✨ The vector record becomes a five-field read contract, and zero-norm embeddings are rejected before any adapter

[ebigunso]

Summary

v0.1.6 Wave 2, Task_3 (ADR-I-0025 record read contract, ADR-I-0024 write-side zero-norm rule, ADR-I-0026 published surface policy), stacked on Task_2.

• The vector record and its typed manifest carry exactly five fields: object id, object type, surface, schema version, and the embedded text as provenance. Relationship, lifecycle, time, ranking, and object-specific hints, the readable text column, the graph URI, and the raw source reference are gone, with their per-field index creation.
• One Display/FromStr per enum in the domain replaces the adapter's private token mappers and the pipeline's copy.
• The vector indexing service rejects a zero-norm record embedding as a typed per-record indexing failure before any adapter sees it; one fixture (behind the non-default test-fixtures feature, enabled for integration tests by a self dev-dependency) serves the unit test now and the two-adapter parity suite in Task_4.
• The maximum number of embedding surfaces per object kind is published as public policy beside the surface policy, with a test that it matches the builders.
• Database documentation (payload design note, schema cheat sheet, README, graph schema design) describes the five-field record.

Validation

• Worker (cm-worker2): fmt; Clippy -D warnings on Rust 1.95 and 1.97; service-up cargo test (415 passed, 4 ignored); ignored Qdrant tests 4/4; zero-hit census for dropped fields, content_text readers, and enum-token copies; collection census unchanged.
• Reviewer (cm-reviewer, Tier D): approved at 36ef8c8 after two revisions on fixture reachability; an independent downstream crate compiled against the feature to prove the fixture is reachable.
• Cross-repository obligation, not a finding: the evaluation repository's exhaustive error-vocabulary conversion is its own prerequisite for re-pinning (ADR-I-0023 impact).

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The reviewed changes have no unresolved approval-blocking issues.

Pull request overview

Simplifies vector records to a five-field read contract and centralizes vector-related domain policy.

Changes:

• Removes obsolete Qdrant payload fields and indexes.
• Centralizes persisted enum tokens and surface limits.
• Rejects zero-norm embeddings before adapter dispatch.

File summaries

File	Description
src/usecases/vector_indexing.rs	Adds pre-adapter zero-norm rejection and tests.
src/usecases/retrieve.rs	Uses the shared domain surface enum.
src/policy/embedding_surface.rs	Simplifies builders and publishes surface limits.
src/models/vector/record.rs	Defines the five-field vector record.
src/models/vector/candidate_record.rs	Adopts the domain surface enum.
src/models/vector.rs	Updates exports and adds the shared fixture.
src/lib.rs	Exposes surface policy and test fixtures.
src/errors.rs	Adds the typed zero-norm error cause.
src/domain.rs	Centralizes enums and persisted token conversions.
src/api/types/retrieval.rs	Reuses the domain surface type.
src/api/types.rs	Removes the duplicate surface export.
src/adapters/qdrant/store.rs	Uses canonical tokens and reduced payloads.
src/adapters/qdrant/payload.rs	Implements the five-field manifest.
docs/design/database/vector_payload_design.md	Documents the new vector contract.
docs/design/database/schema_cheat_sheet.md	Updates the compact storage schema.
docs/design/database/README.md	Refreshes the vector-design summary.
docs/design/database/graph_schema_design.md	Updates cross-store responsibilities.
Cargo.toml	Adds the fixture feature and self dev-dependency.
Cargo.lock	Records the development dependency change.

Review details

• Files reviewed: 18/19 changed files
• Comments generated: 0
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🔵 Needs a closer look

The broad public and persistence-contract changes warrant final human review.

Review details

Suppressed comments (1)

docs/design/database/schema_cheat_sheet.md:56

• This says all provenance is hydrated from Oxigraph, but the preceding field list defines embedding_text as embedding provenance retained in Qdrant. Qualify this as source/graph provenance so the five-field contract is not self-contradictory.

These are the only Qdrant payload fields. Readable content, graph URI, object-specific state, relationships, lifecycle/currentness, ranking, timestamps, provenance, and raw references are hydrated from Oxigraph by `object_id`. Existing obsolete extra fields may remain on old points but readers ignore them.

• Files reviewed: 18/19 changed files
• Comments generated: 0 new
• Review effort level: Balanced

[Copilot]

🟢 Approval recommended

The reviewed changes have no unresolved approval-blocking issues.

Review details

• Files reviewed: 18/19 changed files
• Comments generated: 0 new
• Review effort level: Balanced