Framework

Methodology

Data Privacy Risk Quantification Framework

Overview

Liability Quant (LQ) provides a standardized, reproducible measurement of the inherent privacy exposure in a dataset. It supplements due diligence by showing counsel where the exposure is and what to confirm.

Use Cases

•M&A due diligence
•Acquisition of proprietary data or LLMs
•Cyber insurance underwriting
•General compliance assessments

Outputs

LQ Score (0–100)
Higher scores indicate lower risk
Modeled Exposure
Fixed dollar band for inherent exposure, scaled by record count
Follow-up Questions for Counsel
Generated from the findings: which consents, bases and records to confirm

The LQ Score is a directional risk indicator, analogous to credit ratings: a standardized opinion from a published methodology, applied the same way every time. It is not a prediction of enforcement outcomes.

Risk Assessment Pipeline

LQ processes datasets through a six-stage pipeline:

Data Ingestion

Upload dataset

PII Detection

4-tier hierarchy

Linkage Analysis

Mosaic Effect

Context Adjustment

Jurisdiction

Exposure Modeling

Directional band

Score Generation

LQ Score 0-100

Each stage is traceable. Identical inputs produce identical outputs.

Detection Framework

LQ employs a two-pass, four-tier detection hierarchy to identify 70+ PII types and 20+ linkage combinations. The system is deterministic with no LLM dependencies.

Header, pattern, checksum and keyword checks cover every value in the file. NLP entity recognition, which is far slower per value, analyzes a seeded random sample of up to 300 values per column, so results are reproducible and not biased toward the first rows of a sorted file. Every report includes a coverage statement listing what was checked.

First Pass: Atomic PII Detection

Tier	Method	Description
1	Structural	Headers for metadata and schema mapping
2	Pattern	Deterministic regex and checksums on every value — accounts, IDs, phones
3	Entity	NLP named entity recognition on a reproducible random sample
4	Content	Whole-word keyword matching on every value

Second Pass

•Detect linkage combinations (toxic pairs)
•Conduct subject inference

Validation Layer

Checksums (Luhn, Modulo-97, Modulus-11) reduce false positives by 10–15%.

The Mosaic Effect

Seemingly innocuous data fields, when combined, can uniquely identify individuals:

87%of the U.S. population is identifiable via ZIP code + date of birth + gender
95%of individuals are identifiable from just four spatio-temporal data points

LQ identifies toxic pairs and applies multiplicative amplification—linkable combinations transform the attack surface globally, not incrementally.

Research Foundation

Sweeney (1997), de Montjoye et al. (2013), Gymrek et al. (2013)

Risk Calculation Model

Risk is calculated as a product of compounding factors rather than a linear sum—accounting for the Mosaic Effect.

R_total = R_base × A_linkage × M_context

R_base

Weighted aggregate of detected PII, with weights informed by the severity of statutory regimes (GDPR, HIPAA, CCPA). Health and biometric data weighted heavily; quasi-identifiers weighted lightly.

A_linkage

Mosaic Effect amplifier for re-identification risk from linkable clusters like DOB + ZIP + gender.

M_context

Product of operational multipliers: jurisdiction, data subject risk, and intended use case. Jurisdiction, history and intended use are declared by the user, not detected.

Rating Classification

LQ Score	Risk Class	Key Characteristics	Typical Diligence Considerations
90–100	Minimal	Minimal PII, no toxic pairs	Standard governance
70–89	Low	Low-risk PII, few toxic pairs	Standard representations & warranties typically sufficient
50–69	Moderate	Mixed PII, some toxic pairs	Consider enhanced controls and targeted diligence
30–49	High	High-risk PII, multiple toxic pairs	Consider specific indemnity or escrow
0–29	Critical	Highly sensitive PII, severe toxic pairs	Consider remediation before close, specific indemnity or escrow

Exposure Bands

Reports show modeled exposure as one of six fixed bands rather than a single figure. The same file always lands in the same band, and bands compare cleanly across datasets.

Under $100K

$100K–$500K

$500K–$2.5M

$2.5M–$10M

$10M–$50M

Over $50M

Score and exposure are reported separately. The exposure estimate scales with the number of records, so two datasets with the same score can fall in very different dollar bands. The estimate uses a baseline of $250 per record, loosely anchored to published per-record breach cost research, and is not fitted to enforcement outcomes.

Illustrative Example

A customer database with SSNs, emails, DOBs, and timestamps containing an SSN + DOB toxic pair. EU deployment with 60% high-risk subjects. Result: LQ Score 38 (High Risk), modeled exposure in the $2.5M–$10M band.

Inherent vs. Residual Risk

LQ measures presence, not permissibility. The scan establishes what personal data is in the file and how linkable it is. That is the dataset's inherent exposure: the exposure that exists if no mitigating documentation is found.

Diligence then determines residual exposure by confirming (or failing to confirm) consent, lawful basis, provenance and transfer mechanisms. Those factors can substantially reduce or increase actual exposure. LQ directs that work to where it matters most; it does not replace it.

Each report lists follow-up questions for counsel, generated from its findings by a fixed rules table. Health data prompts questions about HIPAA status and authorization; detected linkages prompt questions about de-identification claims; declared AI training use prompts questions about notice and consent. The scan can't check consent, but it can tell counsel exactly which consents to check.

Transactional Application

For high-risk or critical-risk datasets, counsel and the deal team typically weigh options such as:

Protections

Preserve full data utility

Technical:

• Tokenization
• Encryption-at-rest
• Access controls

Legal/Contractual:

• Specific indemnities
• Escrow arrangements
• Compliance warranties
• Remediation timelines

Data Modification

Privacy-utility trade-off

• Suppression
• Generalization
• Differential privacy

These techniques reduce re-identification risk but may diminish analytical value of the dataset.

Scope & Limitations

What LQ Analyzes

Structured or relational datasets and database exports (CSV, Excel, SQL)

Point-in-Time

• Results reflect dataset state at time of analysis
• Periodic reassessment recommended as data evolves

Limitations

•Exposure figures are directional; actual outcomes depend on jurisdiction, enforcement likelihood, class action exposure and remediation costs
•Cannot detect encoded, encrypted, or intentionally obfuscated data
•Cannot verify that submitted samples are representative or complete

What the Scan Does Not Assess

These factors are material to liability and are established in diligence, not by a scan:

•Existence and validity of consent
•Lawful basis for processing (GDPR, CCPA/CPRA)
•Data provenance and chain of custody
•Prior breaches or regulatory history (unless declared)
•Cross-border transfer restrictions, including Quebec Law 25
•Contractual commitments (privacy policies, DPAs)
•Likelihood of enforcement or class action
•Unstructured content such as long free text, images and documents

This methodology provides quantitative risk indicators that supplement due diligence. It does not replace diligence and does not constitute legal advice. Consult qualified counsel for jurisdiction-specific compliance guidance.