT}"""
 
 DEEP_CHECK_PROGRESSIVE_PROMPT = f"""{REVIEWER_PREAMBLE}
@@ -149,6 +158,8 @@
 
 {DO_NOT_FLAG_BASE}
 
+{SEVERITY_RUBRIC}
+
 Return a JSON object with this structure:
 {{{{
   "overall_feedback": "One paragraph high-level assessment of the paper's quality and main issues",
@@ -157,7 +168,8 @@
       "title": "short descriptive title of the issue",
       "quote": "the exact verbatim text from the paper containing the issue (copy it exactly, preserving LaTeX)",
       "explanation": "deep reasoning — what you initially thought, whether context resolves it, and what specifically remains problematic",
-      "type": "technical" or "logical"
+      "type": "technical" or "logical",
+      "severity": "minor" or "moderate" or "major"
     }}}}
   ]
 }}}}
@@ -194,6 +206,8 @@
 
 {DO_NOT_FLAG_CHUNKED}
 
+{SEVERITY_RUBRIC}
+
 Return a JSON object:
 {{{{
   "overall_feedback": "brief assessment of this section",
@@ -202,7 +216,8 @@
       "title": "short descriptive title of the issue",
       "quote": "the exact verbatim text from the paper containing the issue (copy it exactly, preserving LaTeX)",
       "explanation": "deep reasoning — what you initially thought, whether context resolves it, and what specifically remains problematic",
-      "type": "technical" or "logical"
+      "type": "technical" or "logical",
+      "severity": "minor" or "moderate" or "major"
     }}}}
   ]
 }}}}
diff --git a/src/reviewer/utils.py b/src/reviewer/utils.py
index ec62cce..347e926 100644
--- a/src/reviewer/utils.py
+++ b/src/reviewer/utils.py
@@ -6,7 +6,7 @@
 
 import tiktoken
 
-from .models import Comment
+from .models import Comment, SEVERITY_VALUES
 
 
 def _get_encoding():
@@ -205,12 +205,18 @@ def parse_comments_from_list(items: list[dict]) -> list[Comment]:
         paragraph_index = item.get("paragraph_index", None)
         if paragraph_index is not None:
             paragraph_index = int(paragraph_index)
+        severity = item.get("severity")
+        if isinstance(severity, str):
+            severity = severity.strip().lower()
+        if severity not in SEVERITY_VALUES:
+            severity = "moderate"
         comments.append(Comment(
             title=title,
             quote=quote,
             explanation=explanation,
             comment_type=comment_type,
             paragraph_index=paragraph_index,
+            severity=severity,
         ))
     return comments