diff --git a/custom_nodes/research/__init__.py b/custom_nodes/research/__init__.py
new file mode 100644
index 000000000..24daf8e20
--- /dev/null
+++ b/custom_nodes/research/__init__.py
@@ -0,0 +1,17 @@
+"""Research Workbench custom nodes for ComfyUI."""
+from typing import List
+from typing_extensions import override
+from comfy_api.latest import ComfyExtension, io
+
+
+class ResearchExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> List[type[io.ComfyNode]]:
+        from custom_nodes.research.paper_search import PaperSearch
+        from custom_nodes.research.claim_extract import PaperClaimExtract
+        from custom_nodes.research.evidence_assemble import ClaimEvidenceAssemble
+        return [PaperSearch, PaperClaimExtract, ClaimEvidenceAssemble]
+
+
+async def comfy_entrypoint() -> ComfyExtension:
+    return ResearchExtension()
diff --git a/custom_nodes/research/claim_extract.py b/custom_nodes/research/claim_extract.py
new file mode 100644
index 000000000..5b56cb3e6
--- /dev/null
+++ b/custom_nodes/research/claim_extract.py
@@ -0,0 +1,68 @@
+"""PaperClaimExtract node - extract claims from paper text."""
+import json
+import re
+from typing_extensions import override
+from comfy_api.latest import ComfyNode, io
+
+
+class PaperClaimExtract(io.ComfyNode):
+    """Extract scientific claims from paper text or abstract."""
+
+    @classmethod
+    def define_schema(cls) -> io.Schema:
+        return io.Schema(
+            node_id="PaperClaimExtract",
+            display_name="Extract Claims",
+            category="Research",
+            inputs=[
+                io.String.Input(
+                    "paper_text",
+                    display_name="Paper Text / Abstract",
+                    default="",
+                    multiline=True,
+                ),
+                io.String.Input(
+                    "claim_types",
+                    display_name="Claim Types (comma-separated)",
+                    default="performance,robustness,generalization",
+                ),
+            ],
+            outputs=[
+                io.String.Output(display_name="Claims (JSON)"),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, paper_text: str, claim_types: str) -> io.NodeOutput:
+        if not paper_text.strip():
+            return io.NodeOutput(claims=json.dumps([]))
+
+        claim_type_list = [t.strip() for t in claim_types.split(",")]
+        sentences = re.split(r'(?<=[.!?])\s+', paper_text.strip())
+
+        claims = []
+        for sent in sentences:
+            sent = sent.strip()
+            if not sent or len(sent) < 20:
+                continue
+            lower = sent.lower()
+            # Heuristic: sentences with comparison/performance words
+            if any(w in lower for w in ["achieves", "outperforms", "improves", "reduces", "increases", "demonstrates", "shows", "provides", "enables"]):
+                claims.append({
+                    "text": sent,
+                    "type": _classify_claim(sent, claim_type_list),
+                    "support_level": "unsupported",
+                })
+
+        return io.NodeOutput(claims=json.dumps(claims[:10], indent=2))  # limit to 10
+
+
+def _classify_claim(sentence: str, claim_types) -> str:
+    lower = sentence.lower()
+    if any(w in lower for w in ["accuracy", "performance", "score", "auc", "f1", "precision", "recall"]):
+        return "performance"
+    if any(w in lower for w in ["robust", "noise", "attack", "adversarial", "corruption"]):
+        return "robustness"
+    if any(w in lower for w in ["general", "transfer", "cross-domain", "cross-dataset"]):
+        return "generalization"
+    return claim_types[0] if claim_types else "other"
diff --git a/custom_nodes/research/evidence_assemble.py b/custom_nodes/research/evidence_assemble.py
new file mode 100644
index 000000000..0cb9cc7ea
--- /dev/null
+++ b/custom_nodes/research/evidence_assemble.py
@@ -0,0 +1,54 @@
+"""ClaimEvidenceAssemble node - assemble claim × evidence matrix."""
+import json
+from typing_extensions import override
+from comfy_api.latest import ComfyNode, io
+
+
+class ClaimEvidenceAssemble(io.ComfyNode):
+    """Assemble claims and papers into a structured evidence matrix."""
+
+    @classmethod
+    def define_schema(cls) -> io.Schema:
+        return io.Schema(
+            node_id="ClaimEvidenceAssemble",
+            display_name="Assemble Evidence",
+            category="Research",
+            inputs=[
+                io.String.Input(
+                    "claims",
+                    display_name="Claims (JSON)",
+                    default="[]",
+                    multiline=True,
+                ),
+                io.String.Input(
+                    "papers",
+                    display_name="Papers (JSON)",
+                    default="[]",
+                    multiline=True,
+                ),
+            ],
+            outputs=[
+                io.String.Output(display_name="Evidence Matrix (JSON)"),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, claims: str, papers: str) -> io.NodeOutput:
+        try:
+            claims_list = json.loads(claims) if claims else []
+            papers_list = json.loads(papers) if papers else []
+
+            matrix = []
+            for claim in claims_list:
+                claim_text = claim.get("text", "") if isinstance(claim, dict) else str(claim)
+                matrix.append({
+                    "claim": claim_text,
+                    "claim_type": claim.get("type", "") if isinstance(claim, dict) else "",
+                    "support_level": claim.get("support_level", "unsupported") if isinstance(claim, dict) else "unsupported",
+                    "evidence": [],
+                    "gap_flags": ["No linked evidence yet"],
+                })
+
+            return io.NodeOutput(evidence_matrix=json.dumps(matrix, indent=2))
+        except json.JSONDecodeError as e:
+            return io.NodeOutput(evidence_matrix=json.dumps({"error": f"JSON parse error: {e}"}))
diff --git a/custom_nodes/research/paper_search.py b/custom_nodes/research/paper_search.py
new file mode 100644
index 000000000..aa78e7198
--- /dev/null
+++ b/custom_nodes/research/paper_search.py
@@ -0,0 +1,62 @@
+"""PaperSearch node - search papers via academic APIs."""
+import json
+import urllib.request
+import urllib.parse
+from typing_extensions import override
+from comfy_api.latest import ComfyNode, io
+
+
+class PaperSearch(io.ComfyNode):
+    """Search academic papers from Semantic Scholar."""
+
+    @classmethod
+    def define_schema(cls) -> io.Schema:
+        return io.Schema(
+            node_id="PaperSearch",
+            display_name="Paper Search",
+            category="Research",
+            inputs=[
+                io.String.Input(
+                    "query",
+                    display_name="Search Query",
+                    default="",
+                ),
+                io.Int.Input(
+                    "max_results",
+                    display_name="Max Results",
+                    default=5,
+                    min=1,
+                    max=20,
+                    step=1,
+                ),
+            ],
+            outputs=[
+                io.String.Output(display_name="Papers (JSON)"),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, query: str, max_results: int) -> io.NodeOutput:
+        if not query.strip():
+            return io.NodeOutput(papers=json.dumps([]))
+
+        encoded_query = urllib.parse.quote(query)
+        url = f"https://api.semanticscholar.org/graph/v1/paper/search?query={encoded_query}&limit={max_results}&fields=title,authors,abstract,year,journal,venue"
+
+        try:
+            req = urllib.request.Request(url, headers={"Accept": "application/json"})
+            with urllib.request.urlopen(req, timeout=15) as response:
+                data = json.loads(response.read().decode())
+                papers = []
+                for item in data.get("data", []):
+                    papers.append({
+                        "title": item.get("title", ""),
+                        "authors": [a.get("name", "") for a in item.get("authors", [])],
+                        "abstract": item.get("abstract", ""),
+                        "year": item.get("year", ""),
+                        "venue": item.get("venue", ""),
+                        "paper_id": item.get("paperId", ""),
+                    })
+                return io.NodeOutput(papers=json.dumps(papers, indent=2))
+        except Exception as e:
+            return io.NodeOutput(papers=json.dumps({"error": str(e)}))