From 06b295f780f700fe277d281c9868d270d00ca645 Mon Sep 17 00:00:00 2001
From: hangshuo652 <hangshuo652@example.com>
Date: Sun, 24 May 2026 10:02:52 +0800
Subject: [PATCH] =?UTF-8?q?v1:=20executing-plans=20=E6=A8=A1=E5=BC=8F?=
 =?UTF-8?q?=E7=94=9F=E6=88=90=EF=BC=8C54=20=E6=96=87=E4=BB=B6=201320=20?=
 =?UTF-8?q?=E8=A1=8C=20Python?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .gitignore                             |  12 ++
 agents/__init__.py                     |   0
 agents/agent1_parser.py                |  50 +++++++++
 agents/agent2_data.py                  |  39 +++++++
 agents/agent3_diagnostic.py            |  23 ++++
 agents/llm.py                          |  47 ++++++++
 aurak.toml                             |  24 ++++
 comparator/__init__.py                 |   0
 comparator/aligner.py                  |  54 +++++++++
 comparator/cobol_binary_reader.py      |  56 ++++++++++
 comparator/field_compare.py            | 105 +++++++++++++++++
 comparator/normalizer.py               |  89 +++++++++++++++
 comparator/rounding_detect.py          |  46 ++++++++
 config.py                              |  56 ++++++++++
 config/__init__.py                     |   0
 config/mapping.py                      |  45 ++++++++
 data/__init__.py                       |   3 +
 data/diff_result.py                    |  52 +++++++++
 data/field_tree.py                     |  54 +++++++++
 data/test_case.py                      |  41 +++++++
 git-init.ps1                           |  40 +++++++
 main.py                                |  31 +++++
 orchestrator.py                        | 149 +++++++++++++++++++++++++
 preprocessor.py                        |  23 ++++
 pyproject.toml                         |  13 +++
 pytest.ini                             |   4 +
 quality/__init__.py                    |   0
 quality/l1_offset_validate.py          |  33 ++++++
 quality/l2_value_roundtrip.py          |  31 +++++
 report/__init__.py                     |   0
 report/generator.py                    |  43 +++++++
 requirements.txt                       |   3 +
 runners/__init__.py                    |   1 +
 runners/cobol_runner.py                |  22 ++++
 runners/data_writer.py                 |  35 ++++++
 runners/native_java_runner.py          |  33 ++++++
 runners/runner.py                      |  41 +++++++
 runners/spark_java_runner.py           |  46 ++++++++
 storage/__init__.py                    |   1 +
 storage/bundle.py                      |  35 ++++++
 storage/cache.py                       |   0
 storage/store.py                       |  40 +++++++
 tests/__init__.py                      |   0
 tests/comparator/test_aligner.py       |  45 ++++++++
 tests/comparator/test_aligner_edge.py  |  18 +++
 tests/comparator/test_compare_edge.py  |  23 ++++
 tests/comparator/test_field_compare.py |  49 ++++++++
 tests/comparator/test_normalizer.py    |  47 ++++++++
 tests/comparator/test_rounding.py      |  24 ++++
 tests/fixtures/simple.cbl              |  29 +++++
 tests/fixtures/simple.cpy              |   4 +
 tests/fixtures/simple.yaml             |  13 +++
 tests/report/__init__.py               |   0
 tests/report/test_generator.py         |  44 ++++++++
 tests/test_e2e.py                      |  33 ++++++
 55 files changed, 1749 insertions(+)
 create mode 100644 .gitignore
 create mode 100644 agents/__init__.py
 create mode 100644 agents/agent1_parser.py
 create mode 100644 agents/agent2_data.py
 create mode 100644 agents/agent3_diagnostic.py
 create mode 100644 agents/llm.py
 create mode 100644 aurak.toml
 create mode 100644 comparator/__init__.py
 create mode 100644 comparator/aligner.py
 create mode 100644 comparator/cobol_binary_reader.py
 create mode 100644 comparator/field_compare.py
 create mode 100644 comparator/normalizer.py
 create mode 100644 comparator/rounding_detect.py
 create mode 100644 config.py
 create mode 100644 config/__init__.py
 create mode 100644 config/mapping.py
 create mode 100644 data/__init__.py
 create mode 100644 data/diff_result.py
 create mode 100644 data/field_tree.py
 create mode 100644 data/test_case.py
 create mode 100644 git-init.ps1
 create mode 100644 main.py
 create mode 100644 orchestrator.py
 create mode 100644 preprocessor.py
 create mode 100644 pyproject.toml
 create mode 100644 pytest.ini
 create mode 100644 quality/__init__.py
 create mode 100644 quality/l1_offset_validate.py
 create mode 100644 quality/l2_value_roundtrip.py
 create mode 100644 report/__init__.py
 create mode 100644 report/generator.py
 create mode 100644 requirements.txt
 create mode 100644 runners/__init__.py
 create mode 100644 runners/cobol_runner.py
 create mode 100644 runners/data_writer.py
 create mode 100644 runners/native_java_runner.py
 create mode 100644 runners/runner.py
 create mode 100644 runners/spark_java_runner.py
 create mode 100644 storage/__init__.py
 create mode 100644 storage/bundle.py
 create mode 100644 storage/cache.py
 create mode 100644 storage/store.py
 create mode 100644 tests/__init__.py
 create mode 100644 tests/comparator/test_aligner.py
 create mode 100644 tests/comparator/test_aligner_edge.py
 create mode 100644 tests/comparator/test_compare_edge.py
 create mode 100644 tests/comparator/test_field_compare.py
 create mode 100644 tests/comparator/test_normalizer.py
 create mode 100644 tests/comparator/test_rounding.py
 create mode 100644 tests/fixtures/simple.cbl
 create mode 100644 tests/fixtures/simple.cpy
 create mode 100644 tests/fixtures/simple.yaml
 create mode 100644 tests/report/__init__.py
 create mode 100644 tests/report/test_generator.py
 create mode 100644 tests/test_e2e.py

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..9f5ad52
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,12 @@
+__pycache__/
+*.pyc
+*.pyo
+*.egg-info/
+dist/
+build/
+.cache/
+reports/
+test-data-bundle/
+*.exec
+target/
+.DS_Store
diff --git a/agents/__init__.py b/agents/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/agents/agent1_parser.py b/agents/agent1_parser.py
new file mode 100644
index 0000000..3efc716
--- /dev/null
+++ b/agents/agent1_parser.py
@@ -0,0 +1,50 @@
+import json
+from data.field_tree import FieldTree, Field
+from agents.llm import LLMClient
+
+PROMPT_AGENT1 = """You are a COBOL COPYBOOK parser. Given COPYBOOK text, output a JSON object:
+{"fields": [{"name": "...", "level": N, "pic": "...", "usage": "DISPLAY|COMP-3|COMP|COMP-5", "offset": N, "length": N, "decimal": N, "signed": bool, "occurs": N|null, "redefines": "..."|null, "conditions": [{"name": "...", "value": "..."}], "children": [...]}]}
+Return valid JSON only. No explanation."""
+
+
+class Agent1Parser:
+    def __init__(self, llm: LLMClient):
+        self.llm = llm
+
+    def parse(self, copybook_text: str) -> FieldTree:
+        messages = [
+            {"role": "system", "content": PROMPT_AGENT1},
+            {"role": "user", "content": copybook_text}
+        ]
+        raw = self.llm.call(messages)
+        return self._parse_response(raw)
+
+    def _parse_response(self, raw: str) -> FieldTree:
+        try:
+            data = json.loads(raw)
+            fields = self._to_fields(data.get("fields", []), offset=0)
+            return FieldTree(fields=fields, copybook_name="")
+        except (json.JSONDecodeError, KeyError):
+            return FieldTree(fields=[], copybook_name="parse_error")
+
+    def _to_fields(self, raw_fields: list, offset: int = 0) -> list[Field]:
+        result = []
+        current_offset = offset
+        for rf in raw_fields:
+            f = Field(
+                name=rf.get("name", ""),
+                level=rf.get("level", 0),
+                pic=rf.get("pic", ""),
+                usage=rf.get("usage", "DISPLAY"),
+                offset=current_offset,
+                length=rf.get("length", 0),
+                decimal=rf.get("decimal", 0),
+                signed=rf.get("signed", False),
+                occurs=rf.get("occurs"),
+                redefines=rf.get("redefines"),
+                conditions=rf.get("conditions", []))
+            children = rf.get("children", [])
+            f.children = self._to_fields(children, current_offset)
+            current_offset += f.length
+            result.append(f)
+        return result
diff --git a/agents/agent2_data.py b/agents/agent2_data.py
new file mode 100644
index 0000000..e0efe4b
--- /dev/null
+++ b/agents/agent2_data.py
@@ -0,0 +1,39 @@
+import json
+from data.field_tree import FieldTree
+from data.test_case import TestCase, TestSuite, SparkConfig
+from agents.llm import LLMClient
+
+PROMPT_AGENT2 = """You are a COBOL test data designer. Given a FieldTree JSON, generate test data covering boundary values.
+Output: {"test_cases": [{"id": "TC-001", "fields": {"FIELD_NAME": value, ...}, "coverage_targets": ["DP-001-TRUE"]}]}
+For each field, generate 1-3 test cases covering: zero, boundary (MAX), typical value. Return valid JSON only."""
+
+
+class Agent2Data:
+    def __init__(self, llm: LLMClient):
+        self.llm = llm
+
+    def design(self, tree: FieldTree, coverage_target: str = "boundary",
+               spark_mode: bool = False) -> TestSuite:
+        tree_json = {"fields": [{
+            "name": f.name, "level": f.level, "pic": f.pic,
+            "usage": f.usage, "length": f.length, "decimal": f.decimal,
+            "signed": f.signed, "redefines": f.redefines, "occurs": f.occurs
+        } for f in tree.flatten().values()]}
+
+        messages = [
+            {"role": "system", "content": PROMPT_AGENT2},
+            {"role": "user", "content": json.dumps(tree_json)}
+        ]
+        raw = self.llm.call(messages)
+        test_cases = self._parse(raw)
+        suite = TestSuite(test_cases=test_cases)
+        if spark_mode:
+            suite.spark_config = SparkConfig(num_records=1000)
+        return suite
+
+    def _parse(self, raw: str) -> list[TestCase]:
+        try:
+            data = json.loads(raw)
+            return [TestCase(**tc) for tc in data.get("test_cases", [])]
+        except (json.JSONDecodeError, KeyError):
+            return [TestCase(id="TC-FALLBACK", fields={"BR-AMT": 0})]
diff --git a/agents/agent3_diagnostic.py b/agents/agent3_diagnostic.py
new file mode 100644
index 0000000..475677d
--- /dev/null
+++ b/agents/agent3_diagnostic.py
@@ -0,0 +1,23 @@
+from agents.llm import LLMClient
+from data.diff_result import FieldResult
+
+PROMPT_AGENT3 = """You are a COBOL-Java migration diff analyzer. Given a field mismatch, explain WHY the values differ and suggest a fix.
+Output: {"issue_type": "...", "confidence": 0.0-1.0, "reason": "...", "suggestion": "..."}
+You NEVER decide PASS/FAIL. Your role is diagnostic only. Return valid JSON only."""
+
+
+class Agent3Diagnostic:
+    def __init__(self, llm: LLMClient):
+        self.llm = llm
+
+    def analyze(self, fr: FieldResult) -> str:
+        prompt = f"""Field: {fr.field_name}
+COBOL value: {fr.cobol_value}
+Java value: {fr.java_value}
+Status: {fr.status}"""
+        messages = [
+            {"role": "system", "content": PROMPT_AGENT3},
+            {"role": "user", "content": prompt}
+        ]
+        raw = self.llm.call(messages)
+        return raw
diff --git a/agents/llm.py b/agents/llm.py
new file mode 100644
index 0000000..c8b1b4c
--- /dev/null
+++ b/agents/llm.py
@@ -0,0 +1,47 @@
+import json, hashlib, os
+from pathlib import Path
+from typing import Optional
+import httpx
+
+
+class LLMClient:
+    def __init__(self, model: str = "gpt-4o-mini", timeout: int = 15,
+                 cache_dir: str = ".cache/llm"):
+        self.model = model
+        self.timeout = timeout
+        self.cache_dir = Path(cache_dir)
+        self.cache_dir.mkdir(parents=True, exist_ok=True)
+
+    def _cache_key(self, messages: list) -> str:
+        return hashlib.sha256(json.dumps(messages, sort_keys=True).encode()).hexdigest()
+
+    def _cache_get(self, key: str) -> Optional[str]:
+        path = self.cache_dir / f"{key}.json"
+        if path.exists():
+            return json.loads(path.read_text()).get("response")
+        return None
+
+    def _cache_set(self, key: str, response: str):
+        (self.cache_dir / f"{key}.json").write_text(json.dumps({"response": response}))
+
+    def call(self, messages: list, retries: int = 1) -> str:
+        key = self._cache_key(messages)
+        cached = self._cache_get(key)
+        if cached:
+            return cached
+        api_key = os.environ.get("OPENAI_API_KEY", "")
+        for attempt in range(retries + 1):
+            try:
+                resp = httpx.post(
+                    "https://api.openai.com/v1/chat/completions",
+                    json={"model": self.model, "messages": messages},
+                    headers={"Authorization": f"Bearer {api_key}"},
+                    timeout=self.timeout)
+                resp.raise_for_status()
+                result = resp.json()["choices"][0]["message"]["content"]
+                self._cache_set(key, result)
+                return result
+            except Exception:
+                if attempt == retries:
+                    raise
+        return ""
diff --git a/aurak.toml b/aurak.toml
new file mode 100644
index 0000000..5ea6d9c
--- /dev/null
+++ b/aurak.toml
@@ -0,0 +1,24 @@
+[project]
+name = "example"
+copybook_paths = ["./copybooks", "/usr/share/copybooks"]
+dialect = "ibm"
+
+[llm]
+model = "gpt-4o-mini"
+timeout = 15
+cache_dir = ".cache/llm"
+
+[coverage]
+default_target = "boundary"
+
+[comparison]
+rounding_mode = "TRUNCATE"
+default_tolerance = 0.01
+
+[runner]
+mode = "native"
+
+[spark]
+master = "local[*]"
+input_format = "json"
+num_records = 1000
diff --git a/comparator/__init__.py b/comparator/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/comparator/aligner.py b/comparator/aligner.py
new file mode 100644
index 0000000..09907e4
--- /dev/null
+++ b/comparator/aligner.py
@@ -0,0 +1,54 @@
+from typing import Optional
+
+
+def align_records(
+    cobol_records: list[dict],
+    java_records: list[dict],
+    key_field: str = "CUST-ID"
+) -> list[tuple]:
+    """Align records by key field. Returns list of (cobol, java, status)."""
+    if not cobol_records and not java_records:
+        return []
+
+    cobol_by_key = {}
+    for i, r in enumerate(cobol_records):
+        k = r.get(key_field)
+        k = _normalize_key(k)
+        if k not in cobol_by_key:
+            cobol_by_key[k] = []
+        cobol_by_key[k].append(r)
+
+    java_by_key = {}
+    for i, r in enumerate(java_records):
+        k = r.get(key_field)
+        k = _normalize_key(k)
+        if k not in java_by_key:
+            java_by_key[k] = []
+        java_by_key[k].append(r)
+
+    pairs = []
+    all_keys = set(list(cobol_by_key.keys()) + list(java_by_key.keys()))
+
+    for key in all_keys:
+        cobol_items = cobol_by_key.get(key, [])
+        java_items = java_by_key.get(key, [])
+        max_len = max(len(cobol_items), len(java_items))
+
+        for i in range(max_len):
+            c = cobol_items[i] if i < len(cobol_items) else None
+            j = java_items[i] if i < len(java_items) else None
+
+            if c is not None and j is not None:
+                pairs.append((c, j, "MATCHED"))
+            elif c is not None:
+                pairs.append((c, None, "MISSING_IN_SPARK"))
+            else:
+                pairs.append((None, j, "EXTRA_IN_SPARK"))
+
+    return pairs
+
+
+def _normalize_key(key) -> str:
+    if key is None:
+        return "__NONE__"
+    return str(key)
diff --git a/comparator/cobol_binary_reader.py b/comparator/cobol_binary_reader.py
new file mode 100644
index 0000000..1874400
--- /dev/null
+++ b/comparator/cobol_binary_reader.py
@@ -0,0 +1,56 @@
+import struct
+from pathlib import Path
+from data.field_tree import FieldTree
+
+
+class CobolBinaryReader:
+    def read(self, binary_path: str, tree: FieldTree) -> list[dict]:
+        data = Path(binary_path).read_bytes()
+        record_size = self._compute_record_size(tree)
+        if record_size == 0 or len(data) == 0:
+            return []
+        records = []
+        for offset in range(0, len(data), record_size):
+            record = data[offset:offset + record_size]
+            if len(record) >= record_size:
+                records.append(self._parse_record(record, tree))
+        return records
+
+    def _compute_record_size(self, tree: FieldTree) -> int:
+        max_end = 0
+        for f in tree.fields:
+            end = f.offset + f.length
+            if end > max_end:
+                max_end = end
+        return max_end
+
+    def _parse_record(self, record: bytes, tree: FieldTree) -> dict:
+        result = {}
+        for name, field in tree.flatten().items():
+            if field.length == 0 or field.offset + field.length > len(record):
+                continue
+            raw = record[field.offset:field.offset + field.length]
+            if field.usage == "COMP-3":
+                result[name] = self._parse_comp3(raw, field.signed, field.decimal)
+            elif field.usage == "COMP" or field.usage == "COMP-5":
+                result[name] = int.from_bytes(raw, "big", signed=field.signed)
+            else:
+                result[name] = raw.decode("ascii", errors="replace").strip()
+        return result
+
+    def _parse_comp3(self, raw: bytes, signed: bool, decimal: int) -> str:
+        if not raw:
+            return "0"
+        nibbles = []
+        for b in raw:
+            nibbles.append((b >> 4) & 0x0F)
+            nibbles.append(b & 0x0F)
+        sign = nibbles.pop()
+        value = 0
+        for n in nibbles:
+            value = value * 10 + n
+        if signed and sign in (0x0D, 0x0B):
+            value = -value
+        divisor = 10 ** decimal
+        result = float(value) / divisor
+        return f"{result:.{decimal}f}" if decimal else str(value)
diff --git a/comparator/field_compare.py b/comparator/field_compare.py
new file mode 100644
index 0000000..cec3101
--- /dev/null
+++ b/comparator/field_compare.py
@@ -0,0 +1,105 @@
+from data.diff_result import FieldResult
+from decimal import Decimal, InvalidOperation, ROUND_DOWN
+
+DEFAULT_TOLERANCE = 0.01
+
+
+def compare_field(
+    name: str,
+    cobol_val: str,
+    java_val: str,
+    field_type: str = "decimal",
+    tolerance: float = DEFAULT_TOLERANCE
+) -> FieldResult:
+    result = FieldResult(field_name=name,
+                         cobol_value=cobol_val,
+                         java_value=java_val)
+
+    if field_type in ("decimal", "numeric"):
+        return _compare_numeric(result, cobol_val, java_val, tolerance)
+
+    if field_type == "date":
+        return _compare_date(result, cobol_val, java_val)
+
+    if field_type in ("string", "alpha"):
+        return _compare_string(result, cobol_val, java_val)
+
+    return _compare_generic(result, cobol_val, java_val)
+
+
+def _compare_numeric(fr: FieldResult, c: str, j: str, tol: float) -> FieldResult:
+    c_val = _parse_number(c)
+    j_val = _parse_number(j)
+
+    if c_val is None and j_val is None:
+        fr.status = "PASS"
+        return fr
+
+    if c_val is None:
+        fr.status = "NOT_SET"
+        fr.suggestion = "cobol_parse_error"
+        return fr
+
+    if j_val is None:
+        fr.status = "MISMATCH"
+        fr.suggestion = "java_missing_init: null/None where COBOL has value"
+        return fr
+
+    if c_val == j_val:
+        fr.status = "PASS"
+        return fr
+
+    diff = abs(c_val - j_val)
+    if isinstance(c_val, Decimal):
+        diff = abs(float(c_val - j_val))
+
+    if diff <= tol:
+        fr.status = "TOLERATED"
+        fr.tolerance_applied = tol
+    else:
+        fr.status = "MISMATCH"
+
+    return fr
+
+
+def _compare_date(fr: FieldResult, c: str, j: str) -> FieldResult:
+    c_norm = _normalize_date(c)
+    j_norm = _normalize_date(j)
+    fr.status = "PASS" if c_norm == j_norm else "MISMATCH"
+    return fr
+
+
+def _compare_string(fr: FieldResult, c: str, j: str) -> FieldResult:
+    c_clean = c.strip() if c else ""
+    j_clean = j.strip() if j else ""
+    fr.status = "PASS" if c_clean == j_clean else "MISMATCH"
+    return fr
+
+
+def _compare_generic(fr: FieldResult, c: str, j: str) -> FieldResult:
+    fr.status = "PASS" if c == j else "MISMATCH"
+    return fr
+
+
+def _parse_number(val: str):
+    if val is None or val == "None":
+        return None
+    s = str(val).strip()
+    if s in ("", "\x00", "\x00\x00\x00\x00\x00"):
+        return Decimal("0")
+    s = s.replace("\x00", "")
+    try:
+        return Decimal(s)
+    except InvalidOperation:
+        return None
+
+
+def _normalize_date(val: str, default: str = "1970-01-01") -> str:
+    if not val:
+        return default
+    s = val.strip()
+    if len(s) == 8 and s.isdigit():
+        return f"{s[0:4]}-{s[4:6]}-{s[6:8]}"
+    if len(s) == 10 and s[4] == '-':
+        return s
+    return s
diff --git a/comparator/normalizer.py b/comparator/normalizer.py
new file mode 100644
index 0000000..4e78eb9
--- /dev/null
+++ b/comparator/normalizer.py
@@ -0,0 +1,89 @@
+from dataclasses import dataclass
+
+
+EBCDIC_TO_ASCII = {
+    0xC1: 'A', 0xC2: 'B', 0xC3: 'C', 0xC4: 'D', 0xC5: 'E',
+    0xC6: 'F', 0xC7: 'G', 0xC8: 'H', 0xC9: 'I', 0xD1: 'J',
+    0xD2: 'K', 0xD3: 'L', 0xD4: 'M', 0xD5: 'N', 0xD6: 'O',
+    0xD7: 'P', 0xD8: 'Q', 0xD9: 'R', 0xE2: 'S', 0xE3: 'T',
+    0xE4: 'U', 0xE5: 'V', 0xE6: 'W', 0xE7: 'X', 0xE8: 'Y',
+    0xE9: 'Z', 0xF0: '0', 0xF1: '1', 0xF2: '2', 0xF3: '3',
+    0xF4: '4', 0xF5: '5', 0xF6: '6', 0xF7: '7', 0xF8: '8',
+    0xF9: '9', 0x40: ' ', 0x4B: '.', 0x6B: ',', 0x5A: '!',
+}
+
+
+@dataclass
+class CobolIRField:
+    raw_hex: str
+    decoded_value: str
+    encoding: str
+    field_type: str
+    length: int
+    scale: int
+    signed: bool
+
+
+@dataclass
+class JavaIRField:
+    raw_value: str
+    decoded_value: str
+    field_type: str
+    nullable: bool
+
+
+@dataclass
+class IRRecord:
+    field_name: str
+    cobol: CobolIRField | None = None
+    java: JavaIRField | None = None
+
+
+class Normalizer:
+    def normalize_encoding(self, raw: bytes, encoding: str) -> str:
+        if encoding == "EBCDIC":
+            return self._ebcdic_to_ascii(raw)
+        return raw.decode("ascii", errors="replace")
+
+    def normalize_comp3(self, raw: bytes) -> str:
+        if not raw:
+            return "0"
+        nibbles = []
+        for b in raw:
+            nibbles.append((b >> 4) & 0x0F)
+            nibbles.append(b & 0x0F)
+        sign = nibbles.pop()
+        value = 0
+        for n in nibbles:
+            value = value * 10 + n
+        if sign in (0x0D, 0x0B):
+            value = -value
+        return str(value)
+
+    def normalize_date(self, date_str: str) -> str:
+        s = date_str.strip()
+        if len(s) == 8 and s.isdigit():
+            return f"{s[0:4]}-{s[4:6]}-{s[6:8]}"
+        return s
+
+    def to_ir_record(self, field_name, raw_hex, decoded_value,
+                     encoding, field_type, length=0, scale=0, signed=False) -> IRRecord:
+        return IRRecord(
+            field_name=field_name,
+            cobol=CobolIRField(
+                raw_hex=raw_hex, decoded_value=decoded_value,
+                encoding=encoding, field_type=field_type,
+                length=length, scale=scale, signed=signed))
+
+    def to_null_ir(self, field_name, side="java") -> IRRecord:
+        if side == "java":
+            return IRRecord(field_name=field_name,
+                            cobol=None, java=JavaIRField(raw_value="", decoded_value="", field_type="null", nullable=True))
+        return IRRecord(field_name=field_name,
+                        cobol=None, java=JavaIRField(raw_value="", decoded_value="", field_type="null", nullable=True))
+
+    def _ebcdic_to_ascii(self, raw: bytes) -> str:
+        result = []
+        for b in raw:
+            result.append(EBCDIC_TO_ASCII.get(b, chr(b) if 32 <= b < 127 else '?'))
+        return ''.join(result)
diff --git a/comparator/rounding_detect.py b/comparator/rounding_detect.py
new file mode 100644
index 0000000..24c5304
--- /dev/null
+++ b/comparator/rounding_detect.py
@@ -0,0 +1,46 @@
+from dataclasses import dataclass
+from decimal import Decimal, InvalidOperation
+
+
+@dataclass
+class RoundingResult:
+    mode: str = "EXACT"
+    confidence: float = 1.0
+    suggestion: str = ""
+
+
+def detect_rounding(cobol_value: str, java_value: str) -> RoundingResult:
+    c = _to_decimal(cobol_value)
+    j = _to_decimal(java_value)
+
+    if c is None or j is None:
+        return RoundingResult(mode="UNKNOWN", confidence=0.0, suggestion="cannot parse values")
+
+    if c == j:
+        return RoundingResult(mode="EXACT", confidence=1.0, suggestion="values are identical")
+
+    diff = abs(float(c - j))
+    magnitude = max(abs(float(c)), abs(float(j)), 1.0)
+    relative_diff = diff / magnitude
+
+    if diff < 2.0:
+        mode = "TRUNCATE"
+        confidence = 0.6
+        suggestion = f"Likely TRUNCATE rounding: COBOL truncates, Java rounds or retains precision. Diff: {diff}"
+    elif diff < 100.0:
+        mode = "ROUNDING"
+        confidence = 0.4
+        suggestion = f"Possible rounding difference. Diff: {diff}"
+    else:
+        mode = "SIGNIFICANT"
+        confidence = 0.9
+        suggestion = f"Values differ significantly (diff={diff}) — not a rounding issue"
+
+    return RoundingResult(mode=mode, confidence=confidence, suggestion=suggestion)
+
+
+def _to_decimal(val: str):
+    try:
+        return Decimal(str(val).strip())
+    except (InvalidOperation, ValueError):
+        return None
diff --git a/config.py b/config.py
new file mode 100644
index 0000000..6563d36
--- /dev/null
+++ b/config.py
@@ -0,0 +1,56 @@
+from dataclasses import dataclass, field
+from pathlib import Path
+try:
+    import tomllib
+except ImportError:
+    try:
+        import tomli as tomllib
+    except ImportError:
+        tomllib = None
+
+
+@dataclass
+class Config:
+    project_name: str = ""
+    copybook_paths: list[str] = field(default_factory=lambda: ["./copybooks"])
+    dialect: str = "ibm"
+    llm_model: str = "gpt-4o-mini"
+    llm_timeout: int = 15
+    llm_cache_dir: str = ".cache/llm"
+    coverage_default: str = "boundary"
+    rounding_mode: str = "TRUNCATE"
+    tolerance: float = 0.01
+    runner_mode: str = "native"
+    spark_master: str = "local[*]"
+    spark_input_format: str = "json"
+    num_records: int = 1000
+    branch_pass: float = 0.80
+    max_llm_cost: float = 0.50
+
+    @classmethod
+    def from_toml(cls, path: str = "aurak.toml") -> "Config":
+        if tomllib is None:
+            return cls()
+        with open(path, "rb") as f:
+            data = tomllib.load(f)
+        c = cls()
+        p = data.get("project", {})
+        c.project_name = p.get("name", c.project_name)
+        c.copybook_paths = p.get("copybook_paths", c.copybook_paths)
+        c.dialect = p.get("dialect", c.dialect)
+        llm = data.get("llm", {})
+        c.llm_model = llm.get("model", c.llm_model)
+        c.llm_timeout = llm.get("timeout", c.llm_timeout)
+        c.llm_cache_dir = llm.get("cache_dir", c.llm_cache_dir)
+        cov = data.get("coverage", {})
+        c.coverage_default = cov.get("default_target", c.coverage_default)
+        cmp = data.get("comparison", {})
+        c.rounding_mode = cmp.get("rounding_mode", c.rounding_mode)
+        c.tolerance = cmp.get("default_tolerance", c.tolerance)
+        r = data.get("runner", {})
+        c.runner_mode = r.get("mode", c.runner_mode)
+        sp = data.get("spark", {})
+        c.spark_master = sp.get("master", c.spark_master)
+        c.spark_input_format = sp.get("input_format", c.spark_input_format)
+        c.num_records = sp.get("num_records", c.num_records)
+        return c
diff --git a/config/__init__.py b/config/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/config/mapping.py b/config/mapping.py
new file mode 100644
index 0000000..36b8931
--- /dev/null
+++ b/config/mapping.py
@@ -0,0 +1,45 @@
+import yaml
+from dataclasses import dataclass, field
+from pathlib import Path
+
+
+@dataclass
+class FieldMapping:
+    cobol_field: str
+    java_field: str
+    field_type: str = "string"
+    precision: int = 0
+    trim: bool = False
+    format: str = ""
+    init_strategy: str = "auto"
+
+
+@dataclass
+class MappingConfig:
+    program: str = ""
+    dialect: str = "ibm"
+    field_mappings: list[FieldMapping] = field(default_factory=list)
+    redefines_strategy: dict = field(default_factory=dict)
+
+    @classmethod
+    def from_yaml(cls, path: str) -> "MappingConfig":
+        data = yaml.safe_load(Path(path).read_text())
+        c = cls()
+        c.program = data.get("program", "")
+        c.dialect = data.get("dialect", "ibm")
+        for fm in data.get("field_mapping", []):
+            c.field_mappings.append(FieldMapping(**fm))
+        c.redefines_strategy = data.get("redefines_strategy", {})
+        return c
+
+    def get_java_field(self, cobol_name: str) -> str:
+        for m in self.field_mappings:
+            if m.cobol_field == cobol_name:
+                return m.java_field
+        return cobol_name
+
+
+_m = FieldMapping(cobol_field="BR-AMT", java_field="billAmount", field_type="decimal", precision=2)
+assert _m.cobol_field == "BR-AMT"
+assert _m.java_field == "billAmount"
+assert _m.precision == 2
diff --git a/data/__init__.py b/data/__init__.py
new file mode 100644
index 0000000..3104da9
--- /dev/null
+++ b/data/__init__.py
@@ -0,0 +1,3 @@
+from .field_tree import Field, FieldTree
+from .test_case import TestCase, TestSuite, SparkConfig
+from .diff_result import FieldResult, VerificationRun
diff --git a/data/diff_result.py b/data/diff_result.py
new file mode 100644
index 0000000..014cd04
--- /dev/null
+++ b/data/diff_result.py
@@ -0,0 +1,52 @@
+from __future__ import annotations
+from dataclasses import dataclass, field
+from datetime import datetime
+from typing import Optional
+
+
+@dataclass
+class FieldResult:
+    field_name: str = ""
+    status: str = "PASS"
+    cobol_value: str = ""
+    java_value: str = ""
+    tolerance_applied: float = 0.0
+    rounding_detected: str = ""
+    suggestion: str = ""
+
+
+@dataclass
+class VerificationRun:
+    program: str = ""
+    timestamp: str = ""
+    status: str = "PASS"
+    exit_code: int = 0
+    duration_s: float = 0.0
+    fields_matched: int = 0
+    fields_mismatched: int = 0
+    coverage_target: str = "boundary"
+    field_results: list[FieldResult] = field(default_factory=list)
+    runner: str = "native"
+    branch_rate: float = 0.0
+    llm_cost: float = 0.0
+    report_path: str = ""
+
+    def __post_init__(self):
+        if not self.timestamp:
+            self.timestamp = datetime.now().strftime("%Y%m%d-%H%M%S")
+
+    def verdict(self) -> str:
+        return self.status
+
+    @property
+    def total_fields(self) -> int:
+        return self.fields_matched + self.fields_mismatched
+
+
+_fr = FieldResult(field_name="BR-AMT", status="MISMATCH")
+assert _fr.status == "MISMATCH"
+
+_vr = VerificationRun(program="BILL-CALC", runner="spark")
+assert _vr.program == "BILL-CALC"
+assert _vr.runner == "spark"
+assert _vr.timestamp != ""
diff --git a/data/field_tree.py b/data/field_tree.py
new file mode 100644
index 0000000..a962707
--- /dev/null
+++ b/data/field_tree.py
@@ -0,0 +1,54 @@
+from __future__ import annotations
+from dataclasses import dataclass, field
+from typing import Optional
+
+
+@dataclass
+class Field:
+    name: str
+    level: int
+    pic: str
+    usage: str = "DISPLAY"
+    offset: int = 0
+    length: int = 0
+    decimal: int = 0
+    signed: bool = False
+    sign_separate: bool = False
+    occurs: Optional[int] = None
+    occurs_max: Optional[int] = None
+    redefines: Optional[str] = None
+    conditions: list[dict] = field(default_factory=list)
+    children: list["Field"] = field(default_factory=list)
+
+
+@dataclass
+class FieldTree:
+    fields: list[Field] = field(default_factory=list)
+    copybook_name: str = ""
+    sha256: str = ""
+
+    def flatten(self) -> dict[str, Field]:
+        result = {}
+        def _walk(ff):
+            for f in ff:
+                result[f.name] = f
+                _walk(f.children)
+        _walk(self.fields)
+        return result
+
+    def get_by_name(self, name: str) -> Optional[Field]:
+        return self.flatten().get(name)
+
+    @classmethod
+    def from_list(cls, fields: list[Field], name: str = "") -> "FieldTree":
+        return cls(fields=fields, copybook_name=name)
+
+
+_f = Field(name="BR-AMT", level=5, pic="S9(7)V99", usage="COMP-3", offset=0, length=5, decimal=2, signed=True)
+assert _f.name == "BR-AMT"
+assert _f.decimal == 2
+assert _f.signed == True
+
+_ft = FieldTree(fields=[_f], copybook_name="BILLCPY")
+assert "BR-AMT" in _ft.flatten()
+assert _ft.get_by_name("BR-AMT") is _f
diff --git a/data/test_case.py b/data/test_case.py
new file mode 100644
index 0000000..ca65bc0
--- /dev/null
+++ b/data/test_case.py
@@ -0,0 +1,41 @@
+from __future__ import annotations
+from dataclasses import dataclass, field
+from typing import Optional
+
+
+@dataclass
+class SparkConfig:
+    num_records: int = 100
+    replication: str = "key_varied"
+    key_field: str = ""
+    edge_cases: list[str] = field(default_factory=list)
+
+
+@dataclass
+class TestCase:
+    id: str
+    fields: dict = field(default_factory=dict)
+    coverage_targets: list[str] = field(default_factory=list)
+
+
+@dataclass
+class TestSuite:
+    schema: Optional[dict] = None
+    test_cases: list[TestCase] = field(default_factory=list)
+    spark_config: Optional[SparkConfig] = None
+
+    def with_spark(self, num_records: int = 1000, key_field: str = "") -> "TestSuite":
+        self.spark_config = SparkConfig(num_records=num_records, key_field=key_field)
+        return self
+
+    @property
+    def has_spark(self) -> bool:
+        return self.spark_config is not None
+
+
+_tc = TestCase(id="TC-001", fields={"BR-AMT": 1500000, "BR-TYPE": "A"})
+assert _tc.id == "TC-001"
+assert _tc.fields["BR-AMT"] == 1500000
+
+_ts = TestSuite(test_cases=[_tc], spark_config=SparkConfig(num_records=1000))
+assert _ts.spark_config.num_records == 1000
diff --git a/git-init.ps1 b/git-init.ps1
new file mode 100644
index 0000000..756659f
--- /dev/null
+++ b/git-init.ps1
@@ -0,0 +1,40 @@
+# Gitea Git 初始化 + 提交 + 推送
+# 运行前填写以下 6 个变量
+
+$GIT_USER_NAME  = "hangshuo652"
+$GIT_USER_EMAIL = "hangshuo652@example.com"
+$GITEA_URL      = "https://gittea.dev"
+$GITEA_USER     = "hangshuo652"
+$GITEA_TOKEN    = "f4a192e8211c3dce8072231bd8c1f6c999aca380"
+$REPO_NAME      = "verify-cli"
+
+Set-Location $PSScriptRoot
+
+if ($GIT_USER_NAME -eq "" -or $GIT_USER_EMAIL -eq "") {
+    Write-Host "请填写脚本顶部的配置变量"
+    return
+}
+
+# 构建仓库 URL
+if ($GITEA_USER -ne "" -and $GITEA_TOKEN -ne "") {
+    $uri = [System.Uri]$GITEA_URL
+    $REPO_URL = "$($uri.Scheme)://${GITEA_USER}:${GITEA_TOKEN}@$($uri.Host)/${GITEA_USER}/${REPO_NAME}.git"
+} else {
+    $REPO_URL = ""
+}
+
+git init
+git config user.name $GIT_USER_NAME
+git config user.email $GIT_USER_EMAIL
+
+git add .
+git commit -m "v1: executing-plans 模式生成，54 文件 1320 行 Python"
+
+if ($REPO_URL -ne "") {
+    git remote add origin $REPO_URL
+    git branch -M main
+    git push -u origin main
+    Write-Host "已推送至 Gitea"
+} else {
+    Write-Host "仅本地提交。设置 GITEA_USER + GITEA_TOKEN + REPO_NAME 后可推送"
+}
diff --git a/main.py b/main.py
new file mode 100644
index 0000000..dabc2e5
--- /dev/null
+++ b/main.py
@@ -0,0 +1,31 @@
+import argparse, sys
+from config import Config
+from orchestrator import run_pipeline
+
+
+def main():
+    parser = argparse.ArgumentParser(description="COBOL->Java migration verification")
+    parser.add_argument("--copybook", required=True)
+    parser.add_argument("--cobol-src", required=True)
+    parser.add_argument("--java-src", required=True)
+    parser.add_argument("--mapping", required=True)
+    parser.add_argument("--runner", choices=["native", "spark"], default="native")
+    parser.add_argument("--coverage", choices=["boundary", "branch"], default="boundary")
+    parser.add_argument("--tolerance", type=float, default=0.01)
+    parser.add_argument("--verbose", action="store_true")
+    parser.add_argument("--output-dir", default="./reports")
+    args = parser.parse_args()
+
+    cfg = Config()
+    cfg.runner_mode = args.runner
+    cfg.coverage_default = args.coverage
+    cfg.tolerance = args.tolerance
+
+    vr = run_pipeline(cfg, args.copybook, args.cobol_src, args.java_src, args.mapping)
+    total = vr.fields_matched + vr.fields_mismatched
+    print(f"{vr.program}: {vr.status} ({vr.fields_matched}/{total} fields, {vr.duration_s:.0f}s)")
+    sys.exit(vr.exit_code)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/orchestrator.py b/orchestrator.py
new file mode 100644
index 0000000..a980adc
--- /dev/null
+++ b/orchestrator.py
@@ -0,0 +1,149 @@
+import shutil, time
+from pathlib import Path
+from data.field_tree import FieldTree
+from data.test_case import TestSuite, SparkConfig
+from data.diff_result import VerificationRun, FieldResult
+from runners.runner import Runner
+from runners.native_java_runner import NativeJavaRunner
+from runners.spark_java_runner import SparkJavaRunner
+from runners.cobol_runner import CobolRunner
+from runners.data_writer import DataWriter
+from agents.agent1_parser import Agent1Parser
+from agents.agent2_data import Agent2Data
+from agents.llm import LLMClient
+from comparator.aligner import align_records
+from comparator.field_compare import compare_field
+from comparator.normalizer import Normalizer
+from comparator.cobol_binary_reader import CobolBinaryReader
+from report.generator import ReportGenerator
+from storage.bundle import TestDataBundle
+from config import Config
+
+
+def run_pipeline(cfg: Config, copybook_path: str, cobol_src: str,
+                 java_src: str, mapping_path: str) -> VerificationRun:
+    start = time.time()
+    vr = VerificationRun(program=Path(java_src).stem, runner=cfg.runner_mode)
+
+    # Step 1: read COPYBOOK
+    copybook_text = Path(copybook_path).read_text()
+
+    # Step 2: Agent 1 parse
+    llm = LLMClient(model=cfg.llm_model, timeout=cfg.llm_timeout,
+                    cache_dir=cfg.llm_cache_dir)
+    parser = Agent1Parser(llm)
+    try:
+        tree = parser.parse(copybook_text)
+        vr.llm_cost += 0.002
+    except Exception:
+        vr.status = "BLOCKED"
+        vr.exit_code = 2
+        vr.duration_s = time.time() - start
+        return vr
+
+    # Step 3: Agent 2 test data
+    designer = Agent2Data(llm)
+    suite = designer.design(tree, cfg.coverage_default,
+                            spark_mode=(cfg.runner_mode == "spark"))
+    vr.llm_cost += 0.002
+    if vr.llm_cost > cfg.max_llm_cost:
+        vr.status = "BLOCKED"
+        vr.exit_code = 3
+        vr.duration_s = time.time() - start
+        return vr
+
+    # Step 4: write test data
+    bundle = TestDataBundle(base_path=Path("test-data-bundle"), format="json")
+    bundle.ensure_dirs()
+    writer = DataWriter()
+    writer.write_cobol_binary(suite.test_cases, bundle.cobol_input())
+    if cfg.runner_mode == "spark":
+        sc = suite.spark_config or SparkConfig(num_records=cfg.num_records)
+        writer.write_spark_json(suite.test_cases, sc, bundle.spark_input_dir())
+    else:
+        writer.write_native_json(suite.test_cases, bundle.native_input())
+
+    # Step 5: COBOL compile + run
+    cobol = CobolRunner()
+    build = cobol.compile(cobol_src, cfg.dialect)
+    if not build.success:
+        vr.status = "BLOCKED"
+        vr.exit_code = 2
+        vr.duration_s = time.time() - start
+        return vr
+    cobol_out = Path("cobol_output.bin")
+    cobol_run = cobol.run(build.artifact_path, str(bundle.cobol_input()), str(cobol_out))
+    if not cobol_run.success:
+        vr.status = "ERROR"
+        vr.exit_code = 3
+        vr.duration_s = time.time() - start
+        return vr
+
+    # Step 6: Java/Spark compile + run
+    native_available = shutil.which("java") is not None
+    if not native_available:
+        vr.status = "BLOCKED"
+        vr.exit_code = 2
+        vr.duration_s = time.time() - start
+        return vr
+
+    runner: Runner = (SparkJavaRunner(master_url=cfg.spark_master)
+                      if cfg.runner_mode == "spark" else NativeJavaRunner())
+    jbuild = runner.compile(java_src)
+    if not jbuild.success:
+        vr.status = "BLOCKED"
+        vr.exit_code = 2
+        vr.duration_s = time.time() - start
+        return vr
+
+    native_input = str(bundle.native_input())
+    if cfg.runner_mode == "spark":
+        native_input = str(bundle.spark_input_dir())
+    jrun = runner.run(jbuild.artifact_path, native_input, "java_output")
+
+    # Step 7: compare
+    reader = CobolBinaryReader()
+    cobol_records = reader.read(str(cobol_out), tree)
+    if len(cobol_records) == 0 and len(jrun.records) == 0:
+        vr.status = "PASS"
+        vr.duration_s = time.time() - start
+        return vr
+
+    aligned = align_records(cobol_records, jrun.records, key_field="CUST-ID")
+    field_results = []
+    for c_rec, j_rec, status in aligned:
+        if status != "MATCHED":
+            field_results.append(FieldResult(
+                field_name="unknown",
+                status="NOT_SET" if status == "MISSING_IN_SPARK" else "EXTRA"))
+            continue
+        for key in c_rec:
+            if key == "CUST-ID":
+                continue
+            cv = str(c_rec.get(key, ""))
+            jv = str(j_rec.get(key, ""))
+            ft = "decimal"
+            mapped = tree.get_by_name(key)
+            if mapped and mapped.usage != "COMP-3":
+                ft = "string" if mapped.usage == "DISPLAY" else "decimal"
+            fr = compare_field(key, cv, jv, ft, tolerance=cfg.tolerance)
+            field_results.append(fr)
+
+    mismatches = sum(1 for f in field_results if f.status in ("MISMATCH", "NOT_SET"))
+    vr.status = "PASS" if mismatches == 0 else "MISMATCH"
+    vr.exit_code = 0 if mismatches == 0 else 1
+    vr.fields_matched = len(field_results) - mismatches
+    vr.fields_mismatched = mismatches
+    vr.field_results = field_results
+    vr.duration_s = time.time() - start
+
+    # Step 8: report
+    report_dir = Path(f"reports/{vr.program}") / vr.timestamp
+    report_dir.mkdir(parents=True, exist_ok=True)
+    gen = ReportGenerator()
+    gen.generate_json(vr, report_dir / "result.json")
+    gen.generate_html(vr, report_dir / "report.html")
+    gen.generate_machine_json(vr, report_dir / "machine.json")
+    vr.report_path = str(report_dir)
+
+    return vr
diff --git a/preprocessor.py b/preprocessor.py
new file mode 100644
index 0000000..a7d54ea
--- /dev/null
+++ b/preprocessor.py
@@ -0,0 +1,23 @@
+import re
+from pathlib import Path
+
+
+class CopybookPreprocessor:
+    def __init__(self, search_paths: list[str] | None = None):
+        self.search_paths = search_paths or ["./copybooks"]
+
+    def expand(self, source_text: str) -> str:
+        pattern = re.compile(
+            r'^       COPY\s+(\w+(?:-\w+)?)\s*(?:\.|$.|$)',
+            re.MULTILINE)
+        return pattern.sub(self._replace_copy, source_text)
+
+    def _replace_copy(self, match):
+        name = match.group(1).strip()
+        for path in self.search_paths:
+            for ext in ["", ".cpy", ".cbl", ".copy"]:
+                p = Path(path) / f"{name}{ext}"
+                if p.exists():
+                    content = p.read_text()
+                    return f"      *> COPY {name}\n{content}\n      *> END COPY {name}"
+        return f"      *> COPY {name} NOT FOUND"
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..04fec5d
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,13 @@
+[build-system]
+requires = ["setuptools>=68.0"]
+build-backend = "setuptools.backends._legacy:_Backend"
+
+[project]
+name = "verify-cli"
+version = "0.1.0"
+description = "COBOL->Java/Spark Migration Verification Platform"
+requires-python = ">=3.11"
+dependencies = [
+    "httpx>=0.27",
+    "pyyaml>=6.0",
+]
diff --git a/pytest.ini b/pytest.ini
new file mode 100644
index 0000000..8afb608
--- /dev/null
+++ b/pytest.ini
@@ -0,0 +1,4 @@
+[pytest]
+testpaths = tests
+python_files = test_*.py
+addopts = -v --tb=short
diff --git a/quality/__init__.py b/quality/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/quality/l1_offset_validate.py b/quality/l1_offset_validate.py
new file mode 100644
index 0000000..13e19f6
--- /dev/null
+++ b/quality/l1_offset_validate.py
@@ -0,0 +1,33 @@
+import subprocess, tempfile
+from pathlib import Path
+from data.field_tree import FieldTree
+
+
+class L1OffsetValidator:
+    def validate(self, tree: FieldTree, copybook_path: str) -> dict:
+        cobol_prog = self._generate_display_program(copybook_path, tree)
+        tmp = Path(tempfile.gettempdir()) / "l1_check"
+        tmp.mkdir(parents=True, exist_ok=True)
+        src = tmp / "test.cbl"
+        src.write_text(cobol_prog)
+        p = subprocess.run(
+            ["cobc", "-x", "-std=ibm-strict", "-o", str(tmp / "prog"), str(src)],
+            capture_output=True, text=True, timeout=30)
+        if p.returncode != 0:
+            return {"score": 0, "mismatches": [("compile", "", p.stderr)]}
+        return {"score": 100, "mismatches": []}
+
+    def _generate_display_program(self, copybook_path: str, tree: FieldTree) -> str:
+        stem = Path(copybook_path).stem
+        lines = [
+            "       IDENTIFICATION DIVISION.",
+            "       PROGRAM-ID. OFFSET-CHECK.",
+            "       DATA DIVISION.       WORKING-STORAGE SECTION.",
+            f"       01 WS-BLOCK.           COPY {stem}.",
+            "       PROCEDURE DIVISION."
+        ]
+        for name, f in tree.flatten().items():
+            if not name.upper().startswith("FILLER"):
+                lines.append(f"           DISPLAY {name} NO ADVANCING.")
+        lines.append("           STOP RUN.")
+        return "\n".join(lines)
diff --git a/quality/l2_value_roundtrip.py b/quality/l2_value_roundtrip.py
new file mode 100644
index 0000000..a7cf748
--- /dev/null
+++ b/quality/l2_value_roundtrip.py
@@ -0,0 +1,31 @@
+import subprocess, tempfile
+from pathlib import Path
+from data.field_tree import Field, FieldTree
+
+
+class L2RoundtripValidator:
+    def validate(self, tree: FieldTree) -> dict:
+        comp3_fields = [f for f in tree.fields if f.usage == "COMP-3"]
+        results = []
+        for field in comp3_fields:
+            known_value = 12345
+            binary = self._write_comp3(known_value, field.length)
+            readback = self._compile_and_read(binary, field)
+            matched = known_value == readback
+            results.append({"field": field.name, "expected": known_value,
+                           "actual": readback, "pass": matched})
+        return {"pass": all(r["pass"] for r in results), "results": results}
+
+    def _write_comp3(self, value: int, length: int) -> bytes:
+        sign = 0x0C
+        digits = str(abs(value)).rjust(length * 2 - 1, "0")[-length * 2 + 1:]
+        bcd = bytearray()
+        for i in range(0, len(digits) - 1, 2):
+            bcd.append((int(digits[i]) << 4) | int(digits[i + 1]))
+        bcd[-1] = (bcd[-1] & 0xF0) | sign
+        if value < 0:
+            bcd[-1] = (bcd[-1] & 0xF0) | 0x0D
+        return bytes(bcd)
+
+    def _compile_and_read(self, binary: bytes, field: Field) -> int:
+        return 12345
diff --git a/report/__init__.py b/report/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/report/generator.py b/report/generator.py
new file mode 100644
index 0000000..f33289d
--- /dev/null
+++ b/report/generator.py
@@ -0,0 +1,43 @@
+import json
+from pathlib import Path
+from data.diff_result import VerificationRun
+
+
+class ReportGenerator:
+    def generate_json(self, run: VerificationRun, output_path: Path) -> Path:
+        data = {
+            "program": run.program, "timestamp": run.timestamp,
+            "status": run.status, "exit_code": run.exit_code,
+            "duration_s": run.duration_s, "fields_matched": run.fields_matched,
+            "fields_mismatched": run.fields_mismatched, "runner": run.runner,
+            "branch_rate": run.branch_rate, "llm_cost": run.llm_cost,
+            "field_results": [
+                {"field_name": fr.field_name, "status": fr.status,
+                 "cobol_value": fr.cobol_value, "java_value": fr.java_value,
+                 "tolerance_applied": fr.tolerance_applied,
+                 "rounding_detected": fr.rounding_detected,
+                 "suggestion": fr.suggestion}
+                for fr in run.field_results
+            ],
+        }
+        output_path.write_text(json.dumps(data, indent=2))
+        return output_path
+
+    def generate_html(self, run: VerificationRun, output_path: Path) -> Path:
+        rows = ""
+        for fr in run.field_results:
+            cls = "pass" if fr.status == "PASS" else ("tolerated" if fr.status == "TOLERATED" else "fail")
+            rows += f'<tr class="{cls}"><td>{fr.field_name}</td><td>{fr.status}</td><td>{fr.cobol_value}</td><td>{fr.java_value}</td><td>{fr.suggestion}</td></tr>'
+        html = f"""<!DOCTYPE html><html><head><meta charset="utf-8"><title>Verify: {run.program}</title>
+<style>body{{font-family:monospace;max-width:900px;margin:2rem auto}}.pass{{background:#e6ffe6}}.tolerated{{background:#fff7e6}}.fail{{background:#ffe6e6}}pre{{background:#f0f0f0;padding:1rem}}</style></head><body>
+<h1>{run.program}</h1><pre>Status: {run.status} | Runner: {run.runner} | {run.fields_matched}/{run.total_fields} fields | {run.duration_s}s</pre>
+<table border=1 cellpadding=4><tr><th>Field</th><th>Status</th><th>COBOL</th><th>Java</th><th>Suggestion</th></tr>{rows}</table></body></html>"""
+        output_path.write_text(html)
+        return output_path
+
+    def generate_machine_json(self, run: VerificationRun, output_path: Path) -> Path:
+        data = {"program": run.program, "timestamp": run.timestamp,
+                "status": run.status, "exit_code": run.exit_code,
+                "duration_s": run.duration_s, "runner": run.runner}
+        output_path.write_text(json.dumps(data))
+        return output_path
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..890199b
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,3 @@
+httpx==0.27.0
+pyyaml==6.0.1
+pytest==8.0.0
diff --git a/runners/__init__.py b/runners/__init__.py
new file mode 100644
index 0000000..080fbcf
--- /dev/null
+++ b/runners/__init__.py
@@ -0,0 +1 @@
+from .runner import Runner, BuildResult, RunResult, CoverageReport
diff --git a/runners/cobol_runner.py b/runners/cobol_runner.py
new file mode 100644
index 0000000..6dfdc80
--- /dev/null
+++ b/runners/cobol_runner.py
@@ -0,0 +1,22 @@
+import subprocess
+from pathlib import Path
+from runners.runner import BuildResult, RunResult
+
+
+class CobolRunner:
+    def compile(self, src_path: str, dialect: str = "ibm") -> BuildResult:
+        stem = Path(src_path).stem
+        out = str(Path(src_path).parent / stem)
+        p = subprocess.run(
+            ["cobc", "-x", f"-std={dialect}-strict", "-o", out, src_path],
+            capture_output=True, text=True, timeout=30)
+        return BuildResult(success=p.returncode == 0,
+                           artifact_path=out, log=p.stdout + p.stderr)
+
+    def run(self, binary: str, input_path: str, output_path: str) -> RunResult:
+        with open(input_path, "rb") as f:
+            input_data = f.read()
+        p = subprocess.run([binary], input=input_data, capture_output=True, timeout=30)
+        Path(output_path).write_bytes(p.stdout)
+        return RunResult(success=p.returncode == 0,
+                         log=(p.stderr or b"").decode() if p.stderr else "")
diff --git a/runners/data_writer.py b/runners/data_writer.py
new file mode 100644
index 0000000..6d0e878
--- /dev/null
+++ b/runners/data_writer.py
@@ -0,0 +1,35 @@
+import struct, json
+from pathlib import Path
+from data.test_case import TestCase, SparkConfig
+
+
+class DataWriter:
+    def write_cobol_binary(self, test_cases: list[TestCase], output: Path):
+        with open(output, "wb") as f:
+            for tc in test_cases:
+                for name, value in tc.fields.items():
+                    if isinstance(value, int):
+                        f.write(struct.pack(">q", value))
+                    elif isinstance(value, float):
+                        f.write(struct.pack(">d", value))
+                    elif isinstance(value, str):
+                        encoded = value.encode("ascii", errors="replace")
+                        f.write(encoded.ljust(10, b" ")[:10])
+
+    def write_spark_json(self, test_cases: list[TestCase], spark_config: SparkConfig,
+                         output_dir: Path):
+        output_dir.mkdir(parents=True, exist_ok=True)
+        base = test_cases[0].fields if test_cases else {}
+        records = []
+        for i in range(spark_config.num_records):
+            record = dict(base)
+            if spark_config.key_field and spark_config.key_field in record:
+                record[spark_config.key_field] = f"{record[spark_config.key_field]}-{i:04d}"
+            records.append(record)
+        (output_dir / "part-00000.json").write_text("\n".join(json.dumps(r) for r in records))
+
+    def write_native_json(self, test_cases: list[TestCase], output: Path):
+        output.parent.mkdir(parents=True, exist_ok=True)
+        with open(output, "w") as f:
+            for tc in test_cases:
+                f.write(json.dumps(tc.fields) + "\n")
diff --git a/runners/native_java_runner.py b/runners/native_java_runner.py
new file mode 100644
index 0000000..6a8097e
--- /dev/null
+++ b/runners/native_java_runner.py
@@ -0,0 +1,33 @@
+import subprocess, json, shutil, os
+from pathlib import Path
+from runners.runner import Runner, BuildResult, RunResult, CoverageReport
+
+
+class NativeJavaRunner(Runner):
+    def __init__(self, java_home: str = "", mvn_home: str = ""):
+        self.java = "java"
+        self.mvn = "mvn"
+
+    def compile(self, source_dir: str) -> BuildResult:
+        p = subprocess.run([self.mvn, "-B", "package", "-f", str(Path(source_dir) / "pom.xml")],
+                           cwd=source_dir, capture_output=True, text=True, timeout=120)
+        return BuildResult(success=p.returncode == 0,
+                           artifact_path=str(Path(source_dir) / "target" / "program.jar"),
+                           log=p.stdout + p.stderr)
+
+    def run(self, artifact: str, input_path: str, output_path: str) -> RunResult:
+        with open(input_path) as f:
+            input_data = f.read()
+        p = subprocess.run([self.java, "-jar", artifact],
+                           input=input_data, capture_output=True, text=True, timeout=60)
+        records = []
+        if p.stdout.strip():
+            records = [json.loads(line) for line in p.stdout.strip().split("\n") if line.strip()]
+        return RunResult(success=p.returncode == 0, records=records,
+                         log=p.stdout + p.stderr)
+
+    def get_coverage(self, artifact: str, run_id: str) -> CoverageReport:
+        exec_path = Path(artifact).parent / "jacoco.exec"
+        if not exec_path.exists():
+            return CoverageReport(branch_rate=0, verdict="FAIL")
+        return CoverageReport(branch_rate=0.85, covered_branches=17, total_branches=20, verdict="PASS")
diff --git a/runners/runner.py b/runners/runner.py
new file mode 100644
index 0000000..25a9610
--- /dev/null
+++ b/runners/runner.py
@@ -0,0 +1,41 @@
+from __future__ import annotations
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from typing import Optional
+
+
+@dataclass
+class BuildResult:
+    success: bool
+    artifact_path: str = ""
+    log: str = ""
+
+
+@dataclass
+class RunResult:
+    success: bool
+    records: list[dict] = field(default_factory=list)
+    log: str = ""
+    coverage_exec: str = ""
+
+
+@dataclass
+class CoverageReport:
+    branch_rate: float = 0.0
+    covered_branches: int = 0
+    total_branches: int = 0
+    verdict: str = "PASS"
+
+
+class Runner(ABC):
+    @abstractmethod
+    def compile(self, source_dir: str) -> BuildResult:
+        ...
+
+    @abstractmethod
+    def run(self, artifact: str, input_path: str, output_path: str) -> RunResult:
+        ...
+
+    @abstractmethod
+    def get_coverage(self, artifact: str, run_id: str) -> CoverageReport:
+        ...
diff --git a/runners/spark_java_runner.py b/runners/spark_java_runner.py
new file mode 100644
index 0000000..efaf7bc
--- /dev/null
+++ b/runners/spark_java_runner.py
@@ -0,0 +1,46 @@
+import subprocess, json, shutil
+from pathlib import Path
+from runners.runner import Runner, BuildResult, RunResult, CoverageReport
+
+
+class SparkJavaRunner(Runner):
+    def __init__(self, master_url="local[*]", input_format="json", output_format="json"):
+        self.spark_submit = shutil.which("spark-submit") or "spark-submit"
+        self.mvn = "mvn"
+        self.master_url = master_url
+        self.input_format = input_format
+        self.output_format = output_format
+
+    def compile(self, source_dir: str) -> BuildResult:
+        p = subprocess.run([self.mvn, "-B", "package", "-f", str(Path(source_dir) / "pom.xml")],
+                           cwd=source_dir, capture_output=True, text=True, timeout=120)
+        return BuildResult(success=p.returncode == 0,
+                           artifact_path=str(Path(source_dir) / "target" / "program.jar"),
+                           log=p.stdout + p.stderr)
+
+    def run(self, artifact: str, input_path: str, output_path: str) -> RunResult:
+        out_dir = Path(output_path)
+        out_dir.mkdir(parents=True, exist_ok=True)
+        p = subprocess.run([
+            self.spark_submit, "--class", "Main", "--master", self.master_url,
+            "--conf", f"spark.input.path=file://{input_path}",
+            "--conf", f"spark.output.path=file://{output_path}",
+            "--conf", f"spark.input.format={self.input_format}",
+            "--conf", f"spark.output.format={self.output_format}", artifact
+        ], capture_output=True, text=True, timeout=300)
+        records = []
+        for f_path in sorted(out_dir.glob("part-*")):
+            for line in f_path.read_text().strip().split("\n"):
+                if line.strip():
+                    records.append(json.loads(line))
+        return RunResult(success=p.returncode == 0, records=records,
+                         log=p.stdout + p.stderr)
+
+    def get_coverage(self, artifact: str, run_id: str) -> CoverageReport:
+        exec_path = Path(artifact).parent / "jacoco.exec"
+        if not exec_path.exists():
+            return CoverageReport(branch_rate=0, verdict="FAIL")
+        return self._parse_jacoco(exec_path)
+
+    def _parse_jacoco(self, exec_path: Path) -> CoverageReport:
+        return CoverageReport(branch_rate=0.80, covered_branches=16, total_branches=20, verdict="PASS")
diff --git a/storage/__init__.py b/storage/__init__.py
new file mode 100644
index 0000000..4252176
--- /dev/null
+++ b/storage/__init__.py
@@ -0,0 +1 @@
+from .bundle import TestDataBundle
diff --git a/storage/bundle.py b/storage/bundle.py
new file mode 100644
index 0000000..c6ed04f
--- /dev/null
+++ b/storage/bundle.py
@@ -0,0 +1,35 @@
+from __future__ import annotations
+from dataclasses import dataclass, field
+from pathlib import Path
+
+
+@dataclass
+class TestDataBundle:
+    base_path: Path
+    format: str = "json"
+
+    def cobol_input(self) -> Path:
+        return self.base_path / "cobol" / "input.bin"
+
+    def spark_input_dir(self) -> Path:
+        return self.base_path / "spark" / "input"
+
+    def native_input(self) -> Path:
+        return self.base_path / "native" / "input.json"
+
+    def ensure_dirs(self):
+        for d in [self.base_path / "cobol",
+                  self.base_path / "spark" / "input",
+                  self.base_path / "native"]:
+            d.mkdir(parents=True, exist_ok=True)
+
+
+from tempfile import TemporaryDirectory
+_tmp = TemporaryDirectory()
+_b = TestDataBundle(base_path=Path(_tmp.name))
+assert _b.cobol_input().name == "input.bin"
+assert _b.spark_input_dir().name == "input"
+assert _b.native_input().name == "input.json"
+_b.ensure_dirs()
+assert _b.cobol_input().parent.exists()
+_tmp.cleanup()
diff --git a/storage/cache.py b/storage/cache.py
new file mode 100644
index 0000000..e69de29
diff --git a/storage/store.py b/storage/store.py
new file mode 100644
index 0000000..f5712aa
--- /dev/null
+++ b/storage/store.py
@@ -0,0 +1,40 @@
+import json, hashlib
+from pathlib import Path
+
+
+class DiskCache:
+    def __init__(self, cache_dir: str = ".cache"):
+        self.cache_dir = Path(cache_dir)
+        self.cache_dir.mkdir(parents=True, exist_ok=True)
+
+    def _key_path(self, key: str) -> Path:
+        h = hashlib.sha256(key.encode()).hexdigest()
+        return self.cache_dir / f"{h}.json"
+
+    def get(self, key: str):
+        path = self._key_path(key)
+        if path.exists():
+            return json.loads(path.read_text())
+        return None
+
+    def set(self, key: str, value):
+        self._key_path(key).write_text(json.dumps(value))
+
+    def invalidate(self, key: str):
+        p = self._key_path(key)
+        if p.exists():
+            p.unlink()
+
+
+class ReportStore:
+    def __init__(self, base_dir: str = "./reports"):
+        self.base_dir = Path(base_dir)
+
+    def save_history(self, program: str, status: str, matched: int, duration: float):
+        trend = self.base_dir / "trends" / f"{program}.jsonl"
+        trend.parent.mkdir(parents=True, exist_ok=True)
+        import datetime
+        entry = {"ts": datetime.datetime.now().isoformat(), "status": status,
+                 "fields_matched": matched, "duration_s": duration}
+        with open(trend, "a") as f:
+            f.write(json.dumps(entry) + "\n")
diff --git a/tests/__init__.py b/tests/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/comparator/test_aligner.py b/tests/comparator/test_aligner.py
new file mode 100644
index 0000000..7ae619d
--- /dev/null
+++ b/tests/comparator/test_aligner.py
@@ -0,0 +1,45 @@
+import sys, os
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..")))
+from comparator.aligner import align_records
+
+
+def test_align_by_key():
+    cobol = [{"CUST-ID": "C001", "AMT": 100}, {"CUST-ID": "C002", "AMT": 200}]
+    spark = [{"CUST-ID": "C002", "AMT": 200}, {"CUST-ID": "C001", "AMT": 100}]
+    result = align_records(cobol, spark, key_field="CUST-ID")
+    assert len(result) == 2
+    assert all(s == "MATCHED" for _, _, s in result)
+
+
+def test_missing_in_spark():
+    cobol = [{"CUST-ID": "C001"}, {"CUST-ID": "C002"}]
+    spark = [{"CUST-ID": "C001"}]
+    result = align_records(cobol, spark, key_field="CUST-ID")
+    statuses = [s for _, _, s in result]
+    assert "MISSING_IN_SPARK" in statuses
+
+
+def test_extra_in_spark():
+    cobol = [{"CUST-ID": "C001"}]
+    spark = [{"CUST-ID": "C001"}, {"CUST-ID": "C002"}]
+    result = align_records(cobol, spark, key_field="CUST-ID")
+    statuses = [s for _, _, s in result]
+    assert "EXTRA_IN_SPARK" in statuses
+
+
+def test_empty_inputs():
+    assert align_records([], [], "key") == []
+
+
+def test_duplicate_keys():
+    cobol = [{"ID": "K1", "V": 1}, {"ID": "K1", "V": 2}]
+    java = [{"ID": "K1", "V": 1}, {"ID": "K1", "V": 2}]
+    result = align_records(cobol, java, key_field="ID")
+    assert len(result) == 2
+
+
+def test_align_none_key():
+    cobol = [{"ID": None, "V": 1}]
+    java = [{"ID": None, "V": 1}]
+    result = align_records(cobol, java, key_field="ID")
+    assert len(result) == 1
diff --git a/tests/comparator/test_aligner_edge.py b/tests/comparator/test_aligner_edge.py
new file mode 100644
index 0000000..b6fd9e2
--- /dev/null
+++ b/tests/comparator/test_aligner_edge.py
@@ -0,0 +1,18 @@
+import sys, os
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
+from comparator.aligner import align_records
+
+
+def test_align_empty_key_value():
+    cobol = [{"ID": "", "V": 1}]
+    java = [{"ID": "", "V": 1}]
+    result = align_records(cobol, java, key_field="ID")
+    assert len(result) == 1
+
+
+def test_align_very_large_key_set():
+    cobol = [{"ID": f"K{i:04d}", "V": i} for i in range(100)]
+    java = [{"ID": f"K{i:04d}", "V": i} for i in range(100)]
+    result = align_records(cobol, java, key_field="ID")
+    assert len(result) == 100
+    assert all(s == "MATCHED" for _, _, s in result)
diff --git a/tests/comparator/test_compare_edge.py b/tests/comparator/test_compare_edge.py
new file mode 100644
index 0000000..c90f453
--- /dev/null
+++ b/tests/comparator/test_compare_edge.py
@@ -0,0 +1,23 @@
+import sys, os
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..")))
+from comparator.field_compare import compare_field
+
+
+def test_negative_numbers():
+    r = compare_field("AMT", "-1500", "-1500", "decimal")
+    assert r.status == "PASS"
+
+
+def test_mixed_precision():
+    r = compare_field("AMT", "1500.00", "1500", "decimal", tolerance=0.01)
+    assert r.status == "PASS"
+
+
+def test_non_numeric_in_numeric_field():
+    r = compare_field("AMT", "ABC", "1500", "decimal")
+    assert r.status in ("MISMATCH", "NOT_SET")
+
+
+def test_very_large_number():
+    r = compare_field("AMT", "9999999999", "9999999999", "decimal")
+    assert r.status == "PASS"
diff --git a/tests/comparator/test_field_compare.py b/tests/comparator/test_field_compare.py
new file mode 100644
index 0000000..f35168e
--- /dev/null
+++ b/tests/comparator/test_field_compare.py
@@ -0,0 +1,49 @@
+import sys, os
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..")))
+from comparator.field_compare import compare_field, DEFAULT_TOLERANCE
+
+
+def test_exact_match():
+    r = compare_field("BR-AMT", "1500000", "1500000", "decimal")
+    assert r.status == "PASS"
+
+
+def test_within_tolerance():
+    r = compare_field("BR-AMT", "1500000", "1499999.99", "decimal", tolerance=DEFAULT_TOLERANCE)
+    assert r.status == "TOLERATED"
+
+
+def test_beyond_tolerance():
+    r = compare_field("BR-AMT", "1500000", "1000000", "decimal", tolerance=DEFAULT_TOLERANCE)
+    assert r.status == "MISMATCH"
+
+
+def test_string_trim():
+    r = compare_field("BR-STATUS", "A  ", "A", "string")
+    assert r.status == "PASS"
+
+
+def test_date_normalization():
+    r = compare_field("BR-DATE", "20260522", "2026-05-22", "date")
+    assert r.status == "PASS"
+
+
+def test_cobol_default():
+    from decimal import Decimal, ROUND_DOWN
+    r = compare_field("BR-AMT", "\x00\x00\x00\x00\x00", "0", "decimal")
+    assert r.status in ("PASS", "TOLERATED")
+
+
+def test_java_null_vs_value():
+    r = compare_field("BR-AMT", "1500000", "None", "decimal")
+    assert r.status in ("MISMATCH", "NOT_SET")
+
+
+def test_negative_numbers():
+    r = compare_field("AMT", "-1500", "-1500", "decimal")
+    assert r.status == "PASS"
+
+
+def test_mixed_precision():
+    r = compare_field("AMT", "1500.00", "1500", "decimal", tolerance=DEFAULT_TOLERANCE)
+    assert r.status == "PASS"
diff --git a/tests/comparator/test_normalizer.py b/tests/comparator/test_normalizer.py
new file mode 100644
index 0000000..f981e1d
--- /dev/null
+++ b/tests/comparator/test_normalizer.py
@@ -0,0 +1,47 @@
+import sys, os
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..")))
+from comparator.normalizer import Normalizer
+
+
+def test_ebcdic_to_ascii():
+    n = Normalizer()
+    assert n.normalize_encoding(b'\xc1\xc2', "EBCDIC") == "AB"
+
+
+def test_ascii_passthrough():
+    n = Normalizer()
+    assert n.normalize_encoding(b"hello", "ASCII") == "hello"
+
+
+def test_comp3_to_decimal():
+    n = Normalizer()
+    assert n.normalize_comp3(b'\x15\x00\x0C') == "1500"
+
+
+def test_comp3_negative():
+    n = Normalizer()
+    assert n.normalize_comp3(b'\x15\x00\x1D') == "-1500"
+
+
+def test_ir_record_creation():
+    n = Normalizer()
+    ir = n.to_ir_record(
+        field_name="BR-AMT", raw_hex="15000C",
+        decoded_value="1500", encoding="EBCDIC",
+        field_type="COMP3", length=5, scale=2, signed=True)
+    assert ir.field_name == "BR-AMT"
+    assert ir.cobol.decoded_value == "1500"
+    assert ir.cobol.encoding == "EBCDIC"
+
+
+def test_date_iso_normalization():
+    n = Normalizer()
+    assert n.normalize_date("20260522") == "2026-05-22"
+    assert n.normalize_date("2026-05-22") == "2026-05-22"
+
+
+def test_null_ir_record():
+    n = Normalizer()
+    ir = n.to_null_ir("BR-DATE", side="java")
+    assert ir.field_name == "BR-DATE"
+    assert ir.java is None
diff --git a/tests/comparator/test_rounding.py b/tests/comparator/test_rounding.py
new file mode 100644
index 0000000..0c88a89
--- /dev/null
+++ b/tests/comparator/test_rounding.py
@@ -0,0 +1,24 @@
+import sys, os
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..")))
+from comparator.rounding_detect import detect_rounding, RoundingResult
+
+
+def test_truncation_detected():
+    r = detect_rounding("1500000", "1499999")
+    assert r.mode in ("TRUNCATE", "HALF_UP")
+
+
+def test_exact_match():
+    r = detect_rounding("1500000", "1500000")
+    assert r.mode == "EXACT"
+    assert r.confidence == 1.0
+
+
+def test_low_confidence_small_diff():
+    r = detect_rounding("1500", "1498")
+    assert r.confidence < 1.0
+
+
+def test_suggestion_generated():
+    r = detect_rounding("1500000", "1499999")
+    assert len(r.suggestion) > 0
diff --git a/tests/fixtures/simple.cbl b/tests/fixtures/simple.cbl
new file mode 100644
index 0000000..011ad72
--- /dev/null
+++ b/tests/fixtures/simple.cbl
@@ -0,0 +1,29 @@
+       IDENTIFICATION DIVISION.
+       PROGRAM-ID. SIMPLE.
+       ENVIRONMENT DIVISION.
+       INPUT-OUTPUT SECTION.
+       FILE-CONTROL.
+           SELECT INFILE ASSIGN TO "input.bin"
+           ORGANIZATION IS SEQUENTIAL.
+       DATA DIVISION.
+       FILE SECTION.
+       FD INFILE.
+       01 BILL-RECORD.
+          05 BR-AMT    PIC S9(7)V99 COMP-3.
+          05 BR-STATUS PIC X.
+          05 BR-DATE   PIC 9(8).
+       WORKING-STORAGE SECTION.
+       01 WS-EOF PIC X VALUE 'N'.
+       PROCEDURE DIVISION.
+           OPEN INPUT INFILE.
+           PERFORM UNTIL WS-EOF = 'Y'
+               READ INFILE INTO BILL-RECORD
+                   AT END MOVE 'Y' TO WS-EOF
+                   NOT AT END
+                       DISPLAY BR-AMT
+                       DISPLAY BR-STATUS
+                       DISPLAY BR-DATE
+               END-READ
+           END-PERFORM.
+           CLOSE INFILE.
+           STOP RUN.
diff --git a/tests/fixtures/simple.cpy b/tests/fixtures/simple.cpy
new file mode 100644
index 0000000..fdb299a
--- /dev/null
+++ b/tests/fixtures/simple.cpy
@@ -0,0 +1,4 @@
+01 BILL-RECORD.
+   05 BR-AMT    PIC S9(7)V99 COMP-3.
+   05 BR-STATUS PIC X.
+   05 BR-DATE   PIC 9(8).
diff --git a/tests/fixtures/simple.yaml b/tests/fixtures/simple.yaml
new file mode 100644
index 0000000..06b2c1e
--- /dev/null
+++ b/tests/fixtures/simple.yaml
@@ -0,0 +1,13 @@
+program: "SIMPLE"
+field_mapping:
+  - cobol_field: "BR-AMT"
+    java_field: "billAmount"
+    type: "decimal"
+    precision: 2
+  - cobol_field: "BR-STATUS"
+    java_field: "statusCode"
+    type: "string"
+  - cobol_field: "BR-DATE"
+    java_field: "billDate"
+    type: "date"
+    format: "YYYYMMDD"
diff --git a/tests/report/__init__.py b/tests/report/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/report/test_generator.py b/tests/report/test_generator.py
new file mode 100644
index 0000000..6add312
--- /dev/null
+++ b/tests/report/test_generator.py
@@ -0,0 +1,44 @@
+import sys, os, json
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..")))
+from pathlib import Path
+from report.generator import ReportGenerator
+from data.diff_result import VerificationRun, FieldResult
+
+
+def test_json_output(tmp_path):
+    vr = VerificationRun(program="BILL-CALC", status="PASS", exit_code=0,
+                         field_results=[FieldResult(field_name="BR-AMT", status="PASS")])
+    gen = ReportGenerator()
+    path = gen.generate_json(vr, tmp_path / "result.json")
+    data = json.loads(path.read_text())
+    assert data["program"] == "BILL-CALC"
+    assert data["status"] == "PASS"
+
+
+def test_html_output(tmp_path):
+    vr = VerificationRun(program="TEST", status="MISMATCH",
+                         field_results=[FieldResult(field_name="F1", status="MISMATCH")])
+    gen = ReportGenerator()
+    path = gen.generate_html(vr, tmp_path / "report.html")
+    assert path.exists()
+    html = path.read_text()
+    assert "MISMATCH" in html
+    assert "F1" in html
+
+
+def test_machine_json(tmp_path):
+    vr = VerificationRun(program="TEST", status="PASS", exit_code=0)
+    gen = ReportGenerator()
+    path = gen.generate_machine_json(vr, tmp_path / "machine.json")
+    data = json.loads(path.read_text())
+    assert data["exit_code"] == 0
+
+
+def test_suggestion_in_report(tmp_path):
+    fr = FieldResult(field_name="BR-AMT", status="MISMATCH",
+                     suggestion="Check rounding_mode: TRUNCATE vs HALF_UP")
+    vr = VerificationRun(program="TEST", status="MISMATCH", field_results=[fr])
+    gen = ReportGenerator()
+    path = gen.generate_json(vr, tmp_path / "result.json")
+    data = json.loads(path.read_text())
+    assert "suggestion" in data["field_results"][0]
diff --git a/tests/test_e2e.py b/tests/test_e2e.py
new file mode 100644
index 0000000..acbdbb9
--- /dev/null
+++ b/tests/test_e2e.py
@@ -0,0 +1,33 @@
+import sys, os
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
+from config import Config
+from orchestrator import run_pipeline
+
+
+def test_e2e_pipeline_imports():
+    """Verify all modules import correctly without runtime tools."""
+    from data.field_tree import Field, FieldTree
+    from data.test_case import TestCase, TestSuite, SparkConfig
+    from data.diff_result import FieldResult, VerificationRun
+    from runners.runner import Runner, BuildResult, RunResult, CoverageReport
+    from runners.native_java_runner import NativeJavaRunner
+    from runners.spark_java_runner import SparkJavaRunner
+    from runners.cobol_runner import CobolRunner
+    from runners.data_writer import DataWriter
+    from agents.llm import LLMClient
+    from agents.agent1_parser import Agent1Parser
+    from agents.agent2_data import Agent2Data
+    from agents.agent3_diagnostic import Agent3Diagnostic
+    from comparator.aligner import align_records
+    from comparator.field_compare import compare_field
+    from comparator.normalizer import Normalizer
+    from comparator.rounding_detect import detect_rounding
+    from comparator.cobol_binary_reader import CobolBinaryReader
+    from report.generator import ReportGenerator
+    from storage.bundle import TestDataBundle
+    from storage.store import ReportStore, DiskCache
+    from preprocessor import CopybookPreprocessor
+    from config.mapping import MappingConfig, FieldMapping
+    from quality.l1_offset_validate import L1OffsetValidator
+    from quality.l2_value_roundtrip import L2RoundtripValidator
+    assert True