Overview of the Scoring System
The LEWS scoring system is based on a linear weighted model that combines seven key variables to produce a 0-100 lock-in risk score. The methodology is inspired by GiveWell-style weighted scoring, early-warning systems in AI governance, and epidemiological risk models.
Core Variables and Weights
The scoring formula uses the following weights:
0.30 × scale_norm +
0.20 × suffering_norm +
0.15 × sentience_norm +
0.20 × momentum_norm +
0.10 × advocacy_gap_norm +
0.05 × time_to_lockin_norm
Where each variable is normalized to a 0-1 scale before applying the weights.
Variable Normalization
All inputs are normalized to a 0-1 scale using the following approach:
- Population Scale: Normalized by comparing to historical maximums in similar systems
- Senience Probability: Already on a 0-1 scale (0-100%)
- Suffering Intensity: Normalized using the -10 to +10 valence scale
- Industry Momentum: Normalized based on historical growth rates
- Advocacy Gap: Normalized with 0 being strong advocacy and 1 being no advocacy
- Time to Lock-in: Normalized with 0 being already locked in and 1 being far from lock-in
Data Sources
The scoring methodology relies on multiple authoritative data sources:
Population Data
- Our World in Data (slaughter numbers)
- FAO Aquaculture Databases
- Rethink Priorities (insect, shrimp, fish estimations)
- Rowe (2020-2023) for insect-scale estimates
- Mood & Brooke for fish and shrimp
Sentience Research
- Rethink Priorities Sentience Project (2020-2023)
- Birch, Browning, Crump papers on invertebrate sentience (2022-2023)
- EFSA insect sentience evaluations
- Cambridge Declaration on Consciousness
Suffering Intensity
- Welfare Footprint Project
- Peer-reviewed ethology and animal welfare science
- Species-specific welfare audits
- Behavioral restriction indicators
Scoring Stages
The system classifies lock-in risk into three stages:
Research Stage
(0-30) Early development with limited deployment
Scaling Stage
(31-65) Growing adoption and institutional support
Lock-in Approaching
(66-100) System becoming self-reinforcing and difficult to change
Historical Calibration
The scoring methodology was calibrated using historical examples of lock-in:
- Battery cages in poultry (locked in between 1950-1970)
- Modern factory farming systems for chickens
- Aquaculture development patterns
- Insect farming industry progression
These historical patterns help establish baseline weights and thresholds for the model.
Pattern Matching
For trajectory comparison, the system uses simple similarity matching to determine how current technologies compare to historical examples:
Output: "Insect farming today ≈ 1952"
This comparison helps users understand how close current systems are to historical lock-in points.
Quality Assurance
The methodology includes several quality assurance measures:
- Explicit uncertainty ranges for all estimates
- Sensitivity analysis for key parameters
- Historical validation against known outcomes
- Peer review of weight assignments
- Regular calibration updates
Limitations and Assumptions
The scoring methodology makes several important assumptions:
- Linear relationships between variables (simplified version)
- Historical patterns predict future outcomes
- Quantifiable metrics correlate with actual risk
- Current trends continue without major disruptions
These assumptions are clearly communicated to users to ensure appropriate interpretation of results.