fix(vector): separate write transport fields

docs/superpowers/plans/2026-07-11-adapter-foundations.md

@@ -163,7 +163,8 @@ git commit -m "refactor(dwh): adapt direct and REST transports"
 - Modify: `harness/tht/vectorstore/reader.py`
 
 **Interfaces:**
-- Produces: `VectorStore`, `VectorCapabilities`, `VectorHealth`, `VectorRecord`, `VectorHit`, `ThothHttpVectorStore`.
+- Produces: `VectorStore`, `VectorCapabilities`, `VectorHealth`, `VectorRecord`,
+  `VectorWriteRecord`, `VectorHit`, `ThothHttpVectorStore`.
 - Preserves: current `VectorRestClient`, `DirectSearcher`, and `RestSearcher` behavior behind wrappers.
 
 - [ ] **Step 1: Write read/write capability and dual-credential tests**
@@ -193,9 +194,13 @@ class VectorStore(Protocol):
     def search(self, collections: list[str], embedding: list[float], *, limit: int,
                kinds: list[str] | None = None) -> list[VectorHit]: ...
     def existing_hashes(self, collection: str, kinds: list[str]) -> dict[str, str]: ...
-    def upsert(self, collection: str, records: list[VectorRecord]) -> int: ...
+    def upsert(self, collection: str, records: list[VectorWriteRecord]) -> int: ...
 ```
 
+`VectorWriteRecord` is the transport-neutral write envelope: it contains the canonical
+`VectorRecord`, a precomputed embedding, and a content hash. Adapters must preserve
+`VectorRecord.metadata` unchanged, including semantic keys named `embedding` or `content_hash`.
+
 - [ ] **Step 4: Run vector regression tests**
 
 Run: `cd harness && .venv/bin/pytest tests/test_vector_port_contract.py tests/test_vector_dual_key.py tests/test_search_similar_kinds.py tests/test_memory_save_one.py tests/test_solved_question.py -q`  

harness/tests/test_vector_port_contract.py

@@ -3,12 +3,15 @@ from unittest.mock import MagicMock
 import pytest
 
 from tht.adapters.vector.thoth_http import ThothHttpVectorStore
+from tht.evidence.model import EvidenceDoc
 from tht.ports.vector import (
     VectorHit,
     VectorRecord,
     VectorStore,
+    VectorWriteRecord,
     VectorWriteUnavailable,
 )
+from tht.vectorstore.records import evidence_records
 
 
 def test_http_store_reports_reader_without_writer():
@@ -67,13 +70,16 @@ def test_http_store_keeps_reader_and_writer_operations_separate():
     writer.existing_hashes.assert_called_once_with("memory", ["memory"])
 
     records = [
-        VectorRecord(
-            id="m1",
-            kind="memory",
-            ref="session:s1",
-            title="Choice",
-            content="Use the curated table",
-            metadata={"embedding": [0.1, 0.2], "content_hash": "abc"},
+        VectorWriteRecord(
+            record=VectorRecord(
+                id="m1",
+                kind="memory",
+                ref="session:s1",
+                title="Choice",
+                content="Use the curated table",
+            ),
+            embedding=[0.1, 0.2],
+            content_hash="abc",
         )
     ]
     assert store.upsert("memory", records) == 1
@@ -81,11 +87,64 @@ def test_http_store_keeps_reader_and_writer_operations_separate():
     reader.upsert_records.assert_not_called()
 
 
+def test_http_upsert_serializes_a_canonical_builder_record():
+    record = evidence_records(
+        [EvidenceDoc(id="joins", title="Join guidance", body="Use the curated join")],
+        max_chunk_chars=1000,
+    )[0]
+    writer = MagicMock()
+    writer.upsert_records.return_value = 1
+    store = ThothHttpVectorStore(reader=MagicMock(), writer=writer)
+
+    assert store.upsert(
+        "evidence",
+        [VectorWriteRecord(record=record, embedding=[0.2, 0.3], content_hash="digest")],
+    ) == 1
+    row = writer.upsert_records.call_args.args[1][0]
+    assert row["record_key"] == "evidence:joins:0"
+    assert row["metadata"]["status"] == "reviewed"
+    assert row["embedding"] == [0.2, 0.3]
+    assert row["content_hash"] == "digest"
+
+
+def test_http_upsert_preserves_metadata_named_like_transport_fields():
+    record = VectorRecord(
+        id="collision",
+        kind="memory",
+        ref="session:s1",
+        title="Collision",
+        content="Semantic metadata must survive",
+        metadata={"embedding": "semantic embedding", "content_hash": "semantic hash"},
+    )
+    writer = MagicMock()
+    store = ThothHttpVectorStore(reader=MagicMock(), writer=writer)
+
+    store.upsert(
+        "memory",
+        [VectorWriteRecord(record=record, embedding=[0.4], content_hash="transport hash")],
+    )
+    row = writer.upsert_records.call_args.args[1][0]
+    assert row["embedding"] == [0.4]
+    assert row["content_hash"] == "transport hash"
+    assert row["metadata"]["embedding"] == "semantic embedding"
+    assert row["metadata"]["content_hash"] == "semantic hash"
+
+
 def test_http_store_is_runtime_vector_store():
     store = ThothHttpVectorStore(reader=MagicMock(), writer=None)
     assert isinstance(store, VectorStore)
 
 
+def test_vector_contract_is_exported_from_public_packages():
+    from tht.adapters.vector import ThothHttpVectorStore as PublicHttpStore
+    from tht.ports import VectorStore as PublicVectorStore
+    from tht.ports import VectorWriteRecord as PublicVectorWriteRecord
+
+    assert PublicHttpStore is ThothHttpVectorStore
+    assert PublicVectorStore is VectorStore
+    assert PublicVectorWriteRecord is VectorWriteRecord
+
+
 def test_http_health_uses_reader_list_tables_and_reports_failure():
     reader = MagicMock()
     store = ThothHttpVectorStore(reader=reader, writer=None)

harness/tht/adapters/vector/legacy_direct.py

@@ -2,7 +2,12 @@
 
 from sqlalchemy import Engine
 
-from tht.ports.vector import VectorCapabilities, VectorHealth, VectorRecord, VectorWriteUnavailable
+from tht.ports.vector import (
+    VectorCapabilities,
+    VectorHealth,
+    VectorWriteRecord,
+    VectorWriteUnavailable,
+)
 from tht.vectorstore.store import VectorHit, VectorStore as TableVectorStore
 
 
@@ -43,5 +48,5 @@ class LegacyDirectVectorStore:
     def existing_hashes(self, collection: str, kinds: list[str]) -> dict[str, str]:
         raise VectorWriteUnavailable("Legacy direct reader has no writer interface")
 
-    def upsert(self, collection: str, records: list[VectorRecord]) -> int:
+    def upsert(self, collection: str, records: list[VectorWriteRecord]) -> int:
         raise VectorWriteUnavailable("Legacy direct reader has no writer interface")

harness/tht/adapters/vector/thoth_http.py

@@ -4,12 +4,11 @@ from tht.ports.vector import (
     VectorCapabilities,
     VectorHealth,
     VectorHit,
-    VectorRecord,
-    VectorStoreError,
+    VectorWriteRecord,
     VectorWriteUnavailable,
 )
 from tht.vectorstore.rest_client import VectorRestClient
-from tht.vectorstore.store import content_hash, hit_from_metadata
+from tht.vectorstore.store import hit_from_metadata
 
 
 def _merge(hits: list[VectorHit], limit: int) -> list[VectorHit]:
@@ -63,31 +62,26 @@ class ThothHttpVectorStore:
     def existing_hashes(self, collection: str, kinds: list[str]) -> dict[str, str]:
         return self._require_writer().existing_hashes(collection, kinds)
 
-    def upsert(self, collection: str, records: list[VectorRecord]) -> int:
+    def upsert(self, collection: str, records: list[VectorWriteRecord]) -> int:
         writer = self._require_writer()
         rows = [self._row(record) for record in records]
         return writer.upsert_records(collection, rows)
 
     @staticmethod
-    def _row(record: VectorRecord) -> dict:
-        extra = dict(record.metadata)
-        try:
-            embedding = extra.pop("embedding")
-        except KeyError as exc:
-            raise VectorStoreError(f"Vector record {record.id!r} has no embedding") from exc
-        digest = extra.pop("content_hash", content_hash(record.content))
+    def _row(write_record: VectorWriteRecord) -> dict:
+        record = write_record.record
         metadata = {
             "kind": record.kind,
             "ref": record.ref,
             "record_key": record.id,
             "title": record.title,
             "content": record.content,
-            **extra,
+            **record.metadata,
         }
         return {
             "record_key": record.id,
             "kind": record.kind,
-            "content_hash": digest,
+            "content_hash": write_record.content_hash,
             "metadata": metadata,
-            "embedding": embedding,
+            "embedding": write_record.embedding,
         }

harness/tht/ports/__init__.py

@@ -7,6 +7,16 @@ from tht.ports.dwh import (
     DistinctValues,
     UnsupportedCapability,
 )
+from tht.ports.vector import (
+    VectorCapabilities,
+    VectorHealth,
+    VectorHit,
+    VectorRecord,
+    VectorStore,
+    VectorStoreError,
+    VectorWriteRecord,
+    VectorWriteUnavailable,
+)
 
 __all__ = [
     "DwhAdapter",
@@ -14,4 +24,12 @@ __all__ = [
     "DwhHealth",
     "DistinctValues",
     "UnsupportedCapability",
+    "VectorCapabilities",
+    "VectorHealth",
+    "VectorHit",
+    "VectorRecord",
+    "VectorStore",
+    "VectorStoreError",
+    "VectorWriteRecord",
+    "VectorWriteUnavailable",
 ]

harness/tht/ports/vector.py

@@ -20,6 +20,15 @@ class VectorHealth:
     detail: str | None = None
 
 
+@dataclass(frozen=True)
+class VectorWriteRecord:
+    """A canonical record plus transport-neutral, precomputed vector data."""
+
+    record: VectorRecord
+    embedding: list[float]
+    content_hash: str
+
+
 class VectorStoreError(Exception):
     """Base error exposed by vector adapters."""
 
@@ -46,7 +55,7 @@ class VectorStore(Protocol):
 
     def existing_hashes(self, collection: str, kinds: list[str]) -> dict[str, str]: ...
 
-    def upsert(self, collection: str, records: list[VectorRecord]) -> int: ...
+    def upsert(self, collection: str, records: list[VectorWriteRecord]) -> int: ...
 
 
 __all__ = [
@@ -56,5 +65,6 @@ __all__ = [
     "VectorRecord",
     "VectorStore",
     "VectorStoreError",
+    "VectorWriteRecord",
     "VectorWriteUnavailable",
 ]