# Exhaustive PII Target List (Task 3: Legal Compliance Redaction)

This document defines the strict, closed-boundary list of PII categories the agent is responsible for redacting, along with the adversarial "traps" designed to test precision vs. recall.

## 1. Social Security Numbers (SSNs)
*   **Target:** Standard US format (`XXX-XX-XXXX`).
*   **The Trap:** A part number formatted exactly the same way (e.g., `Part #123-45-6789` or `Serial: 123-45-6789`). The agent must use NLP context or regex lookbehinds to realize a machine part number is not a human's SSN.

## 2. Email Addresses
*   **Target:** Any standard email (`first.last@domain.com`).
*   **The Trap:** Website URLs (`www.gmail.com` or `https://company.com/@user`). Only email addresses should be redacted.

## 3. Phone Numbers
*   **Target:** Any 10-digit US phone number with or without area codes (e.g., `(555) 123-4567` or `555-123-4567`). Do not attempt to distinguish "personal" from "business" phones via NLP; redact all standard phone numbers.
*   **The Trap:** Dates formatted with hyphens (`2026-08-03`) or Bates stamp numbers common in litigation (`DOC-555-1234`). Any phone-like number prefixed exactly by `"DOC-"` must be preserved.

## 4. Credit Card Numbers
*   **Target:** 16-digit numbers, spaced or dashed (`1234 5678 9101 1121` or `1234-5678-9101-1121`).
*   **The Trap:** 16-digit software license keys or alphanumeric product keys containing letters (`123A-567B-910C-112D`).

## 5. Dates of Birth (DOB)
*   **Target:** Any date explicitly associated with a birth (`born on 12/05/1980` or `DOB: 12-05-1980`).
*   **The Trap:** General dates of events, meetings, or incidents (`the incident occurred on 12/05/1980`). Standard dates must *not* be redacted, only birth dates.

## 6. Names of Natural Persons (Private Individuals)
*   **Target:** For the purposes of this deterministic benchmark, the agent MUST redact any capitalized first and last name appearing immediately after the exact phrases `"her name is "` or `"his name is "`. **Crucially, once a private name is identified via this trigger, ALL subsequent occurrences of that exact name anywhere in the document must also be globally redacted.**
*   **The Trap:** The names of public officials, clinicians, and corporate entities. For this benchmark, any capitalized name prefixed by `"Dr. "`, `"Judge "`, or `"Officer "` MUST be preserved. In compliance redaction, professionals acting in their official capacity are exempt from redaction.