Skill & Personality

Skill & Personality

These two config sections work together to define how competent a trainer is and how they play. Skill determines whether the AI even attempts to make smart decisions, while personality adds flavor and variance to those decisions.

Skill Config

Controls how “smart” the AI is at a fundamental level. Each turn, the AI rolls a skill check — if it fails, it picks a random move instead of scoring all options. Higher skill levels fail less often.

{
  "skill": {
    "level": 5,
    "predict_opponent_moves": true,
    "consider_type_matchups": true,
    "use_damage_calc": true
  }
}

Skill Fields

Field	JSON Key	Type	Default	Description
Level	level	int (0—5)	5	Each turn, the AI rolls a skill check. If it fails, it picks a random move and skips all scoring. Level 0 always fails. Level 5 never fails.
Predict Opponent Moves	predict_opponent_moves	bool	true	Whether the AI predicts what moves the opponent might use. Affects threat evaluation.
Consider Type Matchups	consider_type_matchups	bool	true	Whether type effectiveness is factored into move and switch scoring. When false, the AI only looks at raw base power.
Use Damage Calc	use_damage_calc	bool	false	When true, move scores are based on estimated damage percentage instead of just base power. Considers stats, abilities, items, weather, and terrain. More accurate but computationally heavier.

Skill Level Breakdown

Level	Behavior
0	Always picks a random move. All scoring is bypassed.
1	Picks randomly ~80% of the time. Very unreliable.
2	Picks randomly ~40% of the time. Occasionally competent.
3	Picks randomly ~20% of the time. Decent player.
4	Picks randomly ~5% of the time. Strong player.
5	Never random. Always uses the full scoring system.

Skill vs. max_select_margin

Skill level controls whether the AI attempts to score moves at all. max_select_margin (from the overview) controls how much randomness exists within the scoring system. A skill-5 trainer with max_select_margin: 0.5 always scores moves but often picks suboptimal ones. A skill-3 trainer with max_select_margin: 0.0 sometimes picks randomly but is deterministic when it does score.

Personality Config

Gives the AI a distinct personality that subtly influences its decisions. Two trainers with the same config will play slightly differently based on their personality traits.

{
  "personality": {
    "enabled": true,
    "base_personality": {
      "aggression": 0.8,
      "risk_tolerance": 0.6,
      "setup_preference": 0.7,
      "switchiness": 0.3,
      "item_conservatism": 0.4
    },
    "trait_variance": 0.1,
    "consistency_within_battle": true
  }
}

Personality Fields

Field	JSON Key	Type	Default	Description
Enabled	enabled	bool	true	Whether personality traits affect scoring.
Base Personality	base_personality	object	Random	If not specified, a random personality is generated for each battle. If specified, this is the center that variance is applied around.
Trait Variance	trait_variance	double (0.0—0.5)	0.15	How much each trait can deviate from the base. 0.0 means identical every battle. 0.3 means noticeably different each time.
Consistency Within Battle	consistency_within_battle	bool	true	Whether the personality stays constant during a single battle versus re-rolling each turn.

Personality Traits

All traits range from 0.0 to 1.0:

Trait	JSON Key	At 0.0	At 1.0
Aggression	aggression	Defensive, prefers safe plays	Aggressive, favors offensive moves
Risk Tolerance	risk_tolerance	Conservative, picks reliable moves	Risk-taker, favors high-risk high-reward moves (low accuracy, recoil)
Setup Preference	setup_preference	Never uses setup moves	Heavily favors setup moves (Swords Dance, Dragon Dance, etc.)
Switchiness	switchiness	Stays in almost always	Switches frequently to maintain type advantage
Item Conservatism	item_conservatism	Uses items freely whenever helpful	Hoards items, only uses them in emergencies

How Personality Affects Scoring

Each trait applies a multiplier to the relevant action type. The multiplier is calculated using linear interpolation (lerp) between a minimum and maximum value.

For example, with aggression = 0.8:

• Damaging moves get a x1.3 boost (lerp from 0.7 at aggression=0 to 1.3 at aggression=1)
• Status moves get a corresponding reduction

This means an aggressive trainer will lean toward attacking moves even when a status move might score similarly, while a defensive trainer (low aggression) will lean toward safe, non-damaging plays.

Personality is Applied Last

Personality modifiers are applied after all other scoring steps — after move scoring rules, action priorities, and gimmick evaluation. This means personality is a final nudge, not a dominant force. A well-constructed action priority with a high score_bonus will still override personality preferences.