# V2 Phase 3 Structured Review Backend Read Path

Date: 2026-07-01

## Scope

This slice moves backend consumers that derive display/search behavior from review results toward structured `review_risk_items` / `review_citations` first.

Legacy `review_tasks.risk_items` and `review_tasks.citations` remain as fallback data. This does not remove JSON columns or change the review task response contract.

## Changes

- Added `yuqei_contract_api.review_result_read_model`.
- `ContractActionContext` now reads latest review risk/repair state through the shared read model:
  - structured risk rows first
  - legacy JSON fallback only when structured rows are missing
- Unified search review-task indexing now reads risk and citation text through the same read model.
- Search rebuild preloads structured risks and citations for review tasks.
- Added regressions proving structured rows still drive:
  - `next_action` / `next_actions` when legacy `risk_items` is empty
  - unified review search when legacy `risk_items` and `citations` are empty

## Validation

```powershell
python -m pytest yuqei-contract-platform\services\contract-api\tests\test_contract_workflow.py::test_structured_review_results_drive_next_action_when_legacy_json_is_empty yuqei-contract-platform\services\contract-api\tests\test_contract_workflow.py::test_unified_search_returns_contract_file_and_review_task -q
python -m pytest yuqei-contract-platform\services\contract-api\tests\test_structured_review_results.py -q
python -m pytest yuqei-contract-platform\services\contract-api\tests\test_contract_workflow.py -q -k "structured_action_context or structured_review_results_drive_next_action or search_document_generation or unified_search_returns_contract_file_and_review_task or clause_repair"
```

Observed:

```text
..                                                                       [100%]
.........                                                                [100%]
......                                                                   [100%]
```

## Next Step

Continue the same pattern for clause repair orchestration: use structured risks to locate repair targets first, then keep legacy JSON mutation only as compatibility state until the write model is fully structured.