# Phase 3 & Phase 4 — Cursor Test Cases (Locked) These tests validate behavioral correctness, phase boundaries, and determinism. They are written to be used by Cursor test agents or manually executed via CLI. --- ## Phase 3 — Evidence Emission Tests ### Test P3-01 — Zero Evidence Is Valid (Turn14) **Setup** - `turn14_items.item_json` contains no fitment arrays **Execution** - Run `turn14_evidence.php` **Expected** - `fitment_evidence`: 0 new rows - Script exits SUCCESS **Failure Conditions** - Any emitted rows - Any warnings or fallback behavior --- ### Test P3-02 — MotorState Evidence Emission **Setup** - `product_fitment_raw` contains ≥1 row with: - `extracted_json.fitments` array **Execution** - Run Phase 3 MotorState extractor **Expected** - One `fitment_evidence` row per vehicle tuple - Columns populated: - `product_id` - `year`, `make`, `model` - `source = 'motorstate'` - `source_confidence = 1.000` --- ### Test P3-03 — Raw Confidence Archival (Optional) **Setup** - `extracted_json.confidence = 0.873` **Execution** - Run Phase 3 extractor **Expected** - `raw_source_confidence = 0.873` - No change to row count - No branching behavior --- ### Test P3-04 — Phase Boundary Enforcement **Execution** - Static scan Phase 3 code **Expected** - No writes to tables other than `fitment_evidence` - No reads from Phase 4 tables --- ## Phase 4 — Validation Harness Tests ### Test P4-V01 — Validation PASS (Single Source) **Setup** - One `fitment_evidence` row per vehicle tuple **Execution** - Run `validate_reconciliation.php` **Expected** ```json { "status": "PASS" } ``` --- ### Test P4-V02 — Validation FAIL (Mixed Vehicle Tuple) **Setup** - Two evidence rows with same product_id but different model **Execution** - Run validation **Expected** ```json { "status": "FAIL", "error_code": "MIXED_VEHICLE_GROUP" } ``` --- ### Test P4-V03 — Validation FAIL (Invalid Confidence) **Setup** - `raw_source_confidence = 1.500` **Expected** - Validation FAIL - No reconciliation run allowed --- ## Phase 4 — Reconciliation Runner Tests ### Test P4-R01 — Single Source Resolution **Setup** - One evidence row **Execution** - Run reconciliation **Expected** - One `fitment_resolved` row - `final_confidence = source_multiplier` - `confidence_band = HIGH` --- ### Test P4-R02 — Multi-Source Agreement Boost **Setup** - Two evidence rows from different sources **Expected** - Confidence boosted - `confidence_band = CERTAIN` if ≥0.95 --- ### Test P4-R03 — Vendor Confidence Modifier Applied **Setup** - Two evidence rows - Vendor confidence values: 0.9 and 0.7 **Expected** - Modifier uses min = 0.7 - Confidence reduced accordingly --- ### Test P4-R04 — Idempotency **Execution** - Run reconciliation twice **Expected** - Second run produces no net changes - Same row hashes --- ## Phase Boundary Tests ### Test X-01 — No Backward Writes **Execution** - Run Phase 4 **Expected** - `fitment_evidence` unchanged --- ### Test X-02 — Override Safety **Setup** - Insert `fitment_overrides` row **Expected** - Phase 4 does NOT overwrite override - `override_flag = 1` --- ## Test Philosophy (Locked) - Evidence is immutable - Decisions are deterministic - Overrides are explicit - Tests fail fast If any test fails, stop the pipeline immediately.