diff --git a/CHANGELOG.md b/CHANGELOG.md
index 43f56d3..5677ac3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,7 @@
 
 ## Unreleased
 
+- ERP Explorer für CSV-Profiling, Dublettenanalyse und optionalen RollCalc-Abgleich ergänzt.
 - RollCalc-JSON-Importer mit strikter Struktur- und Typvalidierung ergänzt.
 - Initiale Projektstruktur angelegt.
 - Dokumentation fuer Architektur, Datenmodell, Datenwoerterbuch und Merge-Regeln erstellt.
diff --git a/README.md b/README.md
index 692ec64..e9fdf05 100644
--- a/README.md
+++ b/README.md
@@ -100,6 +100,23 @@ articles = load_rollcalc_articles(Path("data/source/rollcalc/article-data.json")
 
 Er validiert die bestehende RollCalc-JSON-Struktur strikt, erhält Artikelnummern als Strings und verändert die Quelldatei nicht. Details stehen in `docs/importers.md`.
 
+## Interne Analysewerkzeuge
+
+Der ERP Explorer profiliert einen ERP-CSV-Export, ohne daraus Import- oder Merge-Regeln abzuleiten:
+
+```python
+from pathlib import Path
+
+from article_data_manager.tools.erp_explorer import write_erp_profile_report
+
+write_erp_profile_report(
+    Path("data/source/erp/production-key-data.csv"),
+    rollcalc_path=Path("data/source/rollcalc/article-data.json"),
+)
+```
+
+Der Textreport wird standardmäßig unter `data/reports/erp_profile_report.txt` erzeugt. Die Quelldaten werden nicht verändert.
+
 ## Tests
 
 ```bash
diff --git a/docs/importers.md b/docs/importers.md
index 29dfe6e..48eda5f 100644
--- a/docs/importers.md
+++ b/docs/importers.md
@@ -58,3 +58,20 @@ Fehlermeldungen enthalten Datei, Datensatzindex und Feldname, soweit anwendbar.
 ## Quelldatei
 
 Der Importer liest die Quelldatei nur. Er schreibt, verändert oder repariert die Datei nicht und erzeugt keine Ausgabe auf stdout oder stderr.
+
+## ERP Explorer
+
+Der ERP Explorer unter `article_data_manager.tools.erp_explorer` ist ein internes Analysewerkzeug und kein ERP-Importer. Er liest einen ERP-CSV-Export, profiliert alle Spalten, zählt doppelte Artikelnummern in `SL_ITEM_NO` und kann optional RollCalc-Artikelnummern gegen den ERP-Export abgleichen.
+
+Öffentliche API:
+
+```python
+from pathlib import Path
+
+from article_data_manager.tools.erp_explorer import profile_erp_csv, write_erp_profile_report
+
+profile = profile_erp_csv(Path("data/source/erp/production-key-data.csv"))
+write_erp_profile_report(Path("data/source/erp/production-key-data.csv"))
+```
+
+Der Report wird standardmäßig als `data/reports/erp_profile_report.txt` geschrieben. Der Explorer verändert weder ERP-CSV noch RollCalc-JSON und erzeugt keine `article-data.json`.
diff --git a/src/article_data_manager/tools/__init__.py b/src/article_data_manager/tools/__init__.py
new file mode 100644
index 0000000..172dd9d
--- /dev/null
+++ b/src/article_data_manager/tools/__init__.py
@@ -0,0 +1,2 @@
+"""Internal development tools."""
+
diff --git a/src/article_data_manager/tools/erp_explorer.py b/src/article_data_manager/tools/erp_explorer.py
new file mode 100644
index 0000000..0cf3955
--- /dev/null
+++ b/src/article_data_manager/tools/erp_explorer.py
@@ -0,0 +1,277 @@
+"""Read-only ERP CSV explorer for data profiling."""
+
+from __future__ import annotations
+
+import csv
+from collections import Counter
+from dataclasses import dataclass
+from pathlib import Path
+
+from article_data_manager.importers.rollcalc_json import load_rollcalc_articles
+
+DEFAULT_ARTICLE_NUMBER_COLUMN = "SL_ITEM_NO"
+DEFAULT_REPORT_PATH = Path("data/reports/erp_profile_report.txt")
+
+
+@dataclass(frozen=True, slots=True)
+class ColumnProfile:
+    """Profile information for one CSV column."""
+
+    name: str
+    filled_count: int
+    empty_count: int
+    distinct_count: int
+    inferred_type: str
+
+
+@dataclass(frozen=True, slots=True)
+class DuplicateArticleNumber:
+    """Repeated ERP article number and its exact occurrence count."""
+
+    nr: str
+    count: int
+
+
+@dataclass(frozen=True, slots=True)
+class ArticleNumberProfile:
+    """Profile information for an ERP article number column."""
+
+    column_name: str
+    distinct_count: int
+    duplicate_count: int
+    duplicates: tuple[DuplicateArticleNumber, ...]
+
+
+@dataclass(frozen=True, slots=True)
+class RollCalcComparison:
+    """Simple ERP/RollCalc article number coverage comparison."""
+
+    rollcalc_article_count: int
+    found_count: int
+    missing_count: int
+    missing_article_numbers: tuple[str, ...]
+
+
+@dataclass(frozen=True, slots=True)
+class ErpProfile:
+    """Complete read-only profile of one ERP CSV export."""
+
+    file_name: str
+    row_count: int
+    column_count: int
+    columns: tuple[ColumnProfile, ...]
+    article_numbers: ArticleNumberProfile
+    rollcalc_comparison: RollCalcComparison | None = None
+
+
+def profile_erp_csv(
+    csv_path: Path,
+    *,
+    article_number_column: str = DEFAULT_ARTICLE_NUMBER_COLUMN,
+    rollcalc_path: Path | None = None,
+) -> ErpProfile:
+    """Profile an ERP CSV export without interpreting or changing its source data."""
+
+    source_path = Path(csv_path)
+    rows, fieldnames = _read_csv(source_path)
+    column_profiles = tuple(
+        _profile_column(name, [row.get(name, "") for row in rows]) for name in fieldnames
+    )
+    article_number_profile = _profile_article_numbers(rows, article_number_column)
+    erp_article_numbers = {row.get(article_number_column, "") for row in rows}
+    comparison = (
+        _compare_rollcalc_articles(rollcalc_path, erp_article_numbers)
+        if rollcalc_path is not None
+        else None
+    )
+    return ErpProfile(
+        file_name=source_path.name,
+        row_count=len(rows),
+        column_count=len(fieldnames),
+        columns=column_profiles,
+        article_numbers=article_number_profile,
+        rollcalc_comparison=comparison,
+    )
+
+
+def write_erp_profile_report(
+    csv_path: Path,
+    *,
+    report_path: Path = DEFAULT_REPORT_PATH,
+    article_number_column: str = DEFAULT_ARTICLE_NUMBER_COLUMN,
+    rollcalc_path: Path | None = None,
+) -> ErpProfile:
+    """Create a text profile report under the requested report path.
+
+    The ERP CSV and optional RollCalc JSON are read only. The only write performed
+    by this function is the requested text report.
+    """
+
+    profile = profile_erp_csv(
+        csv_path,
+        article_number_column=article_number_column,
+        rollcalc_path=rollcalc_path,
+    )
+    destination = Path(report_path)
+    destination.parent.mkdir(parents=True, exist_ok=True)
+    destination.write_text(render_erp_profile_report(profile), encoding="utf-8")
+    return profile
+
+
+def render_erp_profile_report(profile: ErpProfile) -> str:
+    """Render an ERP profile as deterministic plain text."""
+
+    lines = [
+        "ERP Profile Report",
+        f"File: {profile.file_name}",
+        f"Rows: {profile.row_count}",
+        f"Columns: {profile.column_count}",
+        "",
+        "Column profiles",
+    ]
+    for column in profile.columns:
+        lines.extend(
+            [
+                f"Column: {column.name}",
+                f"Filled: {column.filled_count}",
+                f"Empty: {column.empty_count}",
+                f"Distinct: {column.distinct_count}",
+                f"Type: {column.inferred_type}",
+                "",
+            ]
+        )
+
+    lines.extend(
+        [
+            "Article numbers",
+            f"Column: {profile.article_numbers.column_name}",
+            f"Distinct: {profile.article_numbers.distinct_count}",
+            f"Duplicates: {profile.article_numbers.duplicate_count}",
+            "",
+            "Duplicate article numbers",
+        ]
+    )
+    for duplicate in profile.article_numbers.duplicates:
+        lines.append(f"{duplicate.nr}\t{duplicate.count}")
+
+    if profile.rollcalc_comparison is not None:
+        comparison = profile.rollcalc_comparison
+        lines.extend(
+            [
+                "",
+                "RollCalc comparison",
+                f"RollCalc articles: {comparison.rollcalc_article_count}",
+                f"Found: {comparison.found_count}",
+                f"Missing: {comparison.missing_count}",
+                "Missing article numbers",
+            ]
+        )
+        lines.extend(comparison.missing_article_numbers)
+
+    return "\n".join(lines) + "\n"
+
+
+def _read_csv(path: Path) -> tuple[list[dict[str, str]], list[str]]:
+    with path.open("r", encoding="utf-8-sig", newline="") as csv_file:
+        sample = csv_file.read(4096)
+        csv_file.seek(0)
+        dialect = csv.Sniffer().sniff(sample, delimiters=",;\t")
+        reader = csv.DictReader(csv_file, dialect=dialect)
+        fieldnames = list(reader.fieldnames or [])
+        return list(reader), fieldnames
+
+
+def _profile_column(name: str, values: list[str]) -> ColumnProfile:
+    filled_values = [value for value in values if not _is_empty(value)]
+    empty_count = len(values) - len(filled_values)
+    return ColumnProfile(
+        name=name,
+        filled_count=len(filled_values),
+        empty_count=empty_count,
+        distinct_count=len(set(filled_values)),
+        inferred_type=infer_value_type(filled_values),
+    )
+
+
+def infer_value_type(values: list[str]) -> str:
+    """Infer a simple data type for already-filled CSV values."""
+
+    if not values:
+        return "empty"
+
+    value_types = {_infer_single_value_type(value) for value in values}
+    if value_types == {"integer"}:
+        return "integer"
+    if value_types <= {"integer", "decimal"}:
+        return "decimal"
+    if value_types == {"text"}:
+        return "text"
+    return "mixed"
+
+
+def _infer_single_value_type(value: str) -> str:
+    normalized = value.strip()
+    if _is_integer(normalized):
+        return "integer"
+    if _is_decimal(normalized):
+        return "decimal"
+    return "text"
+
+
+def _is_integer(value: str) -> bool:
+    if value.startswith(("+", "-")):
+        value = value[1:]
+    return value.isdecimal()
+
+
+def _is_decimal(value: str) -> bool:
+    if value.count(",") + value.count(".") != 1:
+        return False
+    separator = "," if "," in value else "."
+    left, right = value.split(separator, 1)
+    if left.startswith(("+", "-")):
+        left = left[1:]
+    return left.isdecimal() and right.isdecimal()
+
+
+def _profile_article_numbers(
+    rows: list[dict[str, str]],
+    article_number_column: str,
+) -> ArticleNumberProfile:
+    values = []
+    for row in rows:
+        value = row.get(article_number_column, "")
+        if not _is_empty(value):
+            values.append(value)
+
+    counts = Counter(values)
+    duplicates = tuple(
+        DuplicateArticleNumber(nr=nr, count=count)
+        for nr, count in sorted(counts.items())
+        if count > 1
+    )
+    return ArticleNumberProfile(
+        column_name=article_number_column,
+        distinct_count=len(counts),
+        duplicate_count=len(duplicates),
+        duplicates=duplicates,
+    )
+
+
+def _compare_rollcalc_articles(
+    rollcalc_path: Path,
+    erp_article_numbers: set[str],
+) -> RollCalcComparison:
+    rollcalc_articles = load_rollcalc_articles(Path(rollcalc_path))
+    rollcalc_numbers = tuple(article.nr for article in rollcalc_articles)
+    missing_article_numbers = tuple(nr for nr in rollcalc_numbers if nr not in erp_article_numbers)
+    return RollCalcComparison(
+        rollcalc_article_count=len(rollcalc_numbers),
+        found_count=len(rollcalc_numbers) - len(missing_article_numbers),
+        missing_count=len(missing_article_numbers),
+        missing_article_numbers=missing_article_numbers,
+    )
+
+
+def _is_empty(value: str | None) -> bool:
+    return value is None or value == ""
diff --git a/tests/unit/tools/test_erp_explorer.py b/tests/unit/tools/test_erp_explorer.py
new file mode 100644
index 0000000..8375ae3
--- /dev/null
+++ b/tests/unit/tools/test_erp_explorer.py
@@ -0,0 +1,152 @@
+import json
+from pathlib import Path
+
+from article_data_manager.tools.erp_explorer import (
+    infer_value_type,
+    profile_erp_csv,
+    render_erp_profile_report,
+    write_erp_profile_report,
+)
+
+
+def write_text(path: Path, content: str) -> Path:
+    path.write_text(content, encoding="utf-8")
+    return path
+
+
+def write_rollcalc_json(path: Path, article_numbers: list[str]) -> Path:
+    payload = [
+        {
+            "nr": nr,
+            "name": f"Article {nr}",
+            "thickness": 1.0,
+            "area_weight": 0.0,
+            "core_type": 0.0,
+        }
+        for nr in article_numbers
+    ]
+    path.write_text(json.dumps(payload), encoding="utf-8")
+    return path
+
+
+def test_profiles_columns(tmp_path: Path) -> None:
+    csv_path = write_text(
+        tmp_path / "erp.csv",
+        "\n".join(
+            [
+                "SL_ITEM_NO,ROP_PRODUCT_WIDTH,COMMENT,EMPTY_COLUMN",
+                "214700,\"6,00\",Alpha,",
+                "000123,1.20,Beta,",
+                "777777,,Alpha,",
+            ]
+        ),
+    )
+
+    profile = profile_erp_csv(csv_path)
+
+    assert profile.file_name == "erp.csv"
+    assert profile.row_count == 3
+    assert profile.column_count == 4
+    columns = {column.name: column for column in profile.columns}
+    assert columns["ROP_PRODUCT_WIDTH"].filled_count == 2
+    assert columns["ROP_PRODUCT_WIDTH"].empty_count == 1
+    assert columns["ROP_PRODUCT_WIDTH"].distinct_count == 2
+    assert columns["ROP_PRODUCT_WIDTH"].inferred_type == "decimal"
+    assert columns["COMMENT"].distinct_count == 2
+    assert columns["COMMENT"].inferred_type == "text"
+    assert columns["EMPTY_COLUMN"].inferred_type == "empty"
+
+
+def test_infers_value_types() -> None:
+    assert infer_value_type([]) == "empty"
+    assert infer_value_type(["1", "002", "-3"]) == "integer"
+    assert infer_value_type(["1", "2.5", "3,75"]) == "decimal"
+    assert infer_value_type(["Alpha", "Beta"]) == "text"
+    assert infer_value_type(["1", "Alpha"]) == "mixed"
+
+
+def test_detects_duplicate_article_numbers(tmp_path: Path) -> None:
+    csv_path = write_text(
+        tmp_path / "erp.csv",
+        "\n".join(
+            [
+                "SL_ITEM_NO,NAME",
+                "00001,A",
+                "1,B",
+                "00001,C",
+                "214700,D",
+                "214700,E",
+            ]
+        ),
+    )
+
+    article_numbers = profile_erp_csv(csv_path).article_numbers
+
+    assert article_numbers.distinct_count == 3
+    assert article_numbers.duplicate_count == 2
+    assert [(duplicate.nr, duplicate.count) for duplicate in article_numbers.duplicates] == [
+        ("00001", 2),
+        ("214700", 2),
+    ]
+
+
+def test_compares_rollcalc_articles_when_path_is_provided(tmp_path: Path) -> None:
+    csv_path = write_text(
+        tmp_path / "erp.csv",
+        "\n".join(
+            [
+                "SL_ITEM_NO,NAME",
+                "214700,A",
+                "000123,B",
+            ]
+        ),
+    )
+    rollcalc_path = write_rollcalc_json(tmp_path / "article-data.json", ["214700", "999999"])
+
+    comparison = profile_erp_csv(csv_path, rollcalc_path=rollcalc_path).rollcalc_comparison
+
+    assert comparison is not None
+    assert comparison.rollcalc_article_count == 2
+    assert comparison.found_count == 1
+    assert comparison.missing_count == 1
+    assert comparison.missing_article_numbers == ("999999",)
+
+
+def test_rollcalc_comparison_is_optional(tmp_path: Path) -> None:
+    csv_path = write_text(
+        tmp_path / "erp.csv",
+        "\n".join(
+            [
+                "SL_ITEM_NO,NAME",
+                "214700,A",
+            ]
+        ),
+    )
+
+    assert profile_erp_csv(csv_path).rollcalc_comparison is None
+
+
+def test_renders_and_writes_report(tmp_path: Path) -> None:
+    csv_path = write_text(
+        tmp_path / "erp.csv",
+        "\n".join(
+            [
+                "SL_ITEM_NO,ROP_PRODUCT_WIDTH",
+                "214700,\"6,00\"",
+                "214700,\"6,00\"",
+            ]
+        ),
+    )
+    report_path = tmp_path / "reports" / "erp_profile_report.txt"
+
+    profile = write_erp_profile_report(csv_path, report_path=report_path)
+    report = report_path.read_text(encoding="utf-8")
+
+    assert report == render_erp_profile_report(profile)
+    assert "ERP Profile Report" in report
+    assert "File: erp.csv" in report
+    assert "Rows: 2" in report
+    assert "Columns: 2" in report
+    assert "Column: ROP_PRODUCT_WIDTH" in report
+    assert "Duplicate article numbers" in report
+    assert "214700\t2" in report