diff --git a/README.md b/README.md
index cbbe127..630068c 100644
--- a/README.md
+++ b/README.md
@@ -30,7 +30,7 @@ nodes by purpose:
 | Combine | Append rows, inner/outer/semi/anti joins |
 | Filter | Column filter, typed-expression filter, remove duplicates |
 | Transform | Select, derive, ordered calculations, typed expression, conversion, replacement, aggregate, partitioned rank, sort, limit, reusable subflow |
-| Quality | Quality rules, keyed reconciliation |
+| Quality | Quality rules; keyed reconciliation with stable row identity, explicit before/after evidence, and rerun invalidation hashes |
 | Output | Preview output |
 
 Join nodes have explicit left and right ports. Append nodes accept two or more
diff --git a/fixtures/golden/monthly-reconciliation/expected-output.json b/fixtures/golden/monthly-reconciliation/expected-output.json
index 3c1a505..626d744 100644
--- a/fixtures/golden/monthly-reconciliation/expected-output.json
+++ b/fixtures/golden/monthly-reconciliation/expected-output.json
@@ -7,7 +7,23 @@
     "monthly_status": "open",
     "monthly_amount": 10,
     "_reconciliation_status": "match",
-    "_reconciliation_differences": []
+    "_reconciliation_differences": [],
+    "_reconciliation_changes": [],
+    "_reconciliation_key": [
+      "A-1"
+    ],
+    "_reconciliation_key_hash": "04c89a1f5d4765e5c245cbb4bb2964de94b34494bd5af928428042b450d16c09",
+    "_reconciliation_input_hash": "34e8f55e551b289229822bb93bdd30c9ea42832ed67ad67d99050a3ef9fbdb18",
+    "_reconciliation_before": {
+      "case_id": "A-1",
+      "status": "open",
+      "amount": 10
+    },
+    "_reconciliation_after": {
+      "case_id": "A-1",
+      "status": "open",
+      "amount": 10
+    }
   },
   {
     "case_id": "A-2",
@@ -17,13 +33,50 @@
     "monthly_status": "closed",
     "monthly_amount": 25,
     "_reconciliation_status": "changed",
-    "_reconciliation_differences": ["amount"]
+    "_reconciliation_differences": [
+      "amount"
+    ],
+    "_reconciliation_changes": [
+      {
+        "expected_field": "amount",
+        "observed_field": "amount",
+        "expected": 20,
+        "observed": 25
+      }
+    ],
+    "_reconciliation_key": [
+      "A-2"
+    ],
+    "_reconciliation_key_hash": "20f1552752978558506b159b92bc738d34f484d3d76068173b0c96e1fc4a8c2d",
+    "_reconciliation_input_hash": "182de78bc3297d90cd4b02392bfb4a53fc9d9c29851d567f68e10be477a57dd8",
+    "_reconciliation_before": {
+      "case_id": "A-2",
+      "status": "closed",
+      "amount": 20
+    },
+    "_reconciliation_after": {
+      "case_id": "A-2",
+      "status": "closed",
+      "amount": 25
+    }
   },
   {
     "monthly_case_id": "A-3",
     "monthly_status": "open",
     "monthly_amount": 30,
     "_reconciliation_status": "missing_expected",
-    "_reconciliation_differences": []
+    "_reconciliation_differences": [],
+    "_reconciliation_changes": [],
+    "_reconciliation_key": [
+      "A-3"
+    ],
+    "_reconciliation_key_hash": "fedd5f48e5721d007a161015a4bab211555795d54a9492afa1b018050d3767f2",
+    "_reconciliation_input_hash": "fa82f93e97badd67750af911c545d703bb263a336bf8ba6d995eea5377abc2a0",
+    "_reconciliation_before": null,
+    "_reconciliation_after": {
+      "case_id": "A-3",
+      "status": "open",
+      "amount": 30
+    }
   }
 ]
diff --git a/src/govoplan_dataflow/backend/executor.py b/src/govoplan_dataflow/backend/executor.py
index dc16008..ab9167c 100644
--- a/src/govoplan_dataflow/backend/executor.py
+++ b/src/govoplan_dataflow/backend/executor.py
@@ -878,12 +878,32 @@ def _reconcile_rows(
     right_keys = [str(item) for item in config["right_keys"]]
     right_prefix = str(config.get("right_prefix", "observed_"))
     comparisons = _comparison_fields(config.get("compare_columns"))
-    left_index = _unique_row_index(left_rows, left_keys, node_id=node_id)
-    right_index = _unique_row_index(right_rows, right_keys, node_id=node_id)
+    left_index = _unique_row_index(
+        left_rows,
+        left_keys,
+        node_id=node_id,
+        input_label="expected",
+    )
+    right_index = _unique_row_index(
+        right_rows,
+        right_keys,
+        node_id=node_id,
+        input_label="observed",
+    )
     output: list[dict[str, Any]] = []
     for key in dict.fromkeys((*left_index, *right_index)):
         left = left_index.get(key)
         right = right_index.get(key)
+        key_values = [
+            (left or {}).get(left_name)
+            if left is not None
+            else (right or {}).get(right_name)
+            for left_name, right_name in zip(
+                left_keys,
+                right_keys,
+                strict=True,
+            )
+        ]
         item = dict(left or {})
         if right is not None:
             item.update({f"{right_prefix}{name}": value for name, value in right.items()})
@@ -895,18 +915,55 @@ def _reconcile_rows(
             differences = []
         else:
             fields = comparisons or tuple((name, name) for name in left if name not in left_keys)
-            differences = [
-                left_name
+            changes = [
+                {
+                    "expected_field": left_name,
+                    "observed_field": right_name,
+                    "expected": left.get(left_name),
+                    "observed": right.get(right_name),
+                }
                 for left_name, right_name in fields
                 if left.get(left_name) != right.get(right_name)
             ]
+            differences = [str(change["expected_field"]) for change in changes]
             status = "changed" if differences else "match"
+        if left is None or right is None:
+            changes = []
         item["_reconciliation_status"] = status
         item["_reconciliation_differences"] = differences
+        item["_reconciliation_changes"] = changes
+        item["_reconciliation_key"] = key_values
+        item["_reconciliation_key_hash"] = _reconciliation_hash(
+            {
+                "left_keys": left_keys,
+                "right_keys": right_keys,
+                "values": key_values,
+            }
+        )
+        item["_reconciliation_input_hash"] = _reconciliation_hash(
+            {
+                "key": key_values,
+                "expected": left,
+                "observed": right,
+                "comparisons": comparisons,
+            }
+        )
+        item["_reconciliation_before"] = dict(left) if left is not None else None
+        item["_reconciliation_after"] = dict(right) if right is not None else None
         output.append(item)
     return output
 
 
+def _reconciliation_hash(value: object) -> str:
+    encoded = json.dumps(
+        value,
+        sort_keys=True,
+        separators=(",", ":"),
+        default=str,
+    ).encode("utf-8")
+    return hashlib.sha256(encoded).hexdigest()
+
+
 def _comparison_fields(value: object) -> tuple[tuple[str, str], ...]:
     if not isinstance(value, list):
         return ()
@@ -927,13 +984,14 @@ def _unique_row_index(
     columns: list[str],
     *,
     node_id: str,
+    input_label: str,
 ) -> dict[tuple[Any, ...], dict[str, Any]]:
     index: dict[tuple[Any, ...], dict[str, Any]] = {}
     for row in rows:
         key = tuple(_hashable(row.get(column)) for column in columns)
         if key in index:
             raise PipelineExecutionError(
-                f"Reconciliation keys are not unique: {key!r}.",
+                f"Reconciliation keys are not unique in the {input_label} input.",
                 node_id=node_id,
             )
         index[key] = row
diff --git a/src/govoplan_dataflow/backend/manifest.py b/src/govoplan_dataflow/backend/manifest.py
index c4bcda8..ae6a346 100644
--- a/src/govoplan_dataflow/backend/manifest.py
+++ b/src/govoplan_dataflow/backend/manifest.py
@@ -200,6 +200,7 @@ DOCUMENTATION = (
             "Every graph node declares typed inputs, configuration, output schema, and validation rules. "
             "Source nodes pin inline content or governed Datasource references; combine, filter, transform, "
             "quality, reconciliation, reusable-subflow, and output nodes remain explicit in the canonical graph. "
+            "Reconciliation rows expose stable key hashes, explicit before/after values, and input hashes so a later governed human decision can be replayed only while its inputs still match. "
             "Expressions use the typed Dataflow expression language and never execute arbitrary host or database "
             "code. Selecting a node may request a bounded intermediate preview; preview rows are transient, "
             "privacy-filtered for the actor, and are not retained as run output. SQL editing compiles into the same "
diff --git a/src/govoplan_dataflow/backend/node_library.py b/src/govoplan_dataflow/backend/node_library.py
index 9c4148b..2421924 100644
--- a/src/govoplan_dataflow/backend/node_library.py
+++ b/src/govoplan_dataflow/backend/node_library.py
@@ -510,7 +510,7 @@ _NODE_TYPES = (
         type="reconcile.compare",
         category="quality",
         label="Reconcile tables",
-        description="Compare keyed rows and expose missing, matching, and changed records.",
+        description="Compare keyed rows with stable identity, before/after evidence, and decision invalidation fingerprints.",
         icon="scan-search",
         input_ports=(
             NodePortDefinition(id="left", label="Expected"),
diff --git a/src/govoplan_dataflow/backend/schema_validation.py b/src/govoplan_dataflow/backend/schema_validation.py
index f77836e..968f7ca 100644
--- a/src/govoplan_dataflow/backend/schema_validation.py
+++ b/src/govoplan_dataflow/backend/schema_validation.py
@@ -612,6 +612,12 @@ def _reconcile(context: SchemaPropagationContext) -> SchemaPropagationResult:
             (
                 "_reconciliation_status",
                 "_reconciliation_differences",
+                "_reconciliation_changes",
+                "_reconciliation_key",
+                "_reconciliation_key_hash",
+                "_reconciliation_input_hash",
+                "_reconciliation_before",
+                "_reconciliation_after",
             )
         ),
         open=left_state.open or right_state.open,
@@ -623,6 +629,12 @@ def _reconcile(context: SchemaPropagationContext) -> SchemaPropagationResult:
             },
             "_reconciliation_status": "string",
             "_reconciliation_differences": "array",
+            "_reconciliation_changes": "array",
+            "_reconciliation_key": "array",
+            "_reconciliation_key_hash": "string",
+            "_reconciliation_input_hash": "string",
+            "_reconciliation_before": "object",
+            "_reconciliation_after": "object",
         },
     )
     return SchemaPropagationResult(state, tuple(diagnostics))
diff --git a/tests/test_operators.py b/tests/test_operators.py
index b15b3af..3ba2fca 100644
--- a/tests/test_operators.py
+++ b/tests/test_operators.py
@@ -2,7 +2,7 @@ from __future__ import annotations
 
 import unittest
 
-from govoplan_dataflow.backend.executor import execute_preview
+from govoplan_dataflow.backend.executor import PipelineExecutionError, execute_preview
 from govoplan_dataflow.backend.expressions import (
     ExpressionError,
     evaluate_expression,
@@ -389,6 +389,128 @@ class DataflowOperatorTests(unittest.TestCase):
             ["changed", "missing_observed", "missing_expected"],
             [item["_reconciliation_status"] for item in result.rows],
         )
+        changed = result.rows[0]
+        self.assertEqual(["1"], changed["_reconciliation_key"])
+        self.assertEqual(
+            {"id": "1", "amount": 10},
+            changed["_reconciliation_before"],
+        )
+        self.assertEqual(
+            {"id": "1", "amount": 11},
+            changed["_reconciliation_after"],
+        )
+        self.assertEqual(
+            [
+                {
+                    "expected_field": "amount",
+                    "observed_field": "amount",
+                    "expected": 10,
+                    "observed": 11,
+                }
+            ],
+            changed["_reconciliation_changes"],
+        )
+        replay = execute_preview(graph, row_limit=100).rows[0]
+        self.assertEqual(
+            changed["_reconciliation_key_hash"],
+            replay["_reconciliation_key_hash"],
+        )
+        self.assertEqual(
+            changed["_reconciliation_input_hash"],
+            replay["_reconciliation_input_hash"],
+        )
+
+        changed_nodes = [
+            (
+                graph_node.model_copy(
+                    update={
+                        "config": {
+                            **graph_node.config,
+                            "rows": [
+                                {"id": "1", "amount": 12},
+                                {"id": "3", "amount": 30},
+                            ],
+                        }
+                    }
+                )
+                if graph_node.id == "observed"
+                else graph_node
+            )
+            for graph_node in graph.nodes
+        ]
+        rerun = execute_preview(
+            graph.model_copy(update={"nodes": changed_nodes}),
+            row_limit=100,
+        ).rows[0]
+        self.assertEqual(
+            changed["_reconciliation_key_hash"],
+            rerun["_reconciliation_key_hash"],
+        )
+        self.assertNotEqual(
+            changed["_reconciliation_input_hash"],
+            rerun["_reconciliation_input_hash"],
+        )
+        self.assertIsNone(result.rows[1]["_reconciliation_after"])
+        self.assertIsNone(result.rows[2]["_reconciliation_before"])
+
+    def test_reconciliation_rejects_ambiguous_keys_without_echoing_values(self) -> None:
+        graph = PipelineGraph(
+            nodes=[
+                node(
+                    "expected",
+                    "source.inline",
+                    {
+                        "source_name": "expected",
+                        "rows": [
+                            {"id": "private-key", "amount": 10},
+                            {"id": "private-key", "amount": 11},
+                        ],
+                    },
+                    x=0,
+                ),
+                node(
+                    "observed",
+                    "source.inline",
+                    {"source_name": "observed", "rows": []},
+                    x=0,
+                ),
+                node(
+                    "reconcile",
+                    "reconcile.compare",
+                    {
+                        "left_keys": ["id"],
+                        "right_keys": ["id"],
+                        "compare_columns": ["amount"],
+                        "right_prefix": "observed_",
+                    },
+                    x=300,
+                ),
+                node("output", "output", {}, x=600),
+            ],
+            edges=[
+                GraphEdge(
+                    id="e1",
+                    source="expected",
+                    target="reconcile",
+                    target_port="left",
+                ),
+                GraphEdge(
+                    id="e2",
+                    source="observed",
+                    target="reconcile",
+                    target_port="right",
+                ),
+                GraphEdge(id="e3", source="reconcile", target="output"),
+            ],
+        )
+
+        with self.assertRaisesRegex(
+            PipelineExecutionError,
+            "not unique in the expected input",
+        ) as raised:
+            execute_preview(graph, row_limit=100)
+
+        self.assertNotIn("private-key", str(raised.exception))
 
     def test_parameterized_subflow_runs_a_pinned_graph(self) -> None:
         nested = {