📦 모든 산출물 (Outputs Showcase)

CINA 프로젝트의 전 산출물 — 학술 페이퍼, 장관급 브리핑, 정량 평가 보고서, Stage 2 시각화, 데이터셋. 모든 항목은 GitHub repository에서 직접 열람 가능합니다.

3
학술 논문 / 브리핑
8
정량 검증 보고서
7
Stage 2 Figures
15
방법론 문서
225
데이터 manifest
98
Stage 1 stances
📑 Table of Contents

§1 Deliverables — 학술 페이퍼 + 장관급 브리핑 + 정량 검증

Track A (수업 제출 — 장관급 브리핑) + Track B (학술 투고 — 페이퍼) 듀얼 트랙 산출물 일체.

📄 Headline 산출물 (3)

Track B · Paper

📘 paper.md

3,428 words · 23 references · 7 sections + KO summary. Target venue: Global Environmental Change / NeurIPS CCAI 2026 / 한국정책학회보. Inline figure refs (1-6).

📐 EN + KO 📊 23 citations View on GitHub →
Track A · Briefing (KO)

🇰🇷 ministerial_briefing_ko.md

외교부 장관급 협상 전략 브리핑 (한국어). 9 섹션 + 4 부록(증거 추적성, 데이터 lineage, 불확실성, LLM provider attribution).

📐 9 sections 📎 4 appendices View on GitHub →
Track A · Briefing (EN)

🇬🇧 ministerial_briefing_en.md

English version of the ministerial briefing — 310 lines. Direct translation with same evidence-grounded structure for international circulation.

📐 310 lines 📐 mirror of KO View on GitHub →

🔬 정량 검증 보고서 (8)

Finding 1

🌍 IRR_Korea.md

IRR_Korea = 0.653, CI [0.55, 0.71]. 30/30 cells (5 NAP 섹터 × 6 GGA 이슈) Korean NAP × GGA crosswalk 적용.

IRR = 0.653 View →
Finding 2 ⭐

🇧🇷 IRR_Brazil.md

Brazil Translation Gap Δ = 0.304 CONFIRMED (Plano Clima 67% → GGA voluntary 48%). Putnam × Howlett gap 정량화 — NeurIPS CCAI signature finding.

Δ = 0.304 View →
Finding 3

📜 L25_formula_control_evidence.md

L.25 pre-crystallized formula 가설 — Tallberg 2010 procedural authority + Steinberg 2002 + Goh 2007 통합 검증.

NeurIPS CCAI 후보 View →
Finding 4

📊 realist_b0_statistics.md

Realist B0 F1 = 0.560, McNemar p<0.0001, κ=0.216. Constructivist + frame variables 경험적 정당성 확보.

F1 = 0.560 View →
Finding 5 ⭐

🌐 AILAC_norm_entrepreneur_quantification.md

AILAC norm entrepreneur 가설 — Finnemore-Sikkink (1998) 4 criteria 3.5/4 PASS. NES = 0.86. Tactical diplomatic framing 정량화.

NES = 0.86 View →
Phase 5 Final ⭐

📈 evaluation_report.md

Phase 5 4-task 정량 검증. Spearman ρ=0.658, P@3=R@3=1.00, Mean 4.53/5, α=0.905. Ablation A0-A5 + 5/5 PASS.

5/5 PASS View →
Crosswalk

📋 korean_nap_gga_crosswalk.csv

Korean National Adaptation Plan × GGA 5 NAP × 6 GGA = 30 cells. IRR_Korea 산출 기반.

30 cells View →
Pre-analytic

🎯 country_selection.md / sector_focus.md / agenda_matrix.md

Brazil 선정 근거 + Adaptation 섹터 정당성 + COP30 6-issue 매트릭스 (수업 과제 사전 분석).

3 docs Folder →

§2 Figures — Stage 2 시각화 (7 PNG · 300 dpi)

모든 figure는 data/processed/figures/ 또는 deliverables/에서 생성 — matplotlib + networkx, reproducible via src/stage2_graph/generate_figures.py.

Country × Issue Stance Heatmap

Figure 1. Country × Issue Stance Heatmap

13 countries × 6 issues stance tensor (Stage 1 산출). 적색=opposing, 청색=supporting. 페이퍼 §3 anchor figure.

Procedural Authority Distribution

Figure 2. Procedural Authority Distribution

Tallberg 2010 framework: chair_role + pen_holder + drafts_text 신호 분포. Brazil COP30 chair authority 정량화.

Frame Consistency by Country Group

Figure 3. Frame Consistency by Group

5-frame typology (scientific/justice/sovereignty/security/development) × 그룹별 빈도. AOSIS=justice, Saudi+India=sovereignty.

Centrality Top-K

Figure 4. Centrality Rankings

5 centrality (PageRank·betweenness·eigenvector·degree·closeness) Top-K. Brazil 의장국 + EU 펜홀더 우위.

Country Similarity Network with Leiden Communities

Figure 5. Leiden Community Network

Leiden 2 communities: C0={Brazil, Multi, AGN, EU} development / C1={AOSIS, India, Korea, LMDC} mixed. Keohane-Victor 'horizontal cleavage' 정량 검증.

Hedging × Red Line 2D

Figure 6. Hedging Density × Red Line 2D

AILAC (high hedging + mod RL — tactical), AOSIS (low hedging + high RL — principled), LDC (mod + low — vulnerable). Finnemore-Sikkink norm typology 시각 검증.

Hedging vs Red Line — 보강 v2

Figure 7. Hedging vs Red Line (보강판)

추가 그룹 라벨링 + 95% confidence ellipse 적용.

§3 Evaluations — Phase 5 4-Task + Ablation A0-A5

0.658
Task A · Spearman ρ ✅
0.42
Task B · ARI proxy ✅
1.00 / 1.00
Task C · P@3 / R@3 ⭐
4.53/5
Task D · Mean panel ✅
0.905
Task D · Krippendorff α ✅
5/5
Quality Gates PASS ⭐

🔍 Ablation Study (A0-A5)

AblationRemovedSpearman ρ Δ
A0 (full)— (baseline)0.658
A1Bayesian CI−0.05
A2NATO 4-axis−0.07
A3Frame typology−0.04
A4Evidence grounding−0.15
A5Multi-LLM ensemble−0.06

Evidence grounding (A4) is the single most load-bearing component — 페이퍼 §5.3 핵심 발견.

Phase 5 raw

📊 evaluation_report.json

raw 평가 데이터 (machine-readable). Task A/B/C/D + Ablation 전체 메트릭 + per-record breakdowns.

Pipeline

🐍 src/evaluation/run_full_evaluation.py

Phase 5 4-task 평가 재현 스크립트. python -m src.evaluation.run_full_evaluation으로 모든 메트릭 재생성.

§4 Methodology Documents — 학술 방법론 (15 docs)

논문 §2-3 (이론·방법론)의 직접 구성 요소. docs/ 폴더 내 15개 markdown 일체.

Theory

01 Theoretical Foundations

Regime Complex (Keohane-Victor) · Two-Level Games (Putnam) · Issue Linkage (Tollison-Willett) · Epistemic Communities (Haas)

View →
Method

02 Methodology

Three-stage pipeline 공식 정의. Stage 1 LLM extraction · Stage 2 graph analysis · Stage 3 generation.

View →
Schema

03 Data Architecture

StanceRecord schema v1.3 + 모든 필드 정의 + JSONL 포맷 사양.

View →
Stage 1

04 Stage 1 Stance Extraction

Multi-sample k=5 + Bayesian credible interval + Platt calibration + evidence_quote 강제.

View →
Stage 2

05 Stage 2 Graph Analysis

R-GAT 이론 사양 + Leiden community + Apriori-style hyperedge mining.

View →
Stage 3

06 Stage 3 Briefing Generation

Graph-Grounded Generation + 7-rule post-hoc verification.

View →
Eval

07 Evaluation Protocol

Task A-D 사양 + threshold 정의 + ablation matrix.

View →
Position

08 Novelty Positioning

학술 기여 C1-C10 + baseline 비교 (NegotiateCOP/RICE-N/Castro 2025).

View →
Brief

09 Ministerial Briefing Template

Track A 산출 양식 — 9 섹션 + 4 부록 사양.

View →
Council

10 Council Protocol

5-agent council 구조 + Phase A-E + Quality Gate 5종.

View →
Sources

11 Source Catalog

Tier 1-4 소스 분류 + 14 source systems + per-source license.

View →
Plan

12 Data Collection Master Plan

225 manifest entries 수집 계획 + checkpoint resumability.

View →
Reference

13 Reference Tables

20 countries × 12 groups × 6 issues × 16 sessions identifiers.

View →
v1.3

14 Schema v1.3 Changes

NATO 4-axis + frame_type + procedural_signals 추가 changelog.

View →
Stage 2 features

15 Stage 2 Features (Procedural Authority)

chair_status + drafts_text edge type + chair authority weighting.

View →

§5 Data Architecture — Manifest + Schema

225
Manifest entries
14
Source systems
~720MB
Raw data (Zenodo 별도)
100%
License + sha256 추적
98
Stage 1 stances
2
Leiden communities
Public

📋 data/manifest/manifest.jsonl

225 raw 문서 metadata 추적 (license, sha256, source, version, collected_at).

View JSONL →
Public

📊 coverage_summary.json

소스별 커버리지 통계 — 14 source systems 수집 현황.

View JSON →
Web demo

🌐 docs/web/assets/data.json

웹사이트 시각화에 임베딩된 실데이터. Stage 1 stances + Leiden + IRR + Task A-D.

View JSON →

§6 Code — Pipeline + Council + Evaluation

전 파이프라인은 MIT 라이선스로 GitHub에 공개 — 재현 가능한 reproducibility 우선.

CLI

🚀 src/pipeline.py

End-to-end CLI: python -m src.pipeline --country Brazil --cop 30 --sector adaptation --stages 1,2,3.

View →
Stage 1

🤖 src/stage1_extract/

5 LLM provider abstraction (Gemini/Groq/Ollama/OpenRouter/Anthropic) + multi-sample + calibration.

Folder →
Stage 2

🕸️ src/stage2_graph/

NetworkX heterogeneous graph + Leiden (igraph) + 5 centralities + figure 생성.

Folder →
Stage 3

📝 src/stage3_brief/

Graph-Grounded Generation + 7-rule verification.

Folder →
Eval

📈 src/evaluation/

Phase 5 4-task 평가 + Ablation A0-A5 + Krippendorff α.

Folder →
Council

🏛️ src/council/

5-agent council Phase A-E orchestration + state_manager + quality gate.

Folder →
Collect

📥 src/collect/

26 collectors (UNFCCC, NDC, ENB, IPCC, IMF, PRIMAP, WRI, Brazil, Korea, Castro 2025).

Folder →