Skip to main content
HomeHow the Model Evolved
Model History

How the Model Evolved

Beastsora began as a straightforward stat comparison tool. Over time it was refined to better reflect what actually matters in a real encounter: environment, maneuverability, weapon usability under conditions, and the ability to explain every result in terms a non-expert can understand.

Where It Started

The first version of the comparison engine mapped seven biological categories directly to a weighted score. It worked well for clear mismatches. A blue whale versus a house cat produces an obvious result under almost any model. The problem was the middle ground.

A tiger and a crocodile share similar weight ranges. A great white shark and a killer whale overlap on several metrics. When two animals are plausibly competitive, a flat stat-comparison tends to produce outcomes that feel arbitrary, because it treats all environments as equivalent and all weapons as equally deployable.

Addressing this required moving away from raw stats and toward derived biological capabilities: traits that already incorporate the context of how and where an animal actually operates.

Model Evolution Phases

The comparison engine has been refined in stages, moving from a simple weighted stat model toward a more realistic and explainable biological matchup system.

Phase 1Core Comparison Model
  • Initial weighted comparison engine across seven biological categories
  • Stat-driven matchup system using body mass, speed, and aggression
  • Categories: size, strength, speed, weapons, endurance, aggression, intelligence
  • Win probability output as a single percentage per animal
Phase 2Transparency Update
  • Added per-category factor breakdown visible to users
  • Introduced Key Advantages section showing decisive traits
  • Added raw model inputs table (weight, speed, bite force)
  • Added Assumptions and Model Limits disclosure
Phase 3Human Comparison Layer
  • Added human-vs-animal comparison pages for all 48 species
  • Size, speed, and bite force placed in human context
  • Introduced visual stat comparison summaries
  • Extended SEO coverage to human-vs pages
Phase 4Environment Update
  • Added four selectable environments: land, shallow water, deep water, aerial advantage
  • Environment modifies mobility and weapon effectiveness before scoring
  • Semi-aquatic animals gain advantages in shallow water; terrestrial animals are penalised in water
  • Aerial advantage environment added for birds of prey and flying species
Phase 5Derived Trait Model
  • Engine rebuilt around seven derived combat traits rather than raw stats
  • Introduced: size power, mobility, weapon effectiveness, durability
  • Raw data (weight, speed, PSI) now feeds derived traits, not the final score directly
  • Improved explainability: each factor now corresponds to a biological capability
Phase 6Realism Pass
  • Added targeted weapon-type penalties: charging, tusks, and body-mass weapons lose effectiveness in deep water
  • Corrected mobility multipliers for aquatic animals in shallow and land environments
  • Tuned global environment modifiers against a benchmark matrix of iconic matchups
  • Lowered bite force reference ceiling so mid-range bites are scored more accurately
Planned
PlannedEngagement Viability Modelling

Estimate whether an animal can effectively engage in the selected environment, separate from raw mobility. This helps model cases like beached whales, sharks in shallow water, and limited land performance from fully aquatic animals.

PlannedGroup Behaviour Modelling

Introduce limited treatment of pack hunting, schooling, coordinated attacks, and social defensive behaviour for species where group dynamics materially change outcomes.

PlannedTerrain Modelling

Differentiate between open ground, dense cover, marsh, shoreline, and ice-like surfaces so that terrain affects traction, movement, and ambush potential more precisely.

PlannedInjury Probability Modelling

Model how likely animals are to suffer disabling injury early in an encounter, helping represent fragile speed-based species versus highly durable animals more realistically.

PlannedConfidence & Data Quality Layer

Surface how complete or uncertain the underlying biological data is for each matchup, especially for poorly studied or rarely measured species.

Planned items represent areas of active exploration, not feature commitments.

What the Model Considers Today

The current engine scores every matchup across seven derived combat traits. Each trait is calculated from raw biological data and then adjusted for the selected environment before being weighted into the final probability.

Size Power
18%

A blend of maximum and average body mass. Larger animals carry greater momentum, reach, and physical presence in an encounter. Mass is the single strongest raw predictor of combat outcomes across species.

Mobility
22%

How effectively an animal can move, reposition, and engage in the selected environment. Top speed is adjusted for body plan and environment. A great white shark moves very differently in open water than in the shallows.

Weapon Effectiveness
20%

How dangerous an animal's natural weapons are in real combat conditions. Combines bite force in PSI with weapon variety (claws, talons, venom, tusks, charge, constriction). Some weapons are penalised in water environments where they lose leverage or traction.

Durability
13%

How hard an animal is to stop. A combination of body mass, endurance, and physical armoring (thick hide, shell, dense bone). A durable animal can absorb damage and keep engaging.

Endurance
12%

The capacity for sustained effort. Scored 1 to 10 based on documented field behaviour, aerobic capacity, and the ability to maintain combat output over an extended encounter.

Aggression
8%

Likelihood to initiate and persist in an aggressive encounter. Scored 1 to 10. Captures unprovoked attack likelihood and combat tenacity, and the willingness to press an attack under pressure.

Intelligence
7%

Tactical adaptability and the complexity of hunting or defensive strategies. Scored 1 to 10. Higher-intelligence animals tend to exploit weaknesses, adapt mid-encounter, and avoid unfavourable exchanges.

Why Explainability Matters

Beastsora is built to be inspectable. Every result can be broken down into the seven contributing factors, and every factor is derived from documented biological data rather than opaque machine-learning weights.

The model is not intended to simulate every real-world encounter. Real encounters depend on terrain, individual condition, element of surprise, and a hundred contextual variables no model can fully replicate. What it can do is compare biological advantages in a structured, consistent, and understandable way.

That means being honest when outcomes are uncertain. A 53-to-47 split is not a confident prediction. It means the animals are closely matched and the result could go either way. The result labels (Slight Edge, Clear Edge, Strong Advantage, Overwhelming Advantage) are designed to communicate this honestly, without overstating confidence.

Every matchup page shows the derived factor scores, the decisive traits, and the raw inputs used to generate them. If you disagree with a result, the breakdown gives you somewhere specific to look.

Limitations

No model can capture every real-world variable. Terrain, individual condition, encounter dynamics, and the element of surprise all affect real outcomes in ways that a biological capability model cannot fully represent.

  • Extreme body size continues to influence derived scores (size power, durability) even when an animal's mobility is reduced by environment. Very large animals may appear more competitive than a real-world encounter would reflect.
  • Behavioural scores (aggression, intelligence) are expert-calibrated ordinal estimates, not precise empirical measurements. They carry inherently greater uncertainty than physical metrics.
  • Data quality varies significantly across species. Deep-sea animals, filter feeders, and rarely-studied species often have incomplete or absent speed and bite force records.
  • Where published measurements are unavailable, conservative estimates are used rather than zero-values. This is a deliberate choice to avoid systematically penalising species with limited field study.
  • The model represents a single encounter between two healthy adult specimens. Pack dynamics, territorial behaviour, and population-level advantages are not captured.

See the full scoring methodology

Data sources, confidence levels, and reference literature.

Read the Methodology