Transparency

Scoring Algorithm

How the composite ESG score is calculated from raw survey responses — fully documented for auditability.

Composite Score Formula

ESG Score = (E × 0.40) + (S × 0.30) + (G × 0.30)

40%

Environmental

Energy, GHG, Water, Waste

30%

Social

Labour, Safety, Gender

30%

Governance

Compliance, Certifications

Scores are expressed on a 0–100 scale. The Environmental pillar carries a higher weight reflecting the sector's primary impact area: energy-intensive manufacturing with significant Scope 1 and 2 emissions.

Pillar Score Derivation

Environmental Score (E)

Metric Sub-weight Direction
Energy intensity (kWh/unit)35%↓ Lower better
Renewable energy share25%↑ Higher better
GHG intensity (kgCO₂e/unit)20%↓ Lower better
Water intensity (L/unit)10%↓ Lower better
Waste diversion rate10%↑ Higher better

Social Score (S)

Metric Sub-weight Direction
Wage compliance rate30%↑ Higher better
Workplace safety score30%↑ Higher better
Gender ratio (female %)20%↑ Higher better
Labour rights compliance20%↑ Higher better

Governance Score (G)

Metric Sub-weight Direction
Compliance section scores (avg)60%↑ Higher better
Certifications held (count/max)25%↑ Higher better
Permit validity rate15%↑ Higher better

Normalization Method

Each metric is normalized to a 0–100 score using min-max scaling within the sector peer group:

normalized = (value − min) / (max − min) × 100

For "lower is better" metrics (energy intensity, GHG, water), the formula is inverted so that the most efficient factory scores 100.

Metrics with fewer than 3 data points in the peer group are excluded from scoring and flagged as insufficient_data.

Peer Group Benchmarking

Factories are benchmarked within peer groups determined by factory type (Cut-to-Pack vs Vertically Integrated) and size band. The peer group selection follows a two-level fallback:

  1. Type + size band (if n ≥ 10 factories)
  2. Type only (if n ≥ 10 but size band too small)
  3. All factories (last resort fallback)

A minimum of n = 10 factories is required before computing any percentile rank, to prevent de-anonymization through inference.