feat(OMN-12130): add node_handler_correctness_gate — golden eval set + correctness gate

docs/evidence/dep-health-sweep/proof-of-life-001/baseline_diff.json

@@ -1 +1 @@
-null
+null

docs/evidence/dep-health-sweep/proof-of-life-001/findings.json

@@ -19,4 +19,4 @@
     "rule_id": "MISSING_TOPIC_EDGE",
     "rule_version": "v1"
   }
-]
+]

docs/evidence/dep-health-sweep/proof-of-life-001/import_graph.json

@@ -3,4 +3,4 @@
   "nodes": [],
   "edges": [],
   "orphan_modules": []
-}
+}

docs/evidence/dep-health-sweep/proof-of-life-001/projection_rows.json

@@ -23,4 +23,4 @@
     "rule_version": "v1",
     "captured_at": "2026-05-25T19:17:15.256588+00:00"
   }
-]
+]

docs/evidence/dep-health-sweep/proof-of-life-001/proof_summary.md

@@ -11,4 +11,4 @@ timestamp: 2026-05-25T19:17:15.259389+00:00
 - **D: projection**: PASS — rows_upserted=2
 
 ## Findings count: 2
-## Projection rows: 2
+## Projection rows: 2

docs/evidence/dep-health-sweep/proof-of-life-001/run_manifest.json

@@ -6,4 +6,4 @@
   ],
   "graphify_version": "0.8.2",
   "rule_version": "v1"
-}
+}

docs/evidence/dep-health-sweep/proof-of-life-001/topology_graph.json

@@ -3,4 +3,4 @@
   "pub_edges": [],
   "sub_edges": [],
   "orphan_topics": []
-}
+}

pyproject.toml

@@ -309,6 +309,7 @@ node_llm_delegation_projection = "omnimarket.nodes.node_llm_delegation_projectio
 node_swarm_registry_compute = "omnimarket.nodes.node_swarm_registry_compute"
 node_knowledge_query_federation_orchestrator = "omnimarket.nodes.node_knowledge_query_federation_orchestrator"
 node_antipattern_index_effect = "omnimarket.nodes.node_antipattern_index_effect"
+node_handler_correctness_gate = "omnimarket.nodes.node_handler_correctness_gate"
 
 [project.scripts]
 onex-integration-test-runner = "omnimarket.nodes.node_integration_test_runner.cli:main"

src/omnimarket/nodes/node_handler_correctness_gate/__init__.py

@@ -0,0 +1,2 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT

src/omnimarket/nodes/node_handler_correctness_gate/contract.yaml

@@ -0,0 +1,71 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT
+name: node_handler_correctness_gate
+contract_name: node_handler_correctness_gate
+node_name: node_handler_correctness_gate
+contract_version: {major: 1, minor: 0, patch: 0}
+node_version: {major: 1, minor: 0, patch: 0}
+node_type: compute
+description: >
+  Pure compute node that scores pre-computed handler outputs against an eval
+  set and returns pass/fail with a correctness score. The gate itself never
+  invokes the handler — it only compares outputs.
+
+input_model:
+  name: ModelCorrectnessCheckRequest
+  class: ModelCorrectnessCheckRequest
+  module: omnimarket.nodes.node_handler_correctness_gate.models.model_correctness_check_request
+
+output_model:
+  name: ModelCorrectnessCheckResult
+  class: ModelCorrectnessCheckResult
+  module: omnimarket.nodes.node_handler_correctness_gate.models.model_correctness_check_result
+
+handler:
+  name: HandlerCorrectnessGate
+  module: omnimarket.nodes.node_handler_correctness_gate.handlers.handler_correctness_gate
+  class: HandlerCorrectnessGate
+
+handler_routing:
+  routing_strategy: operation_match
+  handlers:
+    - handler:
+        name: HandlerCorrectnessGate
+        module: omnimarket.nodes.node_handler_correctness_gate.handlers.handler_correctness_gate
+      event_model:
+        name: ModelCorrectnessCheckRequest
+        module: omnimarket.nodes.node_handler_correctness_gate.models.model_correctness_check_request
+
+descriptor:
+  node_archetype: compute
+  purity: pure
+  idempotent: true
+  timeout_ms: 60000
+  runtime_profiles: [main]
+
+event_bus:
+  version: {major: 1, minor: 0, patch: 0}
+  plugin_managed: false
+  subscribe_topics:
+    - onex.cmd.omnimarket.handler-correctness-check.v1
+  publish_topics:
+    - onex.evt.omnimarket.handler-correctness-result.v1
+
+terminal_event: onex.evt.omnimarket.handler-correctness-result.v1
+
+externally_consumed_topics:
+  - onex.evt.omnimarket.handler-correctness-result.v1
+
+published_events:
+  - event_type: HandlerCorrectnessResult
+    topic: onex.evt.omnimarket.handler-correctness-result.v1
+    description: Handler correctness gate result with score and failure details.
+
+metadata:
+  transport_type: kafka
+  author: OmniNode Team
+  license: MIT
+  created: "2026-05-25"
+  tags: [compute, eval, correctness, gate, OMN-12130]
+  related_tickets:
+    - "OMN-12130"

src/omnimarket/nodes/node_handler_correctness_gate/contracts/example_eval_set.yaml

@@ -0,0 +1,19 @@
+name: "sentiment-classifier-v1"
+description: "Basic sentiment classification correctness check"
+min_score: 0.85
+entries:
+  - input: "This product is terrible and I want a refund"
+    expected: "negative"
+    scoring: "exact_match"
+  - input: "Best purchase I've ever made"
+    expected: "positive"
+    scoring: "exact_match"
+  - input: "It arrived on time"
+    expected: "neutral"
+    scoring: "exact_match"
+  - input: "Absolutely love it, 5 stars"
+    expected: "positive"
+    scoring: "exact_match"
+  - input: "Worst experience ever, never buying again"
+    expected: "negative"
+    scoring: "exact_match"

src/omnimarket/nodes/node_handler_correctness_gate/handlers/__init__.py

@@ -0,0 +1,2 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT

src/omnimarket/nodes/node_handler_correctness_gate/handlers/handler_correctness_gate.py

@@ -0,0 +1,81 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT
+from __future__ import annotations
+
+from omnimarket.nodes.node_handler_correctness_gate.models.enums import (
+    EnumScoringMethod,
+)
+from omnimarket.nodes.node_handler_correctness_gate.models.model_correctness_check_request import (
+    ModelCorrectnessCheckRequest,
+)
+from omnimarket.nodes.node_handler_correctness_gate.models.model_correctness_check_result import (
+    ModelCorrectnessCheckResult,
+)
+from omnimarket.nodes.node_handler_correctness_gate.models.model_eval_entry import (
+    ModelEvalEntry,
+)
+from omnimarket.nodes.node_handler_correctness_gate.models.model_eval_failure import (
+    ModelEvalFailure,
+)
+
+
+def _score_entry(entry: ModelEvalEntry, actual: str) -> bool:
+    expected = entry.expected
+    if entry.scoring == EnumScoringMethod.EXACT_MATCH:
+        return actual == expected
+    if entry.scoring == EnumScoringMethod.CONTAINS:
+        return expected in actual
+    if entry.scoring == EnumScoringMethod.STARTS_WITH:
+        return actual.startswith(expected)
+    return False
+
+
+class HandlerCorrectnessGate:
+    def handle(
+        self, request: ModelCorrectnessCheckRequest
+    ) -> ModelCorrectnessCheckResult:
+        entries = request.eval_set.entries
+        total = len(entries)
+
+        if total == 0:
+            return ModelCorrectnessCheckResult(
+                handler_id=request.handler_id,
+                score=0.0,
+                passed=False,
+                total_entries=0,
+                correct_entries=0,
+                failures=(),
+                eval_set_name=request.eval_set.name,
+            )
+
+        actual_outputs = request.actual_outputs
+        failures: list[ModelEvalFailure] = []
+        correct = 0
+
+        for idx, entry in enumerate(entries):
+            actual = actual_outputs[idx] if idx < len(actual_outputs) else ""
+            if _score_entry(entry, actual):
+                correct += 1
+            else:
+                failures.append(
+                    ModelEvalFailure(
+                        entry_index=idx,
+                        input=entry.input,
+                        expected=entry.expected,
+                        actual=actual,
+                        scoring=entry.scoring,
+                    )
+                )
+
+        score = correct / total
+        passed = score >= request.eval_set.min_score
+
+        return ModelCorrectnessCheckResult(
+            handler_id=request.handler_id,
+            score=score,
+            passed=passed,
+            total_entries=total,
+            correct_entries=correct,
+            failures=tuple(failures),
+            eval_set_name=request.eval_set.name,
+        )

src/omnimarket/nodes/node_handler_correctness_gate/metadata.yaml

@@ -0,0 +1,23 @@
+name: node_handler_correctness_gate
+version: "1.0.0"
+description: "Pure compute node that scores pre-computed handler outputs against an eval set"
+omnibase_core_compat: ">=0.39.0,<1.0.0"
+entry_points:
+  onex.nodes:
+    node_handler_correctness_gate: "omnimarket.nodes.node_handler_correctness_gate"
+capabilities:
+  standalone: true
+  full_runtime: false
+  requires_network: false
+  requires_repo: false
+  requires_secrets: false
+  requires_docker: false
+  side_effect_class: pure
+dependencies:
+  - "pydantic>=2.0"
+authors: ["OmniNode Team"]
+license: "MIT"
+tags: ["compute", "eval", "correctness", "gate", "OMN-12130"]
+pack: "eval"
+display_name: "Handler Correctness Gate"
+node_role: "compute"

src/omnimarket/nodes/node_handler_correctness_gate/models/__init__.py

@@ -0,0 +1,2 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT

src/omnimarket/nodes/node_handler_correctness_gate/models/enums.py

@@ -0,0 +1,11 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT
+from __future__ import annotations
+
+from enum import StrEnum
+
+
+class EnumScoringMethod(StrEnum):
+    EXACT_MATCH = "exact_match"
+    CONTAINS = "contains"
+    STARTS_WITH = "starts_with"

src/omnimarket/nodes/node_handler_correctness_gate/models/model_correctness_check_request.py

@@ -0,0 +1,18 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT
+from __future__ import annotations
+
+from pydantic import BaseModel, ConfigDict
+
+from omnimarket.nodes.node_handler_correctness_gate.models.model_eval_set import (
+    ModelEvalSet,
+)
+
+
+class ModelCorrectnessCheckRequest(BaseModel):
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    handler_id: str
+    eval_set: ModelEvalSet
+    actual_outputs: tuple[str, ...]
+    correlation_id: str

src/omnimarket/nodes/node_handler_correctness_gate/models/model_correctness_check_result.py

@@ -0,0 +1,21 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT
+from __future__ import annotations
+
+from pydantic import BaseModel, ConfigDict
+
+from omnimarket.nodes.node_handler_correctness_gate.models.model_eval_failure import (
+    ModelEvalFailure,
+)
+
+
+class ModelCorrectnessCheckResult(BaseModel):
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    handler_id: str
+    score: float
+    passed: bool
+    total_entries: int
+    correct_entries: int
+    failures: tuple[ModelEvalFailure, ...] = ()
+    eval_set_name: str = ""

src/omnimarket/nodes/node_handler_correctness_gate/models/model_eval_entry.py

@@ -0,0 +1,17 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT
+from __future__ import annotations
+
+from pydantic import BaseModel, ConfigDict
+
+from omnimarket.nodes.node_handler_correctness_gate.models.enums import (
+    EnumScoringMethod,
+)
+
+
+class ModelEvalEntry(BaseModel):
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    input: str
+    expected: str
+    scoring: EnumScoringMethod = EnumScoringMethod.EXACT_MATCH

src/omnimarket/nodes/node_handler_correctness_gate/models/model_eval_failure.py

@@ -0,0 +1,19 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT
+from __future__ import annotations
+
+from pydantic import BaseModel, ConfigDict
+
+from omnimarket.nodes.node_handler_correctness_gate.models.enums import (
+    EnumScoringMethod,
+)
+
+
+class ModelEvalFailure(BaseModel):
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    entry_index: int
+    input: str
+    expected: str
+    actual: str
+    scoring: EnumScoringMethod

src/omnimarket/nodes/node_handler_correctness_gate/models/model_eval_set.py

@@ -0,0 +1,18 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT
+from __future__ import annotations
+
+from pydantic import BaseModel, ConfigDict
+
+from omnimarket.nodes.node_handler_correctness_gate.models.model_eval_entry import (
+    ModelEvalEntry,
+)
+
+
+class ModelEvalSet(BaseModel):
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    entries: tuple[ModelEvalEntry, ...]
+    min_score: float = 0.85
+    name: str = ""
+    description: str = ""

src/omnimarket/nodes/node_handler_correctness_gate/tests/__init__.py

@@ -0,0 +1,2 @@
+# SPDX-FileCopyrightText: 2026 OmniNode.ai Inc.
+# SPDX-License-Identifier: MIT