diff --git a/netx_api/biz_state/compare_service.py b/netx_api/biz_state/compare_service.py
index 5c30368..983f7fb 100644
--- a/netx_api/biz_state/compare_service.py
+++ b/netx_api/biz_state/compare_service.py
@@ -1348,69 +1348,88 @@ def _port_map_dict(db: Session, mapping_id: str) -> dict[str, str]:
 
 
 def _load_metric_rows(db: Session, *, batch_id: str, metric_id: str) -> list[dict[str, Any]]:
-    """Load metric rows with streaming fetch to avoid holding the full ORM set."""
+    """Load metric rows in LIMIT/OFFSET chunks (avoid ORM yield_per + unique() clash)."""
     if metric_id == "lldp_neighbor":
-        q = (
-            db.query(BizStateLldpNeighbor)
-            .filter(BizStateLldpNeighbor.batch_id == batch_id)
-            .execution_options(stream_results=True, yield_per=_LOAD_YIELD_PER)
-        )
         out: list[dict[str, Any]] = []
-        for n in q:
-            out.append(
-                {
-                    "local_if": n.local_if,
-                    "remote_sys": n.remote_sys,
-                    "remote_if": n.remote_if,
-                    "remote_ip": n.remote_ip,
-                    "protocol": n.protocol,
-                    "_netx": {
-                        "batch_id": batch_id,
-                        "batch_command_id": n.batch_command_id or "",
-                        "task_id": n.task_id or "",
-                        "ne_id": n.ne_id or "",
-                        "collected_at": n.collected_at.isoformat() + "Z"
-                        if n.collected_at
-                        else None,
-                        "row_id": n.id,
-                    },
-                }
+        offset = 0
+        while True:
+            chunk = (
+                db.query(BizStateLldpNeighbor)
+                .filter(BizStateLldpNeighbor.batch_id == batch_id)
+                .order_by(BizStateLldpNeighbor.id.asc())
+                .offset(offset)
+                .limit(_LOAD_YIELD_PER)
+                .all()
             )
-            db.expunge(n)
+            if not chunk:
+                break
+            for n in chunk:
+                out.append(
+                    {
+                        "local_if": n.local_if,
+                        "remote_sys": n.remote_sys,
+                        "remote_if": n.remote_if,
+                        "remote_ip": n.remote_ip,
+                        "protocol": n.protocol,
+                        "_netx": {
+                            "batch_id": batch_id,
+                            "batch_command_id": n.batch_command_id or "",
+                            "task_id": n.task_id or "",
+                            "ne_id": n.ne_id or "",
+                            "collected_at": n.collected_at.isoformat() + "Z"
+                            if n.collected_at
+                            else None,
+                            "row_id": n.id,
+                        },
+                    }
+                )
+                db.expunge(n)
+            offset += len(chunk)
+            if len(chunk) < _LOAD_YIELD_PER:
+                break
         return out
     # Generic tabular metrics (ISIS / interface / ARP / ND6 / BGP …)
     from ..models import BizStateMetricRow
 
-    q = (
-        db.query(BizStateMetricRow)
-        .filter(
-            BizStateMetricRow.batch_id == batch_id,
-            BizStateMetricRow.metric_id == metric_id,
-        )
-        .order_by(BizStateMetricRow.seq.asc(), BizStateMetricRow.id.asc())
-        .execution_options(stream_results=True, yield_per=_LOAD_YIELD_PER)
-    )
     out = []
-    for r in q:
-        # Raw rows only — filtering belongs to the compare sheet template
-        # (``row_filters``), not metric-specific branches here.
-        # ``_netx`` is collector provenance (stripped before field compare).
-        out.append(
-            {
-                **dict(r.data_json or {}),
-                "_netx": {
-                    "batch_id": batch_id,
-                    "batch_command_id": r.batch_command_id or "",
-                    "task_id": r.task_id or "",
-                    "ne_id": r.ne_id or "",
-                    "collected_at": r.collected_at.isoformat() + "Z"
-                    if r.collected_at
-                    else None,
-                    "row_id": r.id,
-                },
-            }
+    offset = 0
+    while True:
+        chunk = (
+            db.query(BizStateMetricRow)
+            .filter(
+                BizStateMetricRow.batch_id == batch_id,
+                BizStateMetricRow.metric_id == metric_id,
+            )
+            .order_by(BizStateMetricRow.seq.asc(), BizStateMetricRow.id.asc())
+            .offset(offset)
+            .limit(_LOAD_YIELD_PER)
+            .all()
         )
-        db.expunge(r)
+        if not chunk:
+            break
+        for r in chunk:
+            # Raw rows only — filtering belongs to the compare sheet template
+            # (``row_filters``), not metric-specific branches here.
+            # ``_netx`` is collector provenance (stripped before field compare).
+            out.append(
+                {
+                    **dict(r.data_json or {}),
+                    "_netx": {
+                        "batch_id": batch_id,
+                        "batch_command_id": r.batch_command_id or "",
+                        "task_id": r.task_id or "",
+                        "ne_id": r.ne_id or "",
+                        "collected_at": r.collected_at.isoformat() + "Z"
+                        if r.collected_at
+                        else None,
+                        "row_id": r.id,
+                    },
+                }
+            )
+            db.expunge(r)
+        offset += len(chunk)
+        if len(chunk) < _LOAD_YIELD_PER:
+            break
     if out:
         return out
     # Known metric with zero rows is OK; unknown metric still errors