victor-codegraph 0.9.0__tar.gz → 0.10.0__tar.gz

{victor_codegraph-0.9.0 → victor_codegraph-0.10.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: victor-codegraph
-Version: 0.9.0
+Version: 0.10.0
 Summary: Repository-to-Code-Semantic-Graph compiler with stable identity, bounded chunks, incremental deltas, and ProximaRecord projection.
 Author-email: Vijaykumar Singh <vijay@anvaiops.com>
 License: Apache-2.0
@@ -16,6 +16,8 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Software Development :: Libraries
 Requires-Python: >=3.12
 Description-Content-Type: text/markdown
+Provides-Extra: embeddings
+Requires-Dist: sentence-transformers<6,>=3; extra == "embeddings"
 Provides-Extra: treesitter
 Requires-Dist: tree-sitter>=0.26.0; extra == "treesitter"
 Requires-Dist: tree-sitter-language-pack>=1.21.3; extra == "treesitter"
@@ -54,7 +56,7 @@ gaps:
 ## Install
 
 Published on PyPI as [`victor-codegraph`](https://pypi.org/project/victor-codegraph/).
-This source tree is version **0.9.0**; consumers should use a compatible `<1.0` pin.
+This source tree is version **0.10.0**; consumers should use a compatible `<1.0` pin.
 
 ```bash
 # from PyPI
@@ -126,7 +128,7 @@ records = to_proxima_records(repository, repo_graph_id="myrepo")
 
 ## Status
 
-`0.9.0` source version. Python (`ast`) is the full, offline frontend. JS/TS, Go, Rust,
+`0.10.0` source version. Python (`ast`) is the full, offline frontend. JS/TS, Go, Rust,
 Java, C, and C++ have conformance fixtures; other mapped tree-sitter languages report
 symbols-tier or fallback status rather than implying full fidelity.
 
@@ -143,4 +145,30 @@ workspace-inherited dependencies, re-export chains, macros and unknown receiver
 types retain unresolved references. Glob imports retain a module dependency but
 do not guess call targets. This is a conservative static graph, not rustc name
 resolution or evidence that a call executes. Publish this package before a Victor
-release that requires its `0.9.0` repository graph contract.
+release that requires its exact repository graph contract.
+
+
+### Embedding conformance (0.10.0)
+
+The accepted FEP-0040 supplies store-neutral
+`EmbeddingProfile`, `SymbolEmbedding`, `embed_symbols` and an optional
+`SentenceTransformerEncoder`. The base compiler stays dependency-free. The
+`embeddings` extra loads sentence-transformers only when explicitly selected;
+private source is not sent to a remote model by default.
+
+The contract uses canonical compiler IDs and `symbol-summary-v1` (name,
+signature and documentation), with an immutable model revision, dimensions,
+token limit, normalization, query prefix and separate input/source hashes.
+Inference errors propagate; hash placeholders, zero vectors and malformed batches
+are not semantic evidence. File/container symbols marked `embedding_excluded`
+remain graph evidence only.
+
+These APIs are introduced in 0.10.0 and are unavailable in 0.9.0. Install the
+exact published `victor-codegraph[treesitter,embeddings]==0.10.0` when selecting
+local model execution; the base package has no mandatory model dependency.
+Current evidence covers
+synthetic document-vector persistence parity and local LanceDB collection/query
+profile enforcement. Labeled relevance and hosted lifecycle remain separate
+acceptance gates. Victor binds a new LanceDB table to the complete profile in its
+Arrow schema and rejects incompatible query encoders and mixed/orphan row markers.
+Legacy tables require a fresh destination; they are never silently relabeled.

{victor_codegraph-0.9.0 → victor_codegraph-0.10.0}/README.md

@@ -25,7 +25,7 @@ gaps:
 ## Install
 
 Published on PyPI as [`victor-codegraph`](https://pypi.org/project/victor-codegraph/).
-This source tree is version **0.9.0**; consumers should use a compatible `<1.0` pin.
+This source tree is version **0.10.0**; consumers should use a compatible `<1.0` pin.
 
 ```bash
 # from PyPI
@@ -97,7 +97,7 @@ records = to_proxima_records(repository, repo_graph_id="myrepo")
 
 ## Status
 
-`0.9.0` source version. Python (`ast`) is the full, offline frontend. JS/TS, Go, Rust,
+`0.10.0` source version. Python (`ast`) is the full, offline frontend. JS/TS, Go, Rust,
 Java, C, and C++ have conformance fixtures; other mapped tree-sitter languages report
 symbols-tier or fallback status rather than implying full fidelity.
 
@@ -114,4 +114,30 @@ workspace-inherited dependencies, re-export chains, macros and unknown receiver
 types retain unresolved references. Glob imports retain a module dependency but
 do not guess call targets. This is a conservative static graph, not rustc name
 resolution or evidence that a call executes. Publish this package before a Victor
-release that requires its `0.9.0` repository graph contract.
+release that requires its exact repository graph contract.
+
+
+### Embedding conformance (0.10.0)
+
+The accepted FEP-0040 supplies store-neutral
+`EmbeddingProfile`, `SymbolEmbedding`, `embed_symbols` and an optional
+`SentenceTransformerEncoder`. The base compiler stays dependency-free. The
+`embeddings` extra loads sentence-transformers only when explicitly selected;
+private source is not sent to a remote model by default.
+
+The contract uses canonical compiler IDs and `symbol-summary-v1` (name,
+signature and documentation), with an immutable model revision, dimensions,
+token limit, normalization, query prefix and separate input/source hashes.
+Inference errors propagate; hash placeholders, zero vectors and malformed batches
+are not semantic evidence. File/container symbols marked `embedding_excluded`
+remain graph evidence only.
+
+These APIs are introduced in 0.10.0 and are unavailable in 0.9.0. Install the
+exact published `victor-codegraph[treesitter,embeddings]==0.10.0` when selecting
+local model execution; the base package has no mandatory model dependency.
+Current evidence covers
+synthetic document-vector persistence parity and local LanceDB collection/query
+profile enforcement. Labeled relevance and hosted lifecycle remain separate
+acceptance gates. Victor binds a new LanceDB table to the complete profile in its
+Arrow schema and rejects incompatible query encoders and mixed/orphan row markers.
+Legacy tables require a fresh destination; they are never silently relabeled.

{victor_codegraph-0.9.0 → victor_codegraph-0.10.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "victor-codegraph"
-version = "0.9.0"
+version = "0.10.0"
 description = "Repository-to-Code-Semantic-Graph compiler with stable identity, bounded chunks, incremental deltas, and ProximaRecord projection."
 readme = "README.md"
 requires-python = ">=3.12"
@@ -40,6 +40,7 @@ classifiers = [
 dependencies = []
 
 [project.optional-dependencies]
+embeddings = ["sentence-transformers>=3,<6"]
 treesitter = [
     "tree-sitter>=0.26.0",
     "tree-sitter-language-pack>=1.21.3",
@@ -77,3 +78,6 @@ select = ["E4", "E7", "E9", "F"]
 [tool.pytest.ini_options]
 testpaths = ["tests"]
 addopts = ""
+
+[tool.setuptools.package-data]
+victor_codegraph = ["py.typed"]

victor_codegraph-0.10.0/tests/test_embedding_contract.py

@@ -0,0 +1,145 @@
+"""One semantic input/provenance contract for local and hosted consumers."""
+
+from dataclasses import replace
+import pytest
+from victor_codegraph import parse
+from victor_codegraph.embeddings import (
+    BGE_SMALL_PROFILE,
+    embedding_text,
+    embed_symbols,
+    validate_vectors,
+)
+
+
+def test_same_symbols_produce_store_independent_embedding_evidence():
+    symbols = parse(
+        'def pay(x):\n    """Charge a customer."""\n    return x\n', file_path="billing.py"
+    ).symbols
+    seen = []
+
+    def encode(texts):
+        seen.extend(texts)
+        return [[1.0] + [0.0] * 383 for _ in texts]
+
+    documents = embed_symbols(symbols, encode, BGE_SMALL_PROFILE)
+    assert seen == [embedding_text(s.simple_name, s.signature, s.documentation) for s in symbols]
+    assert [d.symbol_id for d in documents] == [s.id for s in symbols]
+    assert all(d.profile_id == BGE_SMALL_PROFILE.identity for d in documents)
+    assert all(len(d.input_hash) == 64 and len(d.source_hash) == 64 for d in documents)
+    assert all(len(d.values) == 384 for d in documents)
+
+
+@pytest.mark.parametrize(
+    "vectors",
+    [[], [[0.0, 0.0]], [[True, 1.0]], [[float("nan"), 1.0]], [[1.0]], [[float("inf"), 1.0]]],
+)
+def test_invalid_or_missing_vectors_fail_without_substitution(vectors):
+    with pytest.raises(ValueError):
+        validate_vectors(vectors, count=1, dimension=2)
+
+
+def test_model_preprocessing_changes_invalidate_profile_identity():
+    p = BGE_SMALL_PROFILE
+    assert (
+        len(
+            {
+                p.identity,
+                replace(p, revision="a" * 40).identity,
+                replace(p, normalize=True).identity,
+                replace(p, query_prefix="query: ").identity,
+                replace(p, max_tokens=256).identity,
+            }
+        )
+        == 5
+    )
+    with pytest.raises(ValueError, match="revision"):
+        replace(p, revision="main")
+
+
+def test_failed_encoder_propagates_and_duplicate_symbol_ids_are_rejected():
+    symbols = parse("def f(): pass", file_path="m.py").symbols
+
+    def fail(texts):
+        raise RuntimeError("model unavailable")
+
+    with pytest.raises(RuntimeError, match="model unavailable"):
+        embed_symbols(symbols, fail, BGE_SMALL_PROFILE)
+    with pytest.raises(ValueError, match="duplicate"):
+        embed_symbols(symbols * 2, fail, BGE_SMALL_PROFILE)
+
+
+def test_unbounded_encoder_outputs_are_rejected_after_one_overflow_item():
+    from itertools import repeat
+
+    with pytest.raises(ValueError, match="dimension"):
+        validate_vectors([repeat(1.0)], count=1, dimension=2)
+    with pytest.raises(ValueError, match="exactly one"):
+        validate_vectors(repeat([1.0, 2.0]), count=1, dimension=2)
+
+
+def test_optional_encoder_owns_one_pinned_model_and_query_preprocessing(monkeypatch):
+    import asyncio
+    import sys
+    from types import SimpleNamespace
+    from victor_codegraph.embeddings import SentenceTransformerEncoder
+
+    calls = []
+
+    class Model:
+        max_seq_length = 0
+
+        def __init__(self, model_id, **kwargs):
+            calls.append((model_id, kwargs))
+
+        def get_sentence_embedding_dimension(self):
+            return 3
+
+        def encode(self, texts, **kwargs):
+            calls.append((texts, kwargs, self.max_seq_length))
+            return [[1.0, 2.0, 3.0] for _ in texts]
+
+    monkeypatch.setitem(
+        sys.modules, "sentence_transformers", SimpleNamespace(SentenceTransformer=Model)
+    )
+    profile = replace(BGE_SMALL_PROFILE, dimension=3, query_prefix="query: ", max_tokens=256)
+    encoder = SentenceTransformerEncoder(profile)
+    assert encoder.encode([]) == [] and calls == []
+    assert encoder.encode_query("permissions") == [1.0, 2.0, 3.0]
+    assert asyncio.run(encoder.embed_text("symbol")) == [1.0, 2.0, 3.0]
+    assert len(calls) == 3  # one model load, two bounded inference calls
+    assert calls[0] == (
+        profile.model_id,
+        {"revision": profile.revision, "device": "cpu", "trust_remote_code": False},
+    )
+    assert calls[1][0] == ["query: permissions"]
+    assert calls[1][1]["normalize_embeddings"] is False
+    assert calls[1][2] == 256
+    with pytest.raises(ValueError, match="512"):
+        encoder.encode(["x"] * 513)
+    with pytest.raises(ValueError, match="dimension"):
+        SentenceTransformerEncoder(BGE_SMALL_PROFILE).encode(["x"])
+
+
+def test_optional_encoder_missing_dependency_is_explicit(monkeypatch):
+    import sys
+    from victor_codegraph.embeddings import SentenceTransformerEncoder
+
+    monkeypatch.setitem(sys.modules, "sentence_transformers", None)
+    with pytest.raises(ImportError):
+        SentenceTransformerEncoder().encode(["symbol"])
+
+
+@pytest.mark.parametrize(
+    "changes",
+    [
+        {"model_id": ""},
+        {"revision": "main"},
+        {"dimension": True},
+        {"max_tokens": 0},
+        {"text_contract": "body-v1"},
+        {"normalize": 1},
+    ],
+)
+def test_invalid_profile_cannot_claim_semantic_compatibility(changes):
+    with pytest.raises(ValueError):
+        replace(BGE_SMALL_PROFILE, **changes)

{victor_codegraph-0.9.0 → victor_codegraph-0.10.0}/victor_codegraph/__init__.py

@@ -14,6 +14,15 @@ from __future__ import annotations
 
 from .adapter import relation_to_record, symbol_to_record, to_proxima_records
 from .config import ChunkConfig
+from .embeddings import (
+    BGE_SMALL_PROFILE,
+    EmbeddingProfile,
+    SymbolEmbedding,
+    SentenceTransformerEncoder,
+    embedding_text,
+    embed_symbols,
+    validate_vectors,
+)
 from .languages import detect_language
 from .model import (
     CapabilityTier,
@@ -45,10 +54,17 @@ from .repository_index import (
     relation_key,
 )
 
-__version__ = "0.9.0"
+__version__ = "0.10.0"
 
 __all__ = [
     "__version__",
+    "BGE_SMALL_PROFILE",
+    "EmbeddingProfile",
+    "SymbolEmbedding",
+    "SentenceTransformerEncoder",
+    "embedding_text",
+    "embed_symbols",
+    "validate_vectors",
     "chunk",
     "parse",
     "chunk_repo",

victor_codegraph-0.10.0/victor_codegraph/embeddings.py

@@ -0,0 +1,196 @@
+"""Store-neutral semantic embedding evidence; model dependencies are optional.
+
+Inference and publication failures propagate. Graph identity is supplied by the
+compiler; this module neither derives replacement IDs nor writes to a database.
+"""
+
+from __future__ import annotations
+
+from dataclasses import asdict, dataclass
+import hashlib
+import json
+from itertools import islice
+import math
+from numbers import Real
+import re
+from typing import Any, Callable, Iterable, Sequence
+
+from .model import CodeSymbol
+
+
+def embedding_text(name: str, signature: str | None, documentation: str | None) -> str:
+    """The symbol-summary-v1 input, identical to Victor's existing node text."""
+    return " ".join(filter(None, [name, signature or "", documentation or ""]))
+
+
+@dataclass(frozen=True)
+class EmbeddingProfile:
+    """Complete compatibility key for document and query vectors."""
+
+    model_id: str
+    revision: str
+    dimension: int
+    text_contract: str = "symbol-summary-v1"
+    max_tokens: int = 512
+    normalize: bool = False
+    query_prefix: str = ""
+
+    def __post_init__(self) -> None:
+        if not isinstance(self.model_id, str) or not self.model_id.strip():
+            raise ValueError("model_id is required")
+        if not isinstance(self.revision, str) or not re.fullmatch(r"[0-9a-f]{40}", self.revision):
+            raise ValueError("model revision must be an immutable 40-hex commit")
+        if type(self.dimension) is not int or not 1 <= self.dimension <= 65536:
+            raise ValueError("invalid embedding dimension")
+        if type(self.max_tokens) is not int or not 1 <= self.max_tokens <= 32768:
+            raise ValueError("invalid embedding token limit")
+        if self.text_contract != "symbol-summary-v1":
+            raise ValueError("unsupported embedding text contract")
+        if type(self.normalize) is not bool or not isinstance(self.query_prefix, str):
+            raise ValueError("invalid embedding preprocessing")
+
+    @property
+    def identity(self) -> str:
+        return _digest(json.dumps(asdict(self), sort_keys=True, separators=(",", ":")))
+
+
+BGE_SMALL_PROFILE = EmbeddingProfile(
+    model_id="BAAI/bge-small-en-v1.5",
+    revision="5c38ec7c405ec4b44b94cc5a9bb96e735b38267a",
+    dimension=384,
+)
+
+
+@dataclass(frozen=True)
+class SymbolEmbedding:
+    symbol_id: str
+    profile_id: str
+    input_hash: str
+    source_hash: str
+    values: tuple[float, ...]
+
+
+def _digest(text: str) -> str:
+    return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+
+def validate_vectors(
+    vectors: Iterable[Iterable[float]], *, count: int, dimension: int
+) -> list[list[float]]:
+    """Reject failed, truncated or malformed inference before any publication."""
+    if type(count) is not int or not 0 <= count <= 512:
+        raise ValueError("embedding batch must contain at most 512 inputs")
+    if type(dimension) is not int or not 1 <= dimension <= 65536:
+        raise ValueError("invalid embedding dimension")
+    rows = [list(islice(row, dimension + 1)) for row in islice(vectors, count + 1)]
+    if len(rows) != count:
+        raise ValueError("embedding batch must return exactly one vector per input")
+    for row in rows:
+        if len(row) != dimension:
+            raise ValueError("embedding dimension mismatch")
+        if any(
+            isinstance(v, bool) or not isinstance(v, Real) or not math.isfinite(float(v))
+            for v in row
+        ):
+            raise ValueError("embedding values must be finite numeric values")
+        if not any(float(v) != 0.0 for v in row):
+            raise ValueError("zero vector is not a semantic embedding")
+    return [[float(v) for v in row] for row in rows]
+
+
+def embed_symbols(
+    symbols: Sequence[CodeSymbol],
+    encode: Callable[[list[str]], Iterable[Iterable[float]]],
+    profile: EmbeddingProfile,
+    *,
+    batch_size: int = 32,
+) -> list[SymbolEmbedding]:
+    """Prepare, batch and validate compiler symbols without a store dependency.
+
+    ``encode`` is trusted configuration: it must implement the supplied profile.
+    Backend identity and real-model quality require separate deployment acceptance.
+    """
+    if type(batch_size) is not int or not 1 <= batch_size <= 512:
+        raise ValueError("embedding batch size must be between 1 and 512")
+    ids = [s.id for s in symbols]
+    if len(set(ids)) != len(ids) or any(not sid for sid in ids):
+        raise ValueError("empty or duplicate canonical symbol identity")
+    eligible = [s for s in symbols if not s.metadata.get("embedding_excluded")]
+    result = []
+    for start in range(0, len(eligible), batch_size):
+        batch = eligible[start : start + batch_size]
+        texts = [embedding_text(s.simple_name, s.signature, s.documentation) for s in batch]
+        vectors = validate_vectors(encode(texts), count=len(texts), dimension=profile.dimension)
+        for symbol, text, vector in zip(batch, texts, vectors, strict=True):
+            result.append(
+                SymbolEmbedding(
+                    symbol.id,
+                    profile.identity,
+                    _digest(text),
+                    _digest(symbol.source_code),
+                    tuple(vector),
+                )
+            )
+    return result
+
+
+class SentenceTransformerEncoder:
+    """Optional local inference for an explicit immutable profile.
+
+    Uses sentence-transformers itself; no alternative model implementation or
+    remote-code execution. Own one encoder per configured worker/model lifetime.
+    """
+
+    semantic = True
+
+    def __init__(
+        self, profile: EmbeddingProfile = BGE_SMALL_PROFILE, *, device: str = "cpu"
+    ) -> None:
+        import threading
+
+        self.profile = profile
+        self.device = device
+        self._model: Any = None
+        self._lock = threading.Lock()
+
+    def encode(self, texts: list[str]) -> list[list[float]]:
+        """Encode one bounded document batch; propagate model/installation errors."""
+        if len(texts) > 512:
+            raise ValueError("embedding batch exceeds 512 inputs")
+        if not texts:
+            return []
+        with self._lock:
+            if self._model is None:
+                from sentence_transformers import SentenceTransformer
+
+                model = SentenceTransformer(
+                    self.profile.model_id,
+                    revision=self.profile.revision,
+                    device=self.device,
+                    trust_remote_code=False,
+                )
+                if model.get_sentence_embedding_dimension() != self.profile.dimension:
+                    raise ValueError("model/profile embedding dimension mismatch")
+                model.max_seq_length = self.profile.max_tokens
+                self._model = model
+            values = self._model.encode(
+                texts,
+                batch_size=32,
+                normalize_embeddings=self.profile.normalize,
+                convert_to_numpy=True,
+                show_progress_bar=False,
+            )
+        return validate_vectors(values, count=len(texts), dimension=self.profile.dimension)
+
+    def encode_query(self, query: str) -> list[float]:
+        return self.encode([self.profile.query_prefix + query])[0]
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        import asyncio
+
+        return await asyncio.to_thread(self.encode, texts)
+
+    async def embed_text(self, text: str, use_cache: bool = True) -> list[float]:
+        # Compatibility with Victor's injected embedding-service seam. Cache
+        # ownership remains outside this encoder; the flag never changes semantics.
+        return (await self.embed_batch([text]))[0]

{victor_codegraph-0.9.0 → victor_codegraph-0.10.0}/victor_codegraph.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: victor-codegraph
-Version: 0.9.0
+Version: 0.10.0
 Summary: Repository-to-Code-Semantic-Graph compiler with stable identity, bounded chunks, incremental deltas, and ProximaRecord projection.
 Author-email: Vijaykumar Singh <vijay@anvaiops.com>
 License: Apache-2.0
@@ -16,6 +16,8 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Software Development :: Libraries
 Requires-Python: >=3.12
 Description-Content-Type: text/markdown
+Provides-Extra: embeddings
+Requires-Dist: sentence-transformers<6,>=3; extra == "embeddings"
 Provides-Extra: treesitter
 Requires-Dist: tree-sitter>=0.26.0; extra == "treesitter"
 Requires-Dist: tree-sitter-language-pack>=1.21.3; extra == "treesitter"
@@ -54,7 +56,7 @@ gaps:
 ## Install
 
 Published on PyPI as [`victor-codegraph`](https://pypi.org/project/victor-codegraph/).
-This source tree is version **0.9.0**; consumers should use a compatible `<1.0` pin.
+This source tree is version **0.10.0**; consumers should use a compatible `<1.0` pin.
 
 ```bash
 # from PyPI
@@ -126,7 +128,7 @@ records = to_proxima_records(repository, repo_graph_id="myrepo")
 
 ## Status
 
-`0.9.0` source version. Python (`ast`) is the full, offline frontend. JS/TS, Go, Rust,
+`0.10.0` source version. Python (`ast`) is the full, offline frontend. JS/TS, Go, Rust,
 Java, C, and C++ have conformance fixtures; other mapped tree-sitter languages report
 symbols-tier or fallback status rather than implying full fidelity.
 
@@ -143,4 +145,30 @@ workspace-inherited dependencies, re-export chains, macros and unknown receiver
 types retain unresolved references. Glob imports retain a module dependency but
 do not guess call targets. This is a conservative static graph, not rustc name
 resolution or evidence that a call executes. Publish this package before a Victor
-release that requires its `0.9.0` repository graph contract.
+release that requires its exact repository graph contract.
+
+
+### Embedding conformance (0.10.0)
+
+The accepted FEP-0040 supplies store-neutral
+`EmbeddingProfile`, `SymbolEmbedding`, `embed_symbols` and an optional
+`SentenceTransformerEncoder`. The base compiler stays dependency-free. The
+`embeddings` extra loads sentence-transformers only when explicitly selected;
+private source is not sent to a remote model by default.
+
+The contract uses canonical compiler IDs and `symbol-summary-v1` (name,
+signature and documentation), with an immutable model revision, dimensions,
+token limit, normalization, query prefix and separate input/source hashes.
+Inference errors propagate; hash placeholders, zero vectors and malformed batches
+are not semantic evidence. File/container symbols marked `embedding_excluded`
+remain graph evidence only.
+
+These APIs are introduced in 0.10.0 and are unavailable in 0.9.0. Install the
+exact published `victor-codegraph[treesitter,embeddings]==0.10.0` when selecting
+local model execution; the base package has no mandatory model dependency.
+Current evidence covers
+synthetic document-vector persistence parity and local LanceDB collection/query
+profile enforcement. Labeled relevance and hosted lifecycle remain separate
+acceptance gates. Victor binds a new LanceDB table to the complete profile in its
+Arrow schema and rejects incompatible query encoders and mixed/orphan row markers.
+Legacy tables require a fresh destination; they are never silently relabeled.

{victor_codegraph-0.9.0 → victor_codegraph-0.10.0}/victor_codegraph.egg-info/SOURCES.txt

@@ -4,6 +4,7 @@ tests/test_adapter.py
 tests/test_adapter_validation_v2.py
 tests/test_chunk_contract_v2.py
 tests/test_diagnostics_v2.py
+tests/test_embedding_contract.py
 tests/test_identity_v2.py
 tests/test_line_base_contract.py
 tests/test_manifest.py
@@ -21,11 +22,13 @@ tests/test_treesitter_relations.py
 victor_codegraph/__init__.py
 victor_codegraph/adapter.py
 victor_codegraph/config.py
+victor_codegraph/embeddings.py
 victor_codegraph/javascript_repository.py
 victor_codegraph/languages.py
 victor_codegraph/manifest.py
 victor_codegraph/model.py
 victor_codegraph/parser.py
+victor_codegraph/py.typed
 victor_codegraph/python_parser.py
 victor_codegraph/repo.py
 victor_codegraph/repository_index.py

{victor_codegraph-0.9.0 → victor_codegraph-0.10.0}/victor_codegraph.egg-info/requires.txt

@@ -8,6 +8,9 @@ ruff>=0.16.7
 tree-sitter>=0.26.0
 tree-sitter-language-pack>=1.21.3
 
+[embeddings]
+sentence-transformers<6,>=3
+
 [treesitter]
 tree-sitter>=0.26.0
 tree-sitter-language-pack>=1.21.3