---
id: wiki-2026-0508-policy-surveillance
title: Policy Surveillance
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Legal Mapping, Policy Tracking, Regulatory Monitoring]
duplicate_of: none
source_trust_level: A
confidence_score: 0.85
verification_status: applied
tags: [public-policy, governance, compliance, legal-tech]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: python
  framework: legal-mapping
---

# Policy Surveillance

## 매 한 줄
> **"매 you can't evaluate what you can't measure — start by mapping the law."**. 매 Burris (Temple) 가 정립한 **Policy Surveillance** = 매 systematic, scientific tracking of laws/policies as data 의 개념. 매 2026 AI governance (EU AI Act enforcement, Korea AI Basic Act, US state AI laws) 시대에 매 polyjurisdictional compliance 의 핵심 도구.

## 매 핵심

### 매 정의 vs adjacent
- **Policy Surveillance**: 매 ongoing, systematic, scientific 매 monitoring of policies as 매 quantifiable data.
- **vs Legal Research**: 매 case-driven, episodic.
- **vs Compliance Audit**: 매 organization-internal, point-in-time.
- **vs Regulatory Tracking**: 매 news-driven, qualitative.

### 매 5단계 method (Burris)
1. 매 frame the question — what behavior does the law target?
2. 매 define jurisdictional + temporal scope.
3. 매 collect primary sources (statutes, regs).
4. 매 code into structured variables (binary, ordinal, categorical).
5. 매 publish + maintain — 매 LawAtlas-style open data.

### 매 응용
1. AI Act compliance: 매 27 EU 회원국 + 미국 50주의 AI law variation 추적.
2. Public health: 매 LawAtlas COVID closure tracking, opioid policies.
3. Privacy: 매 GDPR vs CPRA vs PIPL 의 cross-walk.

## 💻 패턴

### Pattern 1: Coding scheme YAML
```yaml
# 매 ai_law_codes.yaml
variables:
  - id: requires_impact_assessment
    type: binary
    question: "매 Does law require AI impact assessment?"
  - id: penalty_max
    type: numeric
    unit: USD
  - id: covered_systems
    type: categorical
    values: [foundation_models, biometric, hiring, healthcare, all_high_risk]
jurisdictions: [EU, US-CA, US-CO, KR, UK, CN]
effective_dates: required
```

### Pattern 2: Cross-walk matrix
```python
import pandas as pd

def crosswalk(jurisdictions, variables, codes_df):
    matrix = codes_df.pivot(index="jurisdiction",
                            columns="variable",
                            values="value")
    matrix.to_csv("crosswalk.csv")
    return matrix
```

### Pattern 3: Diff over time
```python
def policy_diff(snapshot_old, snapshot_new):
    changes = []
    for jur in snapshot_new.index:
        for var in snapshot_new.columns:
            if snapshot_old.at[jur, var] != snapshot_new.at[jur, var]:
                changes.append({
                    "jurisdiction": jur, "variable": var,
                    "from": snapshot_old.at[jur, var],
                    "to": snapshot_new.at[jur, var],
                })
    return changes
```

### Pattern 4: LLM-assisted coding (with human verification)
```python
import anthropic
client = anthropic.Anthropic()

def code_statute(statute_text, scheme):
    resp = client.messages.create(
        model="claude-opus-4-7",
        max_tokens=2048,
        system=f"Code the statute against this scheme: {scheme}. Return JSON.",
        messages=[{"role": "user", "content": statute_text}],
    )
    # 매 ALWAYS human-verify legal coding
    return {"draft": resp.content[0].text, "needs_review": True}
```

### Pattern 5: Effective-date timeline
```python
def timeline_view(codes_df):
    return codes_df.sort_values("effective_date")[
        ["jurisdiction", "variable", "value", "effective_date"]
    ]
```

### Pattern 6: Citation chain (provenance)
```python
def store_with_provenance(code, value, statute_section, source_url, retrieved_at):
    return {
        "code": code, "value": value,
        "citation": {"section": statute_section, "url": source_url, "retrieved": retrieved_at},
    }
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| 매 single-org compliance | Standard compliance audit |
| 매 multi-jurisdiction policy comparison | Policy Surveillance |
| 매 academic causal inference (does law X cause outcome Y?) | Policy Surveillance + econometrics |
| 매 real-time regulatory news | News tracker (NOT surveillance) |
| 매 AI Act multi-state US tracking | Policy Surveillance + LLM-draft + lawyer review |

**기본값**: 매 LawAtlas-style codebook + git versioning + LLM-draft + human verification.

## 🔗 Graph
- 응용: [[AI 거버넌스 정책(AI Usage Policy)|AI Governance]] · [[GDPR Compliance]]
- Adjacent: [[EU AI Act]]

## 🤖 LLM 활용
**언제**: 매 first-pass coding of large statute corpus, 매 cross-walk drafting, 매 diff summarization.
**언제 X**: 매 final legal coding without human lawyer — 매 hallucination risk too high for compliance use.

## ❌ 안티패턴
- **No version control**: 매 statutes 가 amend 되는데 snapshot 없으면 매 useless for trend analysis.
- **Coding without scheme**: 매 ad-hoc tags — 매 inter-coder reliability ~0.
- **LLM-only coding**: 매 hallucinated citations — 매 catastrophic for legal use.
- **Single jurisdiction silo**: 매 policy surveillance 의 가치 = comparison.

## 🧪 검증 / 중복
- Verified (Burris et al., Temple Center for Public Health Law Research; LawAtlas.org).
- 신뢰도 A (academic + practitioner standard).

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — Burris method, AI Act 응용, LLM augmentation |