pamola-core 1.0.0.dev1

Pamola Core library for data anonymization, privacy models, metrics, and utilities

PAMOLA.CORE

---

Privacy Engineering for Python. Finally.

PAMOLA.CORE is the open-source foundation of the PAMOLA platform - a Python library for privacy-preserving data operations with pipeline workflow, reproducibility, and audit trail.

Developed by Realm Inveo Inc.

---

The Problem

You need to anonymize sensitive data. You've tried:

• ARX: Powerful, but Java, GUI-focused, opaque operations
• Faker + Presidio + custom scripts: Fragmented, no pipeline, no proof
• DP libraries: Great math, but narrow scope

You're still missing:

• Direct operations (mask, drop, fake - not just "achieve k-anonymity")
• Risk testing (can someone actually re-identify records?)
• Short text handling (job titles, comments - without LLM)
• Reproducibility (what exactly was done?)

---

The Solution

PAMOLA.CORE: operations-first privacy engineering with full audit trail.

from pamola_core.tasks import TaskRunner
from pamola_core.profiling import ProfileOperation
from pamola_core.anonymization import MaskingOperation, GeneralizationOperation
from pamola_core.noise import LaplaceNoiseOperation
from pamola_core.metrics import PrivacyProxyMetricsOperation
from pamola_core.attacks import AttackSuiteOperation

# Define pipeline with reproducible seed
task = TaskRunner(task_dir="./anonymize_customers", seed=42)

task.run([
    ProfileOperation(params={"analyzers": ["all"]}),
    MaskingOperation(params={"fields": ["name", "email", "phone"], "strategy": "partial"}),
    GeneralizationOperation(params={"field": "age", "bins": [0, 18, 35, 50, 65, 100]}),
    LaplaceNoiseOperation(params={"fields": ["salary"], "epsilon": 1.0, "sensitivity": 1000}),
    PrivacyProxyMetricsOperation(params={"metrics": ["k_anonymity", "l_diversity"]}),
    AttackSuiteOperation(params={"policy": "standard"}),
], input_data="customers.csv")

# Result: task_dir/ with data, metrics, attack results, manifest.json

Output structure:

anonymize_customers/
├── manifest.json          # Full reproducibility record
├── output/                # Anonymized data (csv/parquet)
│   └── anonymized.csv
├── metrics/               # Privacy & utility metrics (JSON)
│   ├── metrics_summary.json
│   └── metrics_detail.json
├── attacks/               # Attack simulation results
│   └── suite_report.json
├── plots/                 # Generated visualizations
├── dictionaries/          # Extracted mappings
└── logs/                  # Execution logs

---

PAMOLA Ecosystem

PAMOLA.CORE is part of a comprehensive privacy engineering stack:

Component	Description	Availability
PAMOLA.CORE	Operations, metrics, attacks, pipeline runtime	Open Source
PAMOLA.STUDIO	Visual environment for data transformation and privacy management	Commercial
PAMOLA.SYNT	Synthetic data generation with DP guarantees (CTGAN, TVAE)	Commercial
PAMOLA.TEXT	Long text and document anonymization (NLP/LLM-based)	Commercial
PAMOLA.INSIGHT	Agent modules for LLM integration	Commercial

---

What's In CORE

Category	Operations
Profiling	ProfileOperation, CorrelationOperation, ShortTextProfileOperation
Anonymization	MaskingOperation, GeneralizationOperation, SuppressionOperation, PseudonymizationOperation
Transformation	CleaningOperation, MergeOperation, SplitOperation, AggregateOperation
Noise (DP-semantics)	LaplaceNoiseOperation, GaussianNoiseOperation, DateTimeJitterOperation, RandomizedResponseOperation
Short Text	ShortTextProfileOperation, ShortTextCategorizerOperation, ShortTextMaskOperation, ShortTextNEROperation
Fake Data	FakeNameOperation, FakeEmailOperation, FakePhoneOperation, FakeOrgOperation
Metrics	QualityMetricsOperation, PrivacyProxyMetricsOperation, AttackBasedMetricsOperation, CompositeScoreOperation
Attacks	CVPLAttackOperation, LinkageAttackOperation, SinglingOutOperation, AttributeInferenceOperation, AttackSuiteOperation

---

Data-Release Attacks

PAMOLA.CORE tests practical re-identification risk on your data:

Attack	Question
CVPL	How much information leaks between releases? (PAMOLA signature)
Fellegi-Sunter Linkage	Can records be matched to external data?
Singling-out	Are any records uniquely identifiable?
Attribute inference	Can sensitive attributes be guessed from QI?

from pamola_core.tasks import TaskRunner
from pamola_core.attacks import AttackSuiteOperation

task = TaskRunner(task_dir="./risk_assessment", seed=42)
task.run([
    AttackSuiteOperation(params={
        "policy": "standard",  # or "minimal", "comprehensive"
        "quasi_identifiers": ["age", "gender", "zipcode"],
        "sensitive_columns": ["diagnosis"]
    })
], input_data="anonymized.csv")

# Result: attacks/suite_report.json with risk scores and verdicts

Note: Model-centric attacks (MIA on generators) belong to PAMOLA.SYNT

---

Metrics with Verdicts

Metrics produce actionable signals, not just numbers:

from pamola_core.metrics import CompositeScoreOperation

# Aggregate metrics with weighted scoring
CompositeScoreOperation(params={
    "weights": {
        "quality": 0.3,
        "privacy_proxy": 0.2,
        "privacy_attack": 0.4,  # Attack-based metrics weighted higher
        "utility": 0.1
    }
})
# Output: metrics_summary.json with verdict (PASS/WARN/FAIL)

Output example (metrics_summary.json):

{
  "overall": {
    "quality_score": 0.85,
    "privacy_score": 0.78,
    "composite_score": 0.84,
    "verdict": "PASS"
  },
  "metrics": {
    "k_anonymity": {"value": 5, "verdict": "PASS"},
    "linkage_rate": {"value": 0.02, "verdict": "PASS"}
  }
}

---

DP-Semantics Noise

Add calibrated noise with differential privacy semantics:

from pamola_core.noise import LaplaceNoiseOperation

LaplaceNoiseOperation(params={
    "fields": ["salary", "age"],
    "epsilon": 1.0,
    "sensitivity": {"salary": 1000, "age": 1},
    "seed": 42,
    "clip_bounds": {"salary": [0, None]}  # No negative values
})
# Output includes noise_report.json documenting exactly what was applied

Note: This provides DP-like noise but NOT formal DP guarantees without external accountant. For formal guarantees, use pamola-core[dp] with OpenDP adapter.

---

What's NOT in CORE

Feature	Package	Why separate
Long text + LLM anonymization	pamola-core[text]	Heavy deps (torch, transformers)
Formal DP with accountant	pamola-core[dp]	Use OpenDP/diffprivlib adapters
Synthetic data generation	pamola-synt	Different concern, ML models
Model-centric attacks (MIA)	pamola-synt	Requires trained model access

---

Installation

From source (current):

git clone https://github.com/DGT-Network/PAMOLA.git
cd PAMOLA
pip install -e .

With optional extras:

pip install -e ".[fast]"       # + Polars, ConnectorX, DuckDB
pip install -e ".[profiling]"  # + YData-profiling, Presidio
pip install -e ".[ner]"        # + spaCy for short text NER
pip install -e ".[dp]"         # + OpenDP for formal DP guarantees
pip install -e ".[dev]"        # + pytest, coverage, black, ruff

From PyPI:

pip install pamola-core

---

Supported Python Versions

PAMOLA.CORE supports Python 3.10, 3.11, and 3.12 (requires-python = ">=3.10,<3.13").

Python Version	Supported
3.10	✅
3.11	✅
3.12	✅
3.9 and below	❌
3.13 and above	❌ (blocked by third-party dependencies)

---

Core Dependencies

These packages are declared in pyproject.toml under [project.dependencies] and are automatically installed with pip install pamola-core.

Package	Version Range	Purpose
numpy	==1.26.4	Numerical computation and array operations used throughout privacy metrics, attack simulations, and statistical analysis
pandas	==2.2.2	Tabular data structures and DataFrame processing; the primary data container for all PAMOLA operations
scikit-learn	==1.7.2	Machine learning utilities used by core operations including nearest-neighbor attacks, classification metrics, and model-based privacy risk assessment

---

Versioning

PAMOLA.CORE follows Semantic Versioning and PEP 440.

import pamola_core
print(pamola_core.__version__)  # e.g. "1.0.0.dev1"

Phase	Version	Branch	Tag	Install
Dev	1.0.0.dev1	develop	v1.0.0.dev1	pip install pamola-core==1.0.0.dev1
Stable	1.0.0	main	v1.0.0	pip install pamola-core

• Source of truth: pyproject.toml → version
• Changelog: CHANGELOG.md
• CI/CD: GitHub Actions — lint, test (3.10/3.11/3.12), build, PyPI publish on tag v*
• Release rules: Dev tags (v*dev*) must be on develop, stable tags on main

---

CLI

pamola-core --version
pamola-core list-ops
pamola-core run --task task.json --output ./task_dir
pamola-core run --op MaskingOperation --config config.json --input data.csv
pamola-core schema MaskingOperation

---

Sample Data

Note: Synthetic test datasets are available in data/raw/ for development and testing purposes only.

No real personal data (PII/PHI) is included. All records are artificially generated.

---

Philosophy

• Operations-first: Direct transforms, not constraint optimization
• Measure everything: Quality, privacy, utility - with verdicts
• Test before release: Practical risk via data-release attacks
• Noise with transparency: DP-semantics + detailed reports
• Reproducibility by default: manifest.json tracks everything

---

API Documentation

The project uses Sphinx to generate API reference documentation from Python docstrings.

Build the documentation locally:

cd docs
make html

The generated documentation will be available at:

docs/_build/html/index.html

---

Documentation

Resource	Link
PET Knowledge Base	realmdata.io/kb
Technical Documentation	docs/en/index.md
Glossary	realmdata.io/glossary
Examples	examples/

---

Use Cases

• Data Engineering: Prepare privacy-safe datasets for ML training
• Healthcare: HIPAA-oriented de-identification workflows (Safe Harbor support)
• Finance: Privacy engineering aligned with PCI/GDPR considerations
• Compliance: Audit-ready evidence with manifest.json and attack reports
• Data Sharing: Risk-assessed data exchange between organizations

---

Regulatory Context

PAMOLA.CORE provides technical building blocks for privacy compliance programs:

Regulation	Relevant Capabilities
GDPR	Pseudonymization, data minimization (Art. 25, 32)
HIPAA	Safe Harbor de-identification support
CCPA/CPRA	Data suppression, anonymization workflows

Important: PAMOLA.CORE provides technical capabilities only. Legal compliance requires organizational policies, procedures, and legal guidance beyond software tools.

---

Contributing

git clone https://github.com/DGT-Network/PAMOLA.git
cd PAMOLA
pip install -e ".[dev]"
pytest tests/ -v

See CONTRIBUTING.md for guidelines.

---

Ownership & Licensing

PAMOLA.CORE is developed and owned exclusively by Realm Inveo Inc.

This repository is hosted under DGT-Network GitHub organization, which provides shared development infrastructure for Realm Inveo projects. DGT-Network does not claim ownership of this intellectual property. All IP rights belong exclusively to Realm Inveo Inc.

License: Apache 2.0 - see LICENSE

---

Contact

Purpose	Contact
General inquiries	[e-mail]
Commercial / Sales	[e-mail]
Due diligence / Legal	[e-mail]
Website	realmdata.io

---

_{Built by Realm Inveo Inc.}