diff --git a/verifiers/v1/graph.py b/verifiers/v1/graph.py
index f81480d43b..f039e80cf6 100644
--- a/verifiers/v1/graph.py
+++ b/verifiers/v1/graph.py
@@ -101,6 +101,11 @@ class MessageNode(BaseModel):
     logprobs: list[float] = Field(default_factory=list)
     """Sampling logprobs for the sampled tokens — length equals the number of True entries in
     `mask`; empty for input messages."""
+    advantages: list[float] | None = None
+    """Per-token credit over the sampled tokens, same layout as `logprobs`. `None` until a
+    consumer's RL algorithm assigns it, which is not the same as a credit of zero: a group whose
+    rewards were all equal is assigned zeros and carries no gradient, while an unassigned node was
+    never scored at all."""
     multi_modal_data: SkipJsonSchema[MultiModalData | None] = None
     """The renderer items for the images this message's content introduces (pixel tensors,
     grids, hashes, placeholders) — the only carrier of the pixels from the env server to the
diff --git a/verifiers/v1/trace.py b/verifiers/v1/trace.py
index 2e36ced9d1..52e479bcbd 100644
--- a/verifiers/v1/trace.py
+++ b/verifiers/v1/trace.py
@@ -3,7 +3,7 @@
 import time
 import traceback
 import uuid
-from collections.abc import Mapping
+from collections.abc import Callable, Mapping
 from typing import TYPE_CHECKING, Annotated, Any, Generic, Literal
 
 import numpy as np
@@ -200,28 +200,45 @@ def sampled_mask(self) -> list[bool]:
             mask.extend(node.mask)
         return mask
 
-    @property
-    def logprobs(self) -> list[float]:
-        """Per-token sampling logprobs aligned to `token_ids` — the node logprobs spread onto
-        their sampled positions, 0.0 on every non-sampled token."""
+    def spread(
+        self, values: Callable[[MessageNode], list[float] | None]
+    ) -> list[float]:
+        """A per-sampled-token node field widened to `token_ids`: each node's values land on its
+        sampled positions, 0.0 everywhere else. A node holding nothing contributes zeros."""
         out: list[float] = []
         for node in self.nodes:
-            mask = node.mask
-            sampled = sum(mask) if node.logprobs else 0
+            mask, node_values = node.mask, values(node) or []
+            sampled = sum(mask) if node_values else 0
             # Bulk-fill the canonical unsampled-prefix/sampled-suffix layout.
             if not sampled or all(mask[-sampled:]):
-                out += [0.0] * (len(mask) - sampled) + node.logprobs[:sampled]
-                out += [0.0] * max(0, sampled - len(node.logprobs))
+                out += [0.0] * (len(mask) - sampled) + node_values[:sampled]
+                out += [0.0] * max(0, sampled - len(node_values))
                 continue
             li = 0
-            for sampled in mask:
-                if sampled:
-                    out.append(node.logprobs[li] if li < len(node.logprobs) else 0.0)
+            for is_sampled in mask:
+                if is_sampled:
+                    out.append(node_values[li] if li < len(node_values) else 0.0)
                     li += 1
                 else:
                     out.append(0.0)
         return out
 
+    @property
+    def logprobs(self) -> list[float]:
+        """Per-token sampling logprobs aligned to `token_ids` — the node logprobs spread onto
+        their sampled positions, 0.0 on every non-sampled token."""
+        return self.spread(lambda node: node.logprobs)
+
+    @property
+    def advantages(self) -> list[float] | None:
+        """Per-token credit aligned to `token_ids`, spread like `logprobs` — or `None` when no node
+        on the path was ever assigned any, which a branch of zeros would otherwise be
+        indistinguishable from. A partially assigned path spreads, the unassigned nodes reading
+        0.0."""
+        if all(node.advantages is None for node in self.nodes):
+            return None
+        return self.spread(lambda node: node.advantages)
+
     @property
     def multi_modal_data(self) -> MultiModalData | None:
         """Node image data concatenated in token order for training; never persisted."""