Danki Impact Scoring

Methodology Whitepaper for Regulators and Investors

Author

Nambona Yanguere — Danki Studio

Published

February 23, 2026

1 DANKI IMPACT SCORING

Methodology Whitepaper — Scoring Framework, Data Architecture & Validation

v1.0 · January 2025 · For analysts, compliance officers, regulators & LPs

2 Executive Summary

Danki Impact Scoring is a dual-layer scoring engine designed for impact investors who need to go beyond ESG compliance. It produces a composite impact score (0–100) for any investment project, covering 8 impact dimensions weighted toward social outcomes, with full regulatory compliance verification across 5 EU frameworks.

2.0.1 Key differentiators from ESG scoring:

  • Social-first philosophy: Gender & Social Equity (20%) + Social Mobility (15%) + Governance (15%) = 50% of total weight on social foundations
  • Social veto rule: Projects scoring below 30 on Gender or Social Mobility are capped at Amber regardless of environmental performance
  • Regulatory completeness: EU Taxonomy, SFDR (Art. 6/8/9), CSRD/ESRS, TCFD, MiFID II suitability — all checked automatically
  • Explainable by design: Every score traces to a published formula and regulatory standard — zero black box

2.1 Why Not ESG?

ESG ratings measure risk to the company from sustainability factors. Impact scoring measures the company’s effect on the world. These are fundamentally different questions.

A mining company with excellent ESG governance can still destroy communities. A solar farm with perfect environmental metrics can still exploit workers. ESG would score both highly. Danki Impact Scoring would not.

The social veto rule operationalises this distinction: no project achieves a Green band if it fails people, regardless of how well it treats the planet.

3 The Scoring Framework

3.1 Architecture Overview

┌─────────────────────────────────────────────────────────────────┐
│                   DANKI IMPACT SCORING                          │
│                   Scoring Architecture                          │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  INPUT                                                          │
│  ├── Project metadata (country, sector, asset class, size)      │
│  ├── 8 dimension indicators (quantitative + qualitative)        │
│  ├── PAI values (GHG intensity, gender gap, corruption flag)    │
│  └── SFDR classification intent (Art. 6 / 8 / 9)               │
│                                                                 │
│  LAYER 1 — Regulatory Compliance Gate                           │
│  ├── EU Taxonomy: Substantial Contribution + DNSH               │
│  ├── SFDR: 18 mandatory PAI indicators                          │
│  ├── CSRD/ESRS: Double materiality scope                        │
│  ├── TCFD: Physical + transition risk                           │
│  └── MiFID II: Sustainability preference alignment              │
│                                                                 │
│  LAYER 2 — Impact Composite Score (0–100)                       │
│  ├── 8 weighted dimensions (social-first: 50% social weight)    │
│  ├── Social Veto Rule (Gender<30 OR Social<30 → Amber cap)      │
│  └── Band assignment: Red / Amber / Green / Dark Green          │
│                                                                 │
│  LAYER 3 — ML Calibration (roadmap)                             │
│  ├── XGBoost adjustment for country/sector/size interactions    │
│  └── Trained on validated benchmark data                        │
│                                                                 │
│  OUTPUT                                                         │
│  ├── Composite score + band + radar chart                       │
│  ├── Regulatory gate dashboard (pass/fail per framework)        │
│  ├── PAI traffic-light dashboard                                │
│  ├── Warnings & strengths                                       │
│  └── PDF due diligence report                                   │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

3.2 The 8 Impact Dimensions

3.2.1 Social-First Weighting Philosophy

Academic literature (World Bank, UNDP, GIIN) consistently shows that social determinants are upstream of environmental outcomes. Gender equality is the single strongest predictor of long-term development. Social mobility determines whether economic value stays in a territory. Governance is the multiplier — it either amplifies or destroys every other dimension.

The Danki principle: a solar farm built on grabbed land, staffed by imported labour, with a 40% gender pay gap is not impact. ESG would score it green. Danki would not.

Danki Impact Scoring — Dimension Weights
Dimension Weight Key Indicators Regulatory Basis
Gender & Social Equity 20% Gender pay gap, board diversity, pay ratio, anti-discrimination policy SFDR PAI 12–13, CSDDD Art. 3–7, IFC PS2
Social Mobility 15% Local hiring %, living wage ratio, skills uplift, education access OECD Guidelines Ch. V, SDG 1/4/8/10
Governance & Corruption 15% Beneficial ownership transparency, AML flags, CPI score, sanctions SFDR PAI 15–16, EU AML Directive 6
Climate & Environment 18% GHG avoided, energy transition alignment, biodiversity net gain EU Taxonomy Obj. 1–2, TCFD, SFDR PAI 1–6
Pollution & Health 10% Scope 1–3 emissions, toxic exposure, air/soil quality SFDR PAI 1–4, 8–9, EU Taxonomy Obj. 5
Water & Resources 8% Water intensity, circular economy rate, waste EU Taxonomy Obj. 3, SFDR PAI 7–8
Territory & Local Wealth 8% Local GDP contribution, supply chain localisation, tax transparency CSRD ESRS S3, OECD MNE Guidelines
Innovation & Resilience 6% Tech transferability, employment durability, R&D intensity EU Taxonomy Obj. 6, SDG 9

Social total: 50% (Gender 20% + Social Mobility 15% + Governance 15%) — Environmental total: 36% (Climate 18% + Pollution 10% + Water 8%) — Economic total: 14% (Territory 8% + Innovation 6%)

3.3 Social Veto Rule

Important

SOCIAL VETO: If a project scores below 30/100 on Gender & Social Equity OR Social Mobility, the maximum achievable band is Amber, regardless of the composite score.

A project scoring 85/100 overall but 25/100 on gender is capped at Amber with an explicit veto warning.

Regulatory basis for the veto:

  • SFDR PAI 12–13: Gender pay gap and board gender diversity are mandatory disclosure indicators
  • CSDDD Art. 3–7: Companies must identify, prevent and address adverse human rights impacts
  • IFC Performance Standard 2: Labour and working conditions requirements for all IFC investments
  • OECD Guidelines for MNEs, Ch. V: Employment and industrial relations standards

Implementation:

def _assign_band(composite: float, gender_dim: float, social_dim: float) -> tuple:
    veto = gender_dim < 30.0 or social_dim < 30.0
    if composite >= 75:
        raw_band = "Dark Green"
    elif composite >= 55:
        raw_band = "Green"
    elif composite >= 35:
        raw_band = "Amber"
    else:
        raw_band = "Red"

    if veto and raw_band in ("Dark Green", "Green"):
        return "Amber", True   # Capped
    return raw_band, veto and raw_band == "Amber"

3.4 Score Bands

Score Band Decision Table
Band Range Meaning SFDR Alignment
🟢 Dark Green 75–100 Exceptional positive impact across all dimensions Article 9 eligible
🟢 Green 55–74 Meaningful positive impact, minor gaps Article 8+ eligible
🟡 Amber 35–54 Mixed impact, significant improvement needed Article 8 minimum / Article 6
🔴 Red 0–34 Net negative or negligible impact Article 6 only

4 Dimension Scoring Formulas

Each dimension score is computed from observable indicators using transparent, auditable formulas. All scores are on a 0–100 scale.

4.1 Gender & Social Equity (20%)

\[ G = \min\Bigl(100,\; 100 \times (1 - \text{gender\_pay\_gap}) \times \text{board\_diversity} \times \text{policy\_factor}\Bigr) \]

Where:

  • gender_pay_gap ∈ [0, 1]: Ratio of pay gap (0 = no gap, 1 = complete gap)
  • board_diversity ∈ [0, 1]: Proportion of women on board
  • policy_factor ∈ {0.6, 0.8, 1.0}: Anti-discrimination policy strength (none / basic / comprehensive)

Score interpretation: A company with 0% pay gap, 50% board diversity and comprehensive policy scores ~50. To reach 80+, the company needs near-zero gap with majority-female leadership.

4.2 Social Mobility (15%)

\[ S = \min\Bigl(100,\; 100 \times \text{local\_hire\_ratio} \times \text{living\_wage\_ratio} \times \text{skills\_factor}\Bigr) \]

Where:

  • local_hire_ratio ∈ [0, 1]: Proportion of workforce hired locally
  • living_wage_ratio ∈ [0.5, 2.0]: Ratio of median wage to local living wage
  • skills_factor ∈ {0.6, 0.8, 1.0}: Skills uplift programme (none / basic / comprehensive)

4.3 Governance & Corruption (15%)

\[ V = \min\Bigl(100,\; \text{CPI}_{country} \times \text{transparency\_factor} \times (1 - \text{sanctions\_flag})\Bigr) \]

Where:

  • CPI_country ∈ [0, 100]: Transparency International Corruption Perceptions Index
  • transparency_factor ∈ {0.5, 0.75, 1.0}: Beneficial ownership disclosure level
  • sanctions_flag ∈ {0, 1}: Active sanctions or AML flags (binary penalty)

4.4 Climate & Environment (18%)

\[ C = \min\Bigl(100,\; \text{taxonomy\_alignment} \times 100 \times \text{transition\_factor} \times (1 - \text{stranding\_risk})\Bigr) \]

Where:

  • taxonomy_alignment ∈ [0, 1]: EU Taxonomy substantial contribution score
  • transition_factor ∈ {0.6, 0.8, 1.0}: Energy transition pathway strength
  • stranding_risk ∈ [0, 1]: Probability of asset stranding under 1.5°C scenario

4.5 Pollution & Health (10%)

\[ P = \max\Bigl(0,\; 100 - (\text{emission\_intensity} \times 0.5) - (\text{toxic\_exposure} \times 30)\Bigr) \]

Where:

  • emission_intensity: tCO₂e per €M revenue
  • toxic_exposure ∈ {0, 1, 2, 3}: Categorical (none / low / medium / high)

4.6 Water & Resources (8%)

\[ W = \max\Bigl(0,\; 100 - (\text{water\_intensity} \times 10) + (\text{circular\_rate} \times 30)\Bigr) \]

4.7 Territory & Local Wealth (8%)

\[ T = \min\Bigl(100,\; \text{local\_procurement\_rate} \times 100 + \text{tax\_transparency} \times 20\Bigr) \]

4.8 Innovation & Resilience (6%)

\[ I = \min\Bigl(100,\; \text{R\&D\_intensity} \times 500 + \text{tech\_transfer} \times 30 + \text{employment\_durability} \times 20\Bigr) \]

4.9 Composite Score

\[ \text{Danki Score} = \sum_{d=1}^{8} w_d \times D_d \]

Where \(w_d\) are the dimension weights and \(D_d\) are the dimension scores. The composite is then subject to the social veto rule before band assignment.

5 Regulatory Framework Mapping

5.1 Layer 1 — Compliance Gates

Each investment is checked against 5 regulatory frameworks. The output is pass/fail with specific failure reasons.

5.2 EU Taxonomy (Regulation 2020/852)

Check Logic Source
Substantial Contribution Climate score ≥ 60 AND sector is taxonomy-eligible Art. 3, Delegated Acts
DNSH No dimension below 25 Art. 17
Minimum Safeguards Governance ≥ 40 AND Gender ≥ 40 Art. 18, OECD/ILO/UNGP

5.3 SFDR (Regulation 2019/2088)

Classification Logic
Article 9 Composite ≥ 75 AND taxonomy aligned AND no veto
Article 8 Composite ≥ 45 AND taxonomy partially aligned
Article 6 All other products

5.4 CSRD/ESRS (Directive 2022/2464)

  • In scope: Companies with ≥1,000 employees and ≥€450M turnover (post-Omnibus, December 2025)
  • Out of scope projects: Danki uses proxy data and reasonable estimates — a key differentiator vs. tools that require full CSRD disclosure

5.5 TCFD (FSB Recommendations)

  • Physical risk: Country climate vulnerability × sector exposure
  • Transition risk: Carbon intensity × 1.5°C scenario alignment

5.6 MiFID II Sustainability Preferences

  • Taxonomy alignment percentage
  • PAI consideration flag
  • SFDR article classification

6 Exploratory Data Analysis

The following analysis is based on the synthetic benchmark dataset of 2,000 investment projects across 50 countries, 20 sectors and 6 asset classes.

7 ML Calibration Methodology

7.1 Approach: Composite Weighted Index with XGBoost Adjustment

The Danki scoring approach prioritises explainability — a requirement for regulatory acceptance (SFDR Art. 4, MiFID II suitability assessment). The ML layer is a calibration adjustment, not a replacement for the deterministic composite.

7.1.1 Why Composite First, ML Second

ML Approach Comparison
Approach Explainability Regulatory Acceptance Accuracy
Pure ML (black box) Low Rejected by AMF/EBA High
Pure composite (no ML) High Accepted Medium
Composite + ML calibration High Accepted High

7.1.2 XGBoost Calibration Layer (Phase 3 Roadmap)

The ML layer adjusts the deterministic composite for interaction effects that linear weighting cannot capture:

  1. Country x Sector interactions: Renewable energy in Norway vs. Nigeria has fundamentally different risk profiles that a country score alone doesn’t capture
  2. Investment size non-linearity: A EUR 1,500 microfinance project and a EUR 50M infrastructure project are scored on the same 0-100 scale but behave very differently
  3. Temporal calibration: As benchmark data accumulates, the model recalibrates weights to reflect observed impact outcomes

Training pipeline:

Phase 3 Pipeline:
1. Validated benchmark dataset (2,000+ scored projects)
2. Feature engineering: dimension scores + country/sector/size interactions
3. XGBoost regressor: target = expert-validated Danki score
4. SHAP values for every prediction -> full explainability preserved
5. Human-in-the-loop: ML adjustment capped at +/- 10 points from composite
6. Quarterly retraining on new validated data

7.1.3 ML Architecture

7.2 Simulated Feature Importance

8 Data Sources

8.1 Public / Open Data Sources

Public / Open Data Sources
Source Data Provided Coverage Frequency Access
Transparency International CPI Country corruption perception index (0-100) 180 countries Annual Free
World Bank Open Data GDP per capita, Gini, electricity access, education enrollment 217 countries Annual Free
ILO STAT Gender pay gap, labour force participation, working conditions 189 countries Annual Free
UNDP HDI Human Development Index, gender inequality index 191 countries Annual Free
ND-GAIN Index Climate vulnerability and readiness scores 185 countries Annual Free
EU Taxonomy Compass Taxonomy-eligible activities, technical screening criteria EU Ongoing Free
EDGAR (JRC) GHG emission inventories by country and sector Global Annual Free
Global Living Wage Coalition Living wage benchmarks by country/region 35+ countries Annual Free
OECD DAC ODA flows, development finance statistics 180+ countries Annual Free
OpenSanctions Sanctions lists, PEP data, enforcement actions Global Daily Free
UN Comtrade International trade data (supply chain localisation proxy) 200+ countries Monthly Free
Eurostat EU employment, energy, environment statistics EU-27 Quarterly Free
SFDR PAI RTS Annex I 18 mandatory PAI indicator definitions and methodologies EU regulation Regulatory Free
ESMA/EBA Joint PAI Report Supervisory guidance on PAI indicator calculation EU regulation Annual Free

8.2 Proprietary / Commercial Data Sources

Proprietary / Commercial Data Sources
Source Data Provided Use Case Approx. Cost/yr
MSCI ESG ESG ratings, carbon data, PAI metrics, controversy scores Benchmark calibration, PAI pre-fill EUR 30k-100k
Sustainalytics ESG risk ratings, carbon emissions, PAI indicators Cross-validation EUR 25k-80k
CDP Corporate climate disclosures, water security, forests Climate & water enrichment EUR 5k-20k
Refinitiv ESG 630+ ESG metrics, controversies, carbon data Large-cap coverage EUR 20k-60k
Bloomberg ESG ESG scores, supply chain data, governance metrics Terminal-integrated scoring Bloomberg Terminal
Preqin PE/VC fund data, impact fund benchmarks Private market calibration EUR 15k-50k
S&P Trucost Environmental cost data, carbon earnings at risk Pollution & climate dimensions EUR 20k-40k
RepRisk ESG risk incidents, controversy tracking Governance & corruption alerts EUR 10k-30k
Clarity AI Impact measurement, SDG alignment, PAI automation Full PAI dashboard pre-fill Custom
Moody’s ESG Climate risk, physical risk, transition risk scoring TCFD compliance layer EUR 15k-45k
Note

Danki v1.0 operates entirely on public data. The scoring engine requires only country-level indicators (CPI, HDI, climate vulnerability) plus project-level inputs from the user. Proprietary sources are optional enrichments for Phase 3 ML calibration and automated PAI pre-fill.

8.3 Data Integration Architecture

9 Validation Statistics

9.1 Benchmark Dataset Summary

The synthetic benchmark dataset contains 2,000 investment projects designed to stress-test the scoring engine across extreme conditions.

9.2 Scoring Engine Validation

Three representative test cases demonstrate the engine produces economically coherent results:

Validation Test Cases
Test Case Country Sector Size Danki Score Band SFDR Veto
Renewable energy France Solar PV EUR 5M 77.2 Dark Green Article 9 No
Extractive industry Nigeria Oil & Gas EUR 250k 28.7 Red Article 6 No
Microfinance Bangladesh Financial Inclusion EUR 8k 54.4 Amber Article 8 No
Digital infra (forced veto) Nigeria Telecoms EUR 1M 65.0 Amber (capped) Article 8 Yes — Gender

10 Deployment

10.1 Technical Requirements

  • Python 3.10+ with pandas, numpy, matplotlib
  • Quarto 1.4+ for report rendering
  • No database required — the app runs entirely client-side in the browser

10.2 Installation & Build

# 1. Clone the repository
git clone https://github.com/your-org/danki-impact-scoring.git
cd danki-impact-scoring

# 2. Install Python dependencies
pip install -r requirements.txt

# 3. Generate the benchmark dataset (first time only)
cd src
python generate_data.py
# -> data/processed/investment_impacts.csv (2,000 projects)

# 4. Render the whitepaper
cd ../report
quarto render index.qmd --to html --output-dir ../docs

# 5. Render the scoring app
quarto render app.qmd --to html --output-dir ../docs

# 6. Deploy to GitHub Pages
quarto publish gh-pages

10.3 Deployment Options

Deployment Options
Option Setup Best For
GitHub Pages quarto publish gh-pages Public demo, LP access
Internal server Copy docs/ to any static file server Analyst team, compliance
Offline Open docs/app.html directly in browser Field due diligence

The scoring app is fully static — no server, no database, no API calls. All scoring logic runs in JavaScript in the user’s browser. This means:

  • Zero infrastructure cost
  • Works offline for field due diligence in low-connectivity environments
  • No data leaves the user’s device — important for pre-investment confidentiality
  • Instant deployment — any static hosting works (GitHub Pages, Netlify, S3, internal server)

10.4 Project Structure

danki-impact-scoring/
|-- data/
|   +-- processed/
|       +-- investment_impacts.csv    <- 2,000-row benchmark dataset
|-- docs/                             <- GitHub Pages output
|   +-- assets/
|-- notebooks/
|   +-- eda_marimo.py                 <- exploratory analysis
|-- report/
|   |-- index.qmd                     <- this methodology whitepaper
|   |-- app.qmd                       <- Danki scoring app
|   +-- report-style.css              <- shared brand stylesheet
|-- src/
|   |-- generate_data.py              <- synthetic dataset generator
|   |-- scoring.py                    <- composite scoring engine
|   +-- viz.py                        <- chart functions
|-- requirements.txt
|-- _quarto.yml
+-- README.md

11 Limitations & Roadmap

11.1 Current Limitations (v1.0)

  1. Synthetic data only: The benchmark dataset is generated, not sourced from real investments. Dimension scores are calibrated to be realistic but are not validated against actual outcomes.
  2. No temporal dimension: Scores are point-in-time. Impact trajectories (improving vs. declining) are not captured.
  3. Equal treatment of PAI indicators: All 18 mandatory PAI indicators are tracked but not differentially weighted within their parent dimensions.
  4. No automated data ingestion: Users manually input project data. API connections to data providers are not yet implemented.
  5. ML layer is simulated: The XGBoost calibration layer is designed but not yet trained on validated data.

11.2 Roadmap

Development Roadmap
Phase Deliverable Timeline
Phase 1 (complete) Scoring engine, whitepaper, interactive app, synthetic benchmark Q1 2025
Phase 2 Real investment data onboarding, expert validation of dimension scores Q2 2025
Phase 3 XGBoost calibration layer with SHAP explainability Q3 2025
Phase 4 API connections to public data sources (TI CPI, World Bank, ND-GAIN) Q3 2025
Phase 5 Proprietary data integration (MSCI, CDP, RepRisk) Q4 2025
Phase 6 Portfolio-level scoring (aggregate multiple projects into fund-level impact) Q1 2026
Phase 7 Temporal impact tracking and trajectory scoring Q2 2026

12 References

  1. EU Taxonomy Regulation (2020/852) — Official Journal of the European Union
  2. SFDR (2019/2088) — Regulation on sustainability-related disclosures in the financial services sector
  3. CSRD (2022/2464) — Corporate Sustainability Reporting Directive
  4. CSDDD (2024/1760) — Corporate Sustainability Due Diligence Directive
  5. TCFD Recommendations (2017) — Task Force on Climate-related Financial Disclosures, FSB
  6. MiFID II (2014/65/EU) — Markets in Financial Instruments Directive, as amended for sustainability preferences
  7. ESMA/EBA Joint Report on PAI (JC 2024/68, October 2024) — Principal Adverse Impact disclosures
  8. EU Omnibus Simplification Package (December 2025) — CSRD scope narrowing to 1,000+ employees / EUR 450M turnover
  9. MSCI SFDR Adverse Impact Metrics Methodology (2024) — PAI calculation standards
  10. Transparency International CPI (2024) — Corruption Perceptions Index methodology
  11. UNDP Human Development Report (2024) — HDI and Gender Inequality Index
  12. ND-GAIN Country Index (2024) — Climate vulnerability and readiness
  13. GIIN — Global Impact Investing Network, Annual Impact Investor Survey (2024)
  14. IMP — Impact Management Project, Five Dimensions of Impact framework
  15. IFC Performance Standards (2012) — Environmental and Social Sustainability
  16. OECD Guidelines for Multinational Enterprises (2023 update) — Responsible business conduct

Danki Impact Scoring v1.0 · Methodology Whitepaper Designed for impact investors who refuse to confuse compliance with change.