diff --git a/lending-poc/app/main.py b/lending-poc/app/main.py
index dc9c4d0..5b70c89 100644
--- a/lending-poc/app/main.py
+++ b/lending-poc/app/main.py
@@ -4,10 +4,10 @@
 from fastapi import FastAPI
 from sqlalchemy import text
 
-from app.api.cases import router as cases_router
 from app.api.health import router as health_router
 from app.config import logger, settings
 from app.database import async_session, engine
+from cross_document_validation.api import router as cases_router
 
 
 @asynccontextmanager
diff --git a/lending-poc/app/matching/__init__.py b/lending-poc/app/models/__init__.py
similarity index 100%
rename from lending-poc/app/matching/__init__.py
rename to lending-poc/app/models/__init__.py
diff --git a/lending-poc/app/services/__init__.py b/lending-poc/cross_document_validation/__init__.py
similarity index 100%
rename from lending-poc/app/services/__init__.py
rename to lending-poc/cross_document_validation/__init__.py
diff --git a/lending-poc/app/api/cases.py b/lending-poc/cross_document_validation/api.py
similarity index 76%
rename from lending-poc/app/api/cases.py
rename to lending-poc/cross_document_validation/api.py
index 4d32522..5b3feea 100644
--- a/lending-poc/app/api/cases.py
+++ b/lending-poc/cross_document_validation/api.py
@@ -2,11 +2,11 @@
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.database import get_db
-from app.schemas.case import CaseCreateRequest, CaseCreateResponse, ValidationResultOut
-from app.services.case_parsing import parse_case
-from app.services.persistence import save_pipeline_result
-from app.services.pipeline import run_pipeline
-from app.utils.json_safe import json_safe
+from cross_document_validation.schemas.case import CaseCreateRequest, CaseCreateResponse, ValidationResultOut
+from cross_document_validation.services.case_parsing import parse_case
+from cross_document_validation.services.persistence import save_pipeline_result
+from cross_document_validation.services.pipeline import run_pipeline
+from cross_document_validation.utils.json_safe import json_safe
 
 router = APIRouter(tags=["cases"])
 
diff --git a/lending-poc/cross_document_validation/matching/__init__.py b/lending-poc/cross_document_validation/matching/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/lending-poc/app/matching/embeddings.py b/lending-poc/cross_document_validation/matching/embeddings.py
similarity index 100%
rename from lending-poc/app/matching/embeddings.py
rename to lending-poc/cross_document_validation/matching/embeddings.py
diff --git a/lending-poc/app/matching/exact.py b/lending-poc/cross_document_validation/matching/exact.py
similarity index 100%
rename from lending-poc/app/matching/exact.py
rename to lending-poc/cross_document_validation/matching/exact.py
diff --git a/lending-poc/app/matching/fuzzy.py b/lending-poc/cross_document_validation/matching/fuzzy.py
similarity index 100%
rename from lending-poc/app/matching/fuzzy.py
rename to lending-poc/cross_document_validation/matching/fuzzy.py
diff --git a/lending-poc/cross_document_validation/schemas/__init__.py b/lending-poc/cross_document_validation/schemas/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/lending-poc/app/schemas/case.py b/lending-poc/cross_document_validation/schemas/case.py
similarity index 100%
rename from lending-poc/app/schemas/case.py
rename to lending-poc/cross_document_validation/schemas/case.py
diff --git a/lending-poc/cross_document_validation/services/__init__.py b/lending-poc/cross_document_validation/services/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/lending-poc/app/services/business_validation.py b/lending-poc/cross_document_validation/services/business_validation.py
similarity index 89%
rename from lending-poc/app/services/business_validation.py
rename to lending-poc/cross_document_validation/services/business_validation.py
index ed58b07..c048023 100644
--- a/lending-poc/app/services/business_validation.py
+++ b/lending-poc/cross_document_validation/services/business_validation.py
@@ -6,9 +6,9 @@
 from calendar import monthrange
 from datetime import date, timedelta
 
-from app.matching.fuzzy import employer_similarity
-from app.services import validation_config as cfg
-from app.services.dto import (
+from cross_document_validation.matching.fuzzy import employer_similarity
+from cross_document_validation.services import validation_config as cfg
+from cross_document_validation.services.dto import (
     BankStatementDoc,
     BankTransaction,
     CaseInput,
@@ -135,7 +135,12 @@ def _validate_salary_slip(
             score=0.0,
             document_id=slip.doc_id,
             failure_reason="no_matching_credit_in_window",
-            evidence={"window": window},
+            evidence={
+                "source_text": slip.salary_month.strftime("%B %Y"),
+                "target_text": None,
+                "match_type": "DATE_MATCH",
+                "window": window,
+            },
         )
 
     txn, score = selection
@@ -145,7 +150,12 @@ def _validate_salary_slip(
         passed=True,
         score=score,
         document_id=slip.doc_id,
-        evidence={"matched_transaction": txn},
+        evidence={
+            "source_text": slip.salary_month.strftime("%B %Y"),
+            "target_text": txn.txn_date.strftime("%Y-%m") if txn.txn_date else None,
+            "match_type": "DATE_MATCH",
+            "matched_transaction": txn,
+        },
     )
 
 
@@ -187,6 +197,11 @@ def _employer_match_for_slip(
         score=score,
         document_id=slip.doc_id,
         failure_reason=None if passed else "employer_narration_mismatch",
+        evidence={
+            "source_text": slip.employer_name,
+            "target_text": matched_txn.narration,
+            "match_type": "FUZZY",
+        },
     )
 
 
@@ -206,9 +221,12 @@ def _salary_credit_count(
         check_type=CheckType.SALARY_CREDIT_COUNT,
         passed=(matched_slips == total_slips),
         score=confidence_score,
+        document_id=bank_statement.doc_id,
         evidence={
+            "source_value": total_slips,
+            "target_value": matched_slips,
+            "match_type": "COUNT_MATCH",
             "stmt_duration": stmt_duration,
-            "no_of_matches": matched_slips,
             "total_slips": total_slips,
             "confidence_score": confidence_score,
         },
diff --git a/lending-poc/app/services/case_parsing.py b/lending-poc/cross_document_validation/services/case_parsing.py
similarity index 98%
rename from lending-poc/app/services/case_parsing.py
rename to lending-poc/cross_document_validation/services/case_parsing.py
index b8a0c33..d9c3a2b 100644
--- a/lending-poc/app/services/case_parsing.py
+++ b/lending-poc/cross_document_validation/services/case_parsing.py
@@ -5,7 +5,7 @@
 
 from datetime import date, datetime
 
-from app.services.dto import (
+from cross_document_validation.services.dto import (
     AadhaarDoc,
     AddressProofDoc,
     BankStatementDoc,
diff --git a/lending-poc/app/services/decision_engine.py b/lending-poc/cross_document_validation/services/decision_engine.py
similarity index 88%
rename from lending-poc/app/services/decision_engine.py
rename to lending-poc/cross_document_validation/services/decision_engine.py
index be8abd5..9b55ffd 100644
--- a/lending-poc/app/services/decision_engine.py
+++ b/lending-poc/cross_document_validation/services/decision_engine.py
@@ -1,7 +1,7 @@
 """Final PASS / FAIL / NEEDS_REVIEW logic."""
 
-from app.services import validation_config as cfg
-from app.services.dto import CheckType, Decision, DecisionResult, ScoreResult, ValidationResult
+from cross_document_validation.services import validation_config as cfg
+from cross_document_validation.services.dto import CheckType, Decision, DecisionResult, ScoreResult, ValidationResult
 
 MANDATORY_CHECK_TYPES = {CheckType.NAME, CheckType.AADHAAR, CheckType.PAN, CheckType.DOB}
 
diff --git a/lending-poc/app/services/dto.py b/lending-poc/cross_document_validation/services/dto.py
similarity index 100%
rename from lending-poc/app/services/dto.py
rename to lending-poc/cross_document_validation/services/dto.py
diff --git a/lending-poc/app/services/golden_record.py b/lending-poc/cross_document_validation/services/golden_record.py
similarity index 91%
rename from lending-poc/app/services/golden_record.py
rename to lending-poc/cross_document_validation/services/golden_record.py
index c676e81..9e47e2c 100644
--- a/lending-poc/app/services/golden_record.py
+++ b/lending-poc/cross_document_validation/services/golden_record.py
@@ -11,10 +11,10 @@
 rather than silently adopting an unrelated "fuller" name.
 """
 
-from app.matching.embeddings import get_address_embedding
-from app.matching.fuzzy import name_similarity
-from app.services import validation_config as cfg
-from app.services.dto import CaseInput, GoldenRecord
+from cross_document_validation.matching.embeddings import get_address_embedding
+from cross_document_validation.matching.fuzzy import name_similarity
+from cross_document_validation.services import validation_config as cfg
+from cross_document_validation.services.dto import CaseInput, GoldenRecord
 
 FULLER_NAME_RELATEDNESS_THRESHOLD = cfg.NAME_MATCH_THRESHOLD
 
diff --git a/lending-poc/app/services/identity_validation.py b/lending-poc/cross_document_validation/services/identity_validation.py
similarity index 79%
rename from lending-poc/app/services/identity_validation.py
rename to lending-poc/cross_document_validation/services/identity_validation.py
index 8b544a9..5e6eb71 100644
--- a/lending-poc/app/services/identity_validation.py
+++ b/lending-poc/cross_document_validation/services/identity_validation.py
@@ -3,10 +3,10 @@
 the Golden Record at all, regardless of why they're missing.
 """
 
-from app.matching import exact, fuzzy
-from app.matching.embeddings import address_similarity
-from app.services import validation_config as cfg
-from app.services.dto import CaseInput, CheckType, GoldenRecord, ValidationResult
+from cross_document_validation.matching import exact, fuzzy
+from cross_document_validation.matching.embeddings import address_similarity
+from cross_document_validation.services import validation_config as cfg
+from cross_document_validation.services.dto import CaseInput, CheckType, GoldenRecord, ValidationResult
 
 MANDATORY_GOLDEN_FIELDS = {
     CheckType.NAME: "name",
@@ -31,7 +31,9 @@ def check_mandatory_presence(golden: GoldenRecord) -> list[ValidationResult]:
     return results
 
 
-def _exact_result_to_validation(check_type: CheckType, outcome, document_id: str) -> ValidationResult:
+def _exact_result_to_validation(
+    check_type: CheckType, outcome, document_id: str, source_text, target_text
+) -> ValidationResult:
     passed = outcome.result == exact.MatchResult.MATCH
     score = 100.0 if passed else (50.0 if outcome.result == exact.MatchResult.INCONCLUSIVE else 0.0)
     return ValidationResult(
@@ -40,6 +42,7 @@ def _exact_result_to_validation(check_type: CheckType, outcome, document_id: str
         score=score,
         document_id=document_id,
         failure_reason=None if passed else outcome.reason,
+        evidence={"source_text": source_text, "target_text": target_text, "match_type": "EXACT"},
     )
 
 
@@ -64,6 +67,7 @@ def validate_document_against_golden(
                 score=score,
                 document_id=document_id,
                 failure_reason=None if passed else "name_below_threshold",
+                evidence={"source_text": golden.name, "target_text": doc_name, "match_type": "FUZZY"},
             )
         )
 
@@ -78,20 +82,31 @@ def validate_document_against_golden(
                 score=score,
                 document_id=document_id,
                 failure_reason=None if passed else "address_below_threshold",
+                evidence={"source_text": golden.address, "target_text": doc_address, "match_type": "SEMANTIC"},
             )
         )
 
     if doc_aadhaar is not None:
         outcome = exact.aadhaar_match(golden.aadhaar_number, doc_aadhaar)
-        results.append(_exact_result_to_validation(CheckType.AADHAAR, outcome, document_id))
+        results.append(
+            _exact_result_to_validation(
+                CheckType.AADHAAR, outcome, document_id, golden.aadhaar_number, doc_aadhaar
+            )
+        )
 
     if doc_pan is not None:
         outcome = exact.pan_match(golden.pan_number, doc_pan)
-        results.append(_exact_result_to_validation(CheckType.PAN, outcome, document_id))
+        results.append(
+            _exact_result_to_validation(CheckType.PAN, outcome, document_id, golden.pan_number, doc_pan)
+        )
 
     if doc_dob is not None:
         outcome = exact.dob_match(golden.date_of_birth, doc_dob)
-        results.append(_exact_result_to_validation(CheckType.DOB, outcome, document_id))
+        results.append(
+            _exact_result_to_validation(
+                CheckType.DOB, outcome, document_id, golden.date_of_birth, doc_dob
+            )
+        )
 
     return results
 
diff --git a/lending-poc/app/services/persistence.py b/lending-poc/cross_document_validation/services/persistence.py
similarity index 91%
rename from lending-poc/app/services/persistence.py
rename to lending-poc/cross_document_validation/services/persistence.py
index 6e41534..011a9a5 100644
--- a/lending-poc/app/services/persistence.py
+++ b/lending-poc/cross_document_validation/services/persistence.py
@@ -1,9 +1,9 @@
 """Persists one pipeline run (CaseInput + PipelineResult) to the database.
 
-The in-memory dataclasses in app.services.dto reference documents by their
-string doc_id (e.g. "AADHAAR", "SALARY_SLIP-0"). This module inserts the
-Document rows first and keeps a doc_id -> Document.id map so
-ValidationResult.document_id (also a doc_id string) can be resolved to the
+The in-memory dataclasses in cross_document_validation.services.dto reference
+documents by their string doc_id (e.g. "AADHAAR", "SALARY_SLIP-0"). This
+module inserts the Document rows first and keeps a doc_id -> Document.id map
+so ValidationResult.document_id (also a doc_id string) can be resolved to the
 real foreign key.
 """
 
@@ -11,9 +11,9 @@
 
 from sqlalchemy.ext.asyncio import AsyncSession
 
-from app.services.dto import CaseInput, Decision, DocType
-from app.services.dto import PipelineResult as PipelineResultDTO
-from app.utils.json_safe import json_safe
+from cross_document_validation.services.dto import CaseInput, Decision, DocType
+from cross_document_validation.services.dto import PipelineResult as PipelineResultDTO
+from cross_document_validation.utils.json_safe import json_safe
 from db.models.case import Case, CaseStatus
 from db.models.document import Document
 from db.models.golden_record import GoldenRecord as GoldenRecordModel
diff --git a/lending-poc/app/services/pipeline.py b/lending-poc/cross_document_validation/services/pipeline.py
similarity index 83%
rename from lending-poc/app/services/pipeline.py
rename to lending-poc/cross_document_validation/services/pipeline.py
index bdee270..78af125 100644
--- a/lending-poc/app/services/pipeline.py
+++ b/lending-poc/cross_document_validation/services/pipeline.py
@@ -2,10 +2,10 @@
 Decision in one call, with a simple in-memory audit log.
 """
 
-from app.services import business_validation, decision_engine, golden_record, scoring
-from app.services import validation_config as cfg
-from app.services.dto import CaseInput, Decision, DecisionResult, PipelineResult
-from app.services.identity_validation import run_identity_validation
+from cross_document_validation.services import business_validation, decision_engine, golden_record, scoring
+from cross_document_validation.services import validation_config as cfg
+from cross_document_validation.services.dto import CaseInput, Decision, DecisionResult, PipelineResult
+from cross_document_validation.services.identity_validation import run_identity_validation
 
 
 def run_pipeline(case: CaseInput) -> PipelineResult:
diff --git a/lending-poc/app/services/scoring.py b/lending-poc/cross_document_validation/services/scoring.py
similarity index 87%
rename from lending-poc/app/services/scoring.py
rename to lending-poc/cross_document_validation/services/scoring.py
index 6f79533..0e3f0eb 100644
--- a/lending-poc/app/services/scoring.py
+++ b/lending-poc/cross_document_validation/services/scoring.py
@@ -2,8 +2,8 @@
 
 from collections import defaultdict
 
-from app.services import validation_config as cfg
-from app.services.dto import ScoreResult, ValidationResult
+from cross_document_validation.services import validation_config as cfg
+from cross_document_validation.services.dto import ScoreResult, ValidationResult
 
 
 def compute_score(validation_results: list[ValidationResult]) -> ScoreResult:
diff --git a/lending-poc/app/services/validation_config.py b/lending-poc/cross_document_validation/services/validation_config.py
similarity index 100%
rename from lending-poc/app/services/validation_config.py
rename to lending-poc/cross_document_validation/services/validation_config.py
diff --git a/lending-poc/cross_document_validation/utils/__init__.py b/lending-poc/cross_document_validation/utils/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/lending-poc/app/utils/json_safe.py b/lending-poc/cross_document_validation/utils/json_safe.py
similarity index 100%
rename from lending-poc/app/utils/json_safe.py
rename to lending-poc/cross_document_validation/utils/json_safe.py
diff --git a/lending-poc/db/migrations/versions/0006_make_document_source_file_ref_nullable.py b/lending-poc/db/migrations/versions/0006_make_document_source_file_ref_nullable.py
deleted file mode 100644
index 3189c29..0000000
--- a/lending-poc/db/migrations/versions/0006_make_document_source_file_ref_nullable.py
+++ /dev/null
@@ -1,25 +0,0 @@
-"""make document source_file_ref nullable
-
-Revision ID: 0006_doc_source_ref_nullable
-Revises: 0005_add_validation_results
-Create Date: 2026-08-13 00:00:00.000000
-
-"""
-from typing import Sequence, Union
-
-from alembic import op
-import sqlalchemy as sa
-
-# revision identifiers, used by Alembic.
-revision: str = "0006_doc_source_ref_nullable"
-down_revision: Union[str, None] = "0005_add_validation_results"
-branch_labels: Union[str, Sequence[str], None] = None
-depends_on: Union[str, Sequence[str], None] = None
-
-
-def upgrade() -> None:
-    op.alter_column("documents", "source_file_ref", existing_type=sa.String(), nullable=True)
-
-
-def downgrade() -> None:
-    op.alter_column("documents", "source_file_ref", existing_type=sa.String(), nullable=False)
diff --git a/lending-poc/docs/cases_api.md b/lending-poc/docs/cases_api.md
index 9bca958..78873e1 100644
--- a/lending-poc/docs/cases_api.md
+++ b/lending-poc/docs/cases_api.md
@@ -92,14 +92,37 @@ Notes:
       "passed": true,
       "score": 94.5,
       "document_id": "PAN",
-      "evidence": null
+      "evidence": {
+        "source_text": "Sneha Sunil Lokhande",
+        "target_text": "Sneha Lokhande",
+        "match_type": "FUZZY"
+      }
     },
     {
       "check_type": "SALARY_DATE",
       "passed": false,
       "score": 0.0,
       "document_id": "SALARY_SLIP-3",
-      "evidence": { "window": ["2026-05-27", "2026-07-31"] }
+      "evidence": {
+        "source_text": "June 2026",
+        "target_text": null,
+        "match_type": "DATE_MATCH",
+        "window": ["2026-05-27", "2026-07-31"]
+      }
+    },
+    {
+      "check_type": "SALARY_CREDIT_COUNT",
+      "passed": false,
+      "score": 75.0,
+      "document_id": "BANK_STATEMENT",
+      "evidence": {
+        "source_value": 4,
+        "target_value": 3,
+        "match_type": "COUNT_MATCH",
+        "stmt_duration": { "start": "2026-04-01", "end": "2026-07-04" },
+        "total_slips": 4,
+        "confidence_score": 75.0
+      }
     }
   ]
 }
diff --git a/lending-poc/docs/features.md b/lending-poc/docs/features.md
index 4b5be5f..f7d7166 100644
--- a/lending-poc/docs/features.md
+++ b/lending-poc/docs/features.md
@@ -4,7 +4,7 @@ This document explains, in detail, what the validation pipeline behind `POST /ca
 
 ## 1. Pipeline overview
 
-`app/services/pipeline.py` orchestrates one end-to-end run in this order:
+`cross_document_validation/services/pipeline.py` orchestrates one end-to-end run in this order:
 
 ```
 INGEST -> required-document precheck -> GOLDEN RECORD -> IDENTITY VALIDATION
@@ -17,7 +17,7 @@ Every step appends a line to an in-memory `audit_log`, giving a readable trail o
 
 Before anything else runs, the pipeline checks that all of `AADHAAR`, `PAN`, `SALARY_SLIP`, `BANK_STATEMENT` are present (`validation_config.REQUIRED_DOCUMENT_TYPES`). If any are missing, the pipeline short-circuits with `decision=FAIL`, `overall_score=0.0`, and reasons like `MISSING_DOCUMENT:PAN` — no golden record or checks are computed.
 
-## 2. Golden Record (`app/services/golden_record.py`)
+## 2. Golden Record (`cross_document_validation/services/golden_record.py`)
 
 The Golden Record is the single trusted identity profile for the applicant, built by merging the KYC documents:
 
@@ -29,11 +29,11 @@ The Golden Record is the single trusted identity profile for the applicant, buil
   - If both have a name, and they're recognizably the same person (`fuzzy.name_similarity` >= `NAME_MATCH_THRESHOLD`, 85), the **fuller** name (more tokens) wins — e.g. "Sneha Sunil Lokhande" over "Sneha Lokhande" — because it carries strictly more identity information.
   - If the two names *aren't* recognizably related, Aadhaar stays authoritative and the mismatch is left for the NAME identity check to flag, rather than silently trusting an unrelated "fuller" name.
 - The chosen name is split into `first_name` / `middle_name` / `last_name`.
-- If an address was resolved, an address embedding is computed (`app.matching.embeddings.get_address_embedding`) and stored for later similarity checks.
+- If an address was resolved, an address embedding is computed (`cross_document_validation.matching.embeddings.get_address_embedding`) and stored for later similarity checks.
 
 Each golden field also records its `*_source` (which document it came from), useful for traceability.
 
-## 3. Identity Validation (`app/services/identity_validation.py`)
+## 3. Identity Validation (`cross_document_validation/services/identity_validation.py`)
 
 Two parts:
 
@@ -53,12 +53,12 @@ Every document that carries an identity field is compared against the Golden Rec
 | Each SALARY_SLIP | name |
 | BANK_STATEMENT | name |
 
-Matching strategies (`app/matching/`):
+Matching strategies (`cross_document_validation/matching/`):
 - **NAME** — fuzzy string similarity (`fuzzy.name_similarity`), handles reordering (surname-first), initials, and minor spelling differences. Passes at >= 85.
 - **ADDRESS** — embedding cosine similarity (`embeddings.address_similarity`), tolerant of differently-worded but equivalent addresses (e.g. "Apartment" vs "Flat", "MH" vs "Maharashtra"). Passes at >= 0.55 similarity (scored as similarity × 100).
-- **AADHAAR / PAN / DOB** — exact matching (`app.matching.exact`). Result is `MATCH` (score 100), `NO_MATCH` (score 0), or `INCONCLUSIVE` (score 50, e.g. one side missing/unparseable).
+- **AADHAAR / PAN / DOB** — exact matching (`cross_document_validation.matching.exact`). Result is `MATCH` (score 100), `NO_MATCH` (score 0), or `INCONCLUSIVE` (score 50, e.g. one side missing/unparseable).
 
-## 4. Business Validation (`app/services/business_validation.py`)
+## 4. Business Validation (`cross_document_validation/services/business_validation.py`)
 
 Verifies that declared income (salary slips) is corroborated by actual bank activity. Only runs if both salary slips and a bank statement are present.
 
@@ -81,7 +81,7 @@ Each slip's declared `employer_name` is compared — via fuzzy similarity — on
 
 An aggregate check: `matched_slips / total_slips × 100`. It passes only if *every* slip matched a transaction, but a partial match (e.g. 3 of 4 months) doesn't hard-fail the case — it only lowers this component's score, which feeds into the weighted overall score. Evidence includes the bank statement's observed date range and match counts.
 
-## 5. Scoring (`app/services/scoring.py`)
+## 5. Scoring (`cross_document_validation/services/scoring.py`)
 
 Given every `ValidationResult` produced above:
 1. Scores are grouped by `check_type` and averaged (e.g. if 4 salary slips each produced a `SALARY_DATE` score, they're averaged into one `SALARY_DATE` component score).
@@ -99,7 +99,7 @@ Given every `ValidationResult` produced above:
 
 3. The overall score is the weighted average, **renormalized over only the check types actually observed** in this case (so a case missing an optional check type doesn't get unfairly diluted by a zero for a check that never ran). Note `SALARY_DATE` itself isn't in the weight table — it gates whether a credit was found at all, but the weighted score is driven by `EMPLOYER` and `SALARY_CREDIT_COUNT`.
 
-## 6. Decision Engine (`app/services/decision_engine.py`)
+## 6. Decision Engine (`cross_document_validation/services/decision_engine.py`)
 
 Final decision logic, in priority order:
 
@@ -108,7 +108,7 @@ Final decision logic, in priority order:
 3. **FAIL** — if `overall_score < DECISION_FAIL_THRESHOLD` (60).
 4. **NEEDS_REVIEW** — anything in between (60–90). Reasons list every individual failing check as `<CHECK_TYPE>:<failure_reason>`.
 
-## 7. Persistence (`app/services/persistence.py`)
+## 7. Persistence (`cross_document_validation/services/persistence.py`)
 
 A successful pipeline run is persisted in a single DB transaction:
 - One `Case` row (`applicant_ref`, `status` derived from the decision: PASS/FAIL/NEEDS_REVIEW).
@@ -129,7 +129,7 @@ Document primary keys are resolved via an in-memory `doc_id -> Document.id` map
 | `SALARY_SLIP` | Declared income; multiple allowed per case (one per month) |
 | `BANK_STATEMENT` | Source of truth for actual salary credits |
 
-## 9. Configuration reference (`app/services/validation_config.py`)
+## 9. Configuration reference (`cross_document_validation/services/validation_config.py`)
 
 All thresholds/weights are centralized here as plain constants (intended to move into `app/config.py` / environment-driven settings as the app matures, without touching service logic):