odin-engine 0.1.0__py3-none-any.whl → 0.2.0__py3-none-any.whl

odin_engine-0.2.0.dist-info/RECORD

@@ -0,0 +1,63 @@
+benchmarks/__init__.py,sha256=rollRY8I0xNWexV_jU-uDWYmwTH9Lu9vIxBUwPy1N60,484
+benchmarks/datasets.py,sha256=MyvxpfuAwY2RVBJBvPyEzWgUMc4DcgX3IjMoz9mTVck,8171
+benchmarks/metrics.py,sha256=WXAGSpGFWKZGc_5tU-GNdQNvmApfklaeG7Tve1YPzEE,7751
+benchmarks/run_ablation.py,sha256=qdXTCr5PeakF-XyHGuYkqaDpfO0mAWsxQlyR3JPatUU,8484
+benchmarks/run_npll_benchmark.py,sha256=uQcUq3g-AlbjT13eeY2mALtmGagOwBBmjMi2K9EPa0c,7746
+npll/__init__.py,sha256=2t0fSdqg-P2QygPLeWyYT2LowERyBtifyN6CZVVt8Rk,231
+npll/bootstrap.py,sha256=mb3No4rTPeQ28kqGVqiJsVoTwO6FHNPUVFA280HnCYA,18375
+npll/npll_model.py,sha256=8pQOskXrugkah4OhQ3W4mYIaYfEu_XT8GvUMOJ06ZnU,26540
+npll/core/__init__.py,sha256=8LWXF_z93tf-gzYvRu7ycdvgnXD5Kr6Jaa2SiFZWAUQ,832
+npll/core/knowledge_graph.py,sha256=FwDrBNydCx_0PKBtWOnUbMapjkqvVkuPmmcj_fiK3Ow,11183
+npll/core/logical_rules.py,sha256=x7SrGA8SsszMjy4m8mPUGDyV14uKgy3nFpeRd4qxUDk,17771
+npll/core/mln.py,sha256=ShtA9MK-7vmqzQcS42A4iFE6CWmzFA1XlKCMfEqCaSQ,18585
+npll/inference/__init__.py,sha256=ydHHiRQ8hJAZe0ayHIDgKvDTzugn5q_0vK4O0NkkUN8,1039
+npll/inference/e_step.py,sha256=7u7Xk520WLcyu2f3OlUTRHmURdRh7bG-I91mrIx_nvk,16929
+npll/inference/elbo.py,sha256=vVUFHNzKvUhvW6c-1GSf9YIYy1ZACeY9_Z8WX4qnqRg,16886
+npll/inference/m_step.py,sha256=MG5LKPCcZHyYXr6SjuhKHDPxJOgPKobe_lS-RP_zOp8,22730
+npll/scoring/__init__.py,sha256=y2JMOBGpVGMZRvL0K07OiH1RObARqu9RYkBEr06aCcE,1296
+npll/scoring/embeddings.py,sha256=v6BID7cwjJ5i9-XnrKQ8WZV1YByrMCWaXsHTJQe9ZGQ,19478
+npll/scoring/probability.py,sha256=lgokSRFE7Cu6p5mYFb_PlmjSONaRMwqx2h79hEokA7o,16799
+npll/scoring/scoring_module.py,sha256=1hSNZzbnL6lw8ZYiglj8ok9pT_xQoDkCRIa8V51BwtU,13850
+npll/training/__init__.py,sha256=BLR-CcMMdQ_v9uKHVNuErq2-xE4eRHynOfWaawJZgGA,641
+npll/training/evaluation.py,sha256=rHPpofDOU5zYfbClrC3xrAMrYfDLrIYeMvh5IVZewFM,17155
+npll/training/npll_trainer.py,sha256=1Nt-cdkM3LGLFbe0S87OX7GEzvpABXJbCHIVNObXoPo,17880
+npll/utils/__init__.py,sha256=5xGn-xONVUiH8557_yTWwStBdNF46Yt2Pi7SHTPJa4c,1337
+npll/utils/batch_utils.py,sha256=W1sXyX9b42YOmy-i_7Sv-4KkAIasf8Vip4qUSpKwc1Y,18498
+npll/utils/config.py,sha256=b_s377KZ218jjS0_zWMgX92GkouOqbKk-LLpQHPb3z0,5134
+npll/utils/math_utils.py,sha256=O2-rSZxFDI0WtD2EWx6dnA_YayHFW04WD91V-2d2CxQ,11398
+odin/__init__.py,sha256=rV4ADDt5c43pVexCVcFB3OXlHXWLK3hc_xla2vSDpn8,670
+odin/engine.py,sha256=fG0vof-wcHuU5vrRSnbkzv2l812EvhjXgOZulQFLsu8,9102
+odin/schema.py,sha256=jmaaQRj94IHbppXdqtnIXv00lt3fLBHtZFrYhvqEslw,6524
+odin_engine-0.2.0.dist-info/licenses/LICENSE,sha256=nTutNg_5r6C-zttxTWEYi8nTR8CmF8hMbszt-2qPlgs,1070
+retrieval/__init__.py,sha256=Jf3-Dd82Xr4d78fCDYOpoQCAPCSZXDM_zxMWonp_d3g,2284
+retrieval/adapters.py,sha256=ElU3FzadpeyDc7aajYQQN0QHCKHl5KM4ou_B83NRBXQ,5021
+retrieval/adapters_arango.py,sha256=TP7_c8BeXGWxppyp3DkBCg2b45lOWB8ikwd7JU39zQg,53764
+retrieval/aggregators.py,sha256=Efce66eDSdHiErpEXBzX2dffhdQxOe8IxnLQnUUsGac,23934
+retrieval/beam.py,sha256=we5BV9NbKErFnYH8fOTRI5SklO2pOAycec2h2GW_X6I,4856
+retrieval/budget.py,sha256=uLUUTGOc98q5Ll08_GLNMoLdpY69L33_y7QY6wJ8flQ,1474
+retrieval/cache.py,sha256=OmDEFkOXjLCe5K5g-eiqtYjbslvV2pDJC0WV-DxDLrE,6107
+retrieval/confidence.py,sha256=dlxU-dIK0IK01Y-EAr5xI7pZAbEbhYRdjBnA0_M_p1M,3272
+retrieval/eval.py,sha256=j8_EaTaAFGPykhkIE3ruhbFzch7_1WJykDPTj4-q82M,1483
+retrieval/linker.py,sha256=ZpZuTeLwKDgNWYXPPuPo7_D-af5-0L1-g6tMbFYzx4M,3210
+retrieval/metrics.py,sha256=kDltG_Uu1Ows394Rcc0iWqNgsXbuFgfQSHnzfcevtjs,3288
+retrieval/metrics_motifs.py,sha256=wqVPSpfoIx5F1L9sUX6_LNAy_6zlKg27wUPsQKVcU8I,1346
+retrieval/orchestrator.py,sha256=jm4LzkhUL7P93BEpgPP16gZSkBYOOmMBqc3JAoLexjU,22493
+retrieval/ppr.py,sha256=xR_ovxOIVPMXHcWw5FTkrhnCIzxDqXOOVR6CHjdP894,5376
+retrieval/ppr_cache.py,sha256=mQCGDtwBu-Sw1eFoAfL3f1ryZSfIY2CCE4cMn5QwnsQ,760
+retrieval/scoring.py,sha256=ilQnrmdZvrbQ58io7nIcwvKdDlY2LpbQf2WO8RkU6ys,11022
+retrieval/ppr/__init__.py,sha256=O1f0pM9592kcC2CHRjSHYKO299zAY0RfunlUhk3K7mc,453
+retrieval/ppr/anchors.py,sha256=l_4JuB1UBaTZu3jLvc0z5I0-0QMUDF6QZSoc-7I5DYI,1221
+retrieval/ppr/bippr.py,sha256=byCxF4nCvnu__izsoof2aLSUxx2_sh9KLJd5ie6Ns30,2175
+retrieval/ppr/engines.py,sha256=FzraG3A3Zh2d-iVWIEfnJIrKrNH_lLBYv3sTREYW5sE,10132
+retrieval/ppr/global_pr.py,sha256=dABidLtuhAnfLdbJvSJnQKBbmdTO7woePL6xmQRh05k,2457
+retrieval/ppr/indexes.py,sha256=w1xmqr6xw91WPvShooj3DBV_x1kQWh37kkstcgO1M7I,2480
+retrieval/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+retrieval/utils/pii_redaction.py,sha256=hZyw_8HQM7yUsBm46yIBQCNT7Z-5xGqMXiVmR3ZqUqs,1224
+retrieval/writers/__init__.py,sha256=DvTadpcfbz5-XUI2d8dsP6CoajuYPIJ8vuYpHdDpCKc,203
+retrieval/writers/arango_writer.py,sha256=yfV8Q6EzI3Xso-moDYvt-NsjM5MITqmOkxUtm0cF3-4,1004
+retrieval/writers/base.py,sha256=pe_fgsI6ic8LUyNl2vShjK-OhPEmn2XqZrNW5B-dd1Y,637
+retrieval/writers/janus_writer.py,sha256=SmNEh3ebOZAuO8Iddcs2B7zA3r5GDnad4aMIdk5Wj20,1505
+odin_engine-0.2.0.dist-info/METADATA,sha256=q6EtepkN-Y2v30MR7q1K6tVPj-ANcp3yAiMl_Dgu2Y4,17358
+odin_engine-0.2.0.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+odin_engine-0.2.0.dist-info/top_level.txt,sha256=U53RwIYgQIJ0DBf_OaThVdHosqy06lkB4Zz4crmFadQ,31
+odin_engine-0.2.0.dist-info/RECORD,,

{odin_engine-0.1.0.dist-info → odin_engine-0.2.0.dist-info}/licenses/LICENSE

@@ -1,21 +1,21 @@
-MIT License
-
-Copyright (c) 2026 Prescott Data
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
+MIT License
+
+Copyright (c) 2026 Prescott Data
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

retrieval/__init__.py

@@ -1,50 +1,50 @@
-"""
-Retrieval components: budgets, PPR, beam search, adapters, and confidence providers.
-"""
-
-from .budget import SearchBudget, BudgetTracker
-from .adapters import GraphAccessor, KGCommunityAccessor, OverlayAccessor, JanusGraphAccessor
-from .adapters_arango import ArangoCommunityAccessor, GlobalGraphAccessor
-from .cache import CachedGraphAccessor
-from .confidence import EdgeConfidenceProvider, ConstantConfidence, NPLLConfidence
-from .ppr import (
-    PushPPREngine, MonteCarloPPREngine, BiPPREngine, PPRParams,
-    APPRAnchors, APPRAnchorParams,
-    GlobalPR, GlobalPRParams,
-    RandomWalkIndex, WalkIndexConfig,
-)
-from .beam import beam_search, BeamParams
-from .ppr_cache import PPRCache
-from .scoring import (
-    PathScoreConfig, path_score,
-    aggregate_evidence_strength,
-    InsightScoreConfig, insight_score,
-    score_paths_and_insight,
-)
-from .orchestrator import RetrievalOrchestrator, OrchestratorParams
-from .linker import CoherenceLinker, LinkerConfig, Mention
-from .metrics import MetricsLogger, JSONLSink, RetrievalMetrics, aggregate_latency_and_budget
-from .metrics_motifs import wedge_and_triad_closures
-from .eval import recall_at_k, expected_calibration_error, SimpleLLMCalibrator
-from .writers import PersistenceWriter, ArangoWriter, JanusGraphWriter
-
-__all__ = [
-    'GraphAccessor', 'KGCommunityAccessor', 'OverlayAccessor', 'JanusGraphAccessor', 
-    'ArangoCommunityAccessor', 'GlobalGraphAccessor',
-    'CachedGraphAccessor',
-    'EdgeConfidenceProvider', 'ConstantConfidence', 'NPLLConfidence',
-    'PushPPREngine', 'MonteCarloPPREngine', 'BiPPREngine', 'PPRParams',
-    'APPRAnchors', 'APPRAnchorParams', 'GlobalPR', 'GlobalPRParams',
-    'RandomWalkIndex', 'WalkIndexConfig',
-    'beam_search', 'BeamParams',
-    'PPRCache',
-    'PathScoreConfig', 'path_score', 'aggregate_evidence_strength',
-    'InsightScoreConfig', 'insight_score', 'score_paths_and_insight',
-    'RetrievalOrchestrator', 'OrchestratorParams', 'CoherenceLinker', 'LinkerConfig', 'Mention',
-    'MetricsLogger', 'JSONLSink', 'RetrievalMetrics',
-    'aggregate_latency_and_budget', 'wedge_and_triad_closures',
-    'recall_at_k', 'expected_calibration_error', 'SimpleLLMCalibrator',
-    'PersistenceWriter', 'ArangoWriter', 'JanusGraphWriter',
-    'SearchBudget', 'BudgetTracker'
-]
-
+"""
+Retrieval components: budgets, PPR, beam search, adapters, and confidence providers.
+"""
+
+from .budget import SearchBudget, BudgetTracker
+from .adapters import GraphAccessor, KGCommunityAccessor, OverlayAccessor, JanusGraphAccessor
+from .adapters_arango import ArangoCommunityAccessor, GlobalGraphAccessor
+from .cache import CachedGraphAccessor
+from .confidence import EdgeConfidenceProvider, ConstantConfidence, NPLLConfidence
+from .ppr import (
+    PushPPREngine, MonteCarloPPREngine, BiPPREngine, PPRParams,
+    APPRAnchors, APPRAnchorParams,
+    GlobalPR, GlobalPRParams,
+    RandomWalkIndex, WalkIndexConfig,
+)
+from .beam import beam_search, BeamParams
+from .ppr_cache import PPRCache
+from .scoring import (
+    PathScoreConfig, path_score,
+    aggregate_evidence_strength,
+    InsightScoreConfig, insight_score,
+    score_paths_and_insight,
+)
+from .orchestrator import RetrievalOrchestrator, OrchestratorParams
+from .linker import CoherenceLinker, LinkerConfig, Mention
+from .metrics import MetricsLogger, JSONLSink, RetrievalMetrics, aggregate_latency_and_budget
+from .metrics_motifs import wedge_and_triad_closures
+from .eval import recall_at_k, expected_calibration_error, SimpleLLMCalibrator
+from .writers import PersistenceWriter, ArangoWriter, JanusGraphWriter
+
+__all__ = [
+    'GraphAccessor', 'KGCommunityAccessor', 'OverlayAccessor', 'JanusGraphAccessor', 
+    'ArangoCommunityAccessor', 'GlobalGraphAccessor',
+    'CachedGraphAccessor',
+    'EdgeConfidenceProvider', 'ConstantConfidence', 'NPLLConfidence',
+    'PushPPREngine', 'MonteCarloPPREngine', 'BiPPREngine', 'PPRParams',
+    'APPRAnchors', 'APPRAnchorParams', 'GlobalPR', 'GlobalPRParams',
+    'RandomWalkIndex', 'WalkIndexConfig',
+    'beam_search', 'BeamParams',
+    'PPRCache',
+    'PathScoreConfig', 'path_score', 'aggregate_evidence_strength',
+    'InsightScoreConfig', 'insight_score', 'score_paths_and_insight',
+    'RetrievalOrchestrator', 'OrchestratorParams', 'CoherenceLinker', 'LinkerConfig', 'Mention',
+    'MetricsLogger', 'JSONLSink', 'RetrievalMetrics',
+    'aggregate_latency_and_budget', 'wedge_and_triad_closures',
+    'recall_at_k', 'expected_calibration_error', 'SimpleLLMCalibrator',
+    'PersistenceWriter', 'ArangoWriter', 'JanusGraphWriter',
+    'SearchBudget', 'BudgetTracker'
+]
+

retrieval/adapters.py

@@ -1,140 +1,140 @@
-from __future__ import annotations
-from typing import Iterable, Tuple, Protocol, List, Dict, Optional
-from gremlin_python.process.graph_traversal import __
-from gremlin_python.structure.graph import Graph
-
-NodeId = str
-RelId = str
-
-
-class GraphAccessor(Protocol):
-    """
-    Minimal graph view inside a single community/subgraph.
-    Implement these methods for your KG.
-    """
-
-    def iter_out(self, node: NodeId) -> Iterable[Tuple[NodeId, RelId, float]]:
-        """Yield (neighbor, relation, weight)."""
-        ...
-
-    def community_seed_norm(self, community_id: str, seeds: List[NodeId]) -> List[NodeId]:
-        """Optional mapping from external IDs to internal; default passthrough."""
-        return seeds
-
-    def nodes(self, community_id: str) -> Iterable[NodeId]:
-        """All node IDs in this community."""
-        ...
-
-    def degree(self, node: NodeId) -> int:
-        """Fast out-degree if available; else len(list(iter_out(node)))."""
-        ...
-
-
-class KGCommunityAccessor:
-    """
-    Example adapter to a KnowledgeGraph with a pre-sliced community.
-    Provide a set of allowed node IDs (entity names) for the community.
-    """
-
-    def __init__(self, kg, community_id: str, community_nodes: Optional[set[str]] = None):
-        self.kg = kg
-        self.community_id = community_id
-        self.allowed = community_nodes or set(n.name for n in kg.entities)
-
-    def iter_out(self, node: NodeId):
-        ent = self.kg.get_entity(node)
-        if ent is None:
-            return
-        for triple in self.kg.get_facts_by_head(ent):
-            v = triple.tail.name
-            if v in self.allowed:
-                yield v, triple.relation.name, 1.0
-
-    def nodes(self, community_id: str):
-        return list(self.allowed)
-
-    def degree(self, node: NodeId) -> int:
-        ent = self.kg.get_entity(node)
-        if ent is None:
-            return 0
-        return sum(1 for t in self.kg.get_facts_by_head(ent) if t.tail.name in self.allowed)
-
-
-class OverlayAccessor:
-    """
-    Wrap a base GraphAccessor with soft overlay edges.
-    Overlay edges are tuples (u, rel, v, weight) scoped to a community.
-    """
-
-    def __init__(self, base: GraphAccessor, community_id: str):
-        self.base = base
-        self.cid = community_id
-        self._overlay: dict[NodeId, list[tuple[NodeId, str, float]]] = {}
-
-    def add_edge(self, u: NodeId, rel: str, v: NodeId, weight: float = 1.0):
-        self._overlay.setdefault(u, []).append((v, rel, weight))
-
-    def iter_out(self, node: NodeId):
-        # Base edges
-        for v, r, w in self.base.iter_out(node):
-            yield v, r, w
-        # Overlay edges
-        for v, r, w in self._overlay.get(node, []):
-            yield v, r, w
-
-    def community_seed_norm(self, community_id: str, seeds: list[NodeId]) -> list[NodeId]:
-        return getattr(self.base, 'community_seed_norm', lambda cid, s: s)(community_id, seeds)
-
-    def nodes(self, community_id: str):
-        return getattr(self.base, 'nodes', lambda cid: [])(community_id)
-
-    def degree(self, node: NodeId) -> int:
-        base_deg = getattr(self.base, 'degree', lambda n: 0)(node)
-        return base_deg + len(self._overlay.get(node, []))
-
-
-class JanusGraphAccessor(GraphAccessor):
-    """
-    GraphAccessor implementation for JanusGraph.
-    Assumes connection to a JanusGraph instance via a Gremlin client.
-    """
-    def __init__(self, graph: Graph, community_id_property: str = "communityId"):
-        self.g = graph.traversal()
-        self.community_id_property = community_id_property
-
-    def iter_out(self, node: NodeId) -> Iterable[Tuple[NodeId, RelId, float]]:
-        """
-        Yields (neighbor, relation, weight) for outgoing edges from a node.
-        Example Gremlin query: g.V(node).outE().as_('e').inV().as_('v').select('e', 'v')
-        """
-        traversal = self.g.V(node).outE().as_('e').inV().as_('v').select('e', 'v').toList()
-        for edge, neighbor in traversal:
-            # Note: You might need to adjust how you access IDs and properties based on your graph schema
-            neighbor_id = neighbor.id
-            rel_id = edge.label
-            weight = edge.properties.get('weight', 1.0)
-            yield neighbor_id, rel_id, weight
-
-    def nodes(self, community_id: str) -> Iterable[NodeId]:
-        """
-        Yields all node IDs in a given community.
-        Example Gremlin query: g.V().has('communityId', community_id).id()
-        """
-        traversal = self.g.V().has(self.community_id_property, community_id).id().toList()
-        for node_id in traversal:
-            yield node_id
-
-    def degree(self, node: NodeId) -> int:
-        """
-        Returns the out-degree of a node.
-        Example Gremlin query: g.V(node).outE().count()
-        """
-        return self.g.V(node).outE().count().next()
-
-    def in_degree(self, node: NodeId) -> int:
-        """
-        Returns the in-degree of a node.
-        Example Gremlin query: g.V(node).inE().count()
-        """
-        return self.g.V(node).inE().count().next()
-
+from __future__ import annotations
+from typing import Iterable, Tuple, Protocol, List, Dict, Optional
+from gremlin_python.process.graph_traversal import __
+from gremlin_python.structure.graph import Graph
+
+NodeId = str
+RelId = str
+
+
+class GraphAccessor(Protocol):
+    """
+    Minimal graph view inside a single community/subgraph.
+    Implement these methods for your KG.
+    """
+
+    def iter_out(self, node: NodeId) -> Iterable[Tuple[NodeId, RelId, float]]:
+        """Yield (neighbor, relation, weight)."""
+        ...
+
+    def community_seed_norm(self, community_id: str, seeds: List[NodeId]) -> List[NodeId]:
+        """Optional mapping from external IDs to internal; default passthrough."""
+        return seeds
+
+    def nodes(self, community_id: str) -> Iterable[NodeId]:
+        """All node IDs in this community."""
+        ...
+
+    def degree(self, node: NodeId) -> int:
+        """Fast out-degree if available; else len(list(iter_out(node)))."""
+        ...
+
+
+class KGCommunityAccessor:
+    """
+    Example adapter to a KnowledgeGraph with a pre-sliced community.
+    Provide a set of allowed node IDs (entity names) for the community.
+    """
+
+    def __init__(self, kg, community_id: str, community_nodes: Optional[set[str]] = None):
+        self.kg = kg
+        self.community_id = community_id
+        self.allowed = community_nodes or set(n.name for n in kg.entities)
+
+    def iter_out(self, node: NodeId):
+        ent = self.kg.get_entity(node)
+        if ent is None:
+            return
+        for triple in self.kg.get_facts_by_head(ent):
+            v = triple.tail.name
+            if v in self.allowed:
+                yield v, triple.relation.name, 1.0
+
+    def nodes(self, community_id: str):
+        return list(self.allowed)
+
+    def degree(self, node: NodeId) -> int:
+        ent = self.kg.get_entity(node)
+        if ent is None:
+            return 0
+        return sum(1 for t in self.kg.get_facts_by_head(ent) if t.tail.name in self.allowed)
+
+
+class OverlayAccessor:
+    """
+    Wrap a base GraphAccessor with soft overlay edges.
+    Overlay edges are tuples (u, rel, v, weight) scoped to a community.
+    """
+
+    def __init__(self, base: GraphAccessor, community_id: str):
+        self.base = base
+        self.cid = community_id
+        self._overlay: dict[NodeId, list[tuple[NodeId, str, float]]] = {}
+
+    def add_edge(self, u: NodeId, rel: str, v: NodeId, weight: float = 1.0):
+        self._overlay.setdefault(u, []).append((v, rel, weight))
+
+    def iter_out(self, node: NodeId):
+        # Base edges
+        for v, r, w in self.base.iter_out(node):
+            yield v, r, w
+        # Overlay edges
+        for v, r, w in self._overlay.get(node, []):
+            yield v, r, w
+
+    def community_seed_norm(self, community_id: str, seeds: list[NodeId]) -> list[NodeId]:
+        return getattr(self.base, 'community_seed_norm', lambda cid, s: s)(community_id, seeds)
+
+    def nodes(self, community_id: str):
+        return getattr(self.base, 'nodes', lambda cid: [])(community_id)
+
+    def degree(self, node: NodeId) -> int:
+        base_deg = getattr(self.base, 'degree', lambda n: 0)(node)
+        return base_deg + len(self._overlay.get(node, []))
+
+
+class JanusGraphAccessor(GraphAccessor):
+    """
+    GraphAccessor implementation for JanusGraph.
+    Assumes connection to a JanusGraph instance via a Gremlin client.
+    """
+    def __init__(self, graph: Graph, community_id_property: str = "communityId"):
+        self.g = graph.traversal()
+        self.community_id_property = community_id_property
+
+    def iter_out(self, node: NodeId) -> Iterable[Tuple[NodeId, RelId, float]]:
+        """
+        Yields (neighbor, relation, weight) for outgoing edges from a node.
+        Example Gremlin query: g.V(node).outE().as_('e').inV().as_('v').select('e', 'v')
+        """
+        traversal = self.g.V(node).outE().as_('e').inV().as_('v').select('e', 'v').toList()
+        for edge, neighbor in traversal:
+            # Note: You might need to adjust how you access IDs and properties based on your graph schema
+            neighbor_id = neighbor.id
+            rel_id = edge.label
+            weight = edge.properties.get('weight', 1.0)
+            yield neighbor_id, rel_id, weight
+
+    def nodes(self, community_id: str) -> Iterable[NodeId]:
+        """
+        Yields all node IDs in a given community.
+        Example Gremlin query: g.V().has('communityId', community_id).id()
+        """
+        traversal = self.g.V().has(self.community_id_property, community_id).id().toList()
+        for node_id in traversal:
+            yield node_id
+
+    def degree(self, node: NodeId) -> int:
+        """
+        Returns the out-degree of a node.
+        Example Gremlin query: g.V(node).outE().count()
+        """
+        return self.g.V(node).outE().count().next()
+
+    def in_degree(self, node: NodeId) -> int:
+        """
+        Returns the in-degree of a node.
+        Example Gremlin query: g.V(node).inE().count()
+        """
+        return self.g.V(node).inE().count().next()
+