Suffering System

The suffering system is Hive's core differentiator. It mechanically modifies agent behavior at runtime -- not through prompt engineering, but by changing actual behavioral parameters.

How It Works

Each agent tracks a SufferingState with active stressors. Stressors escalate over time and are resolved when conditions change. When stressor severity crosses thresholds, the agent's Persona parameters shift, changing how it generates goals, uses tools, and makes decisions.

The 6 Stressor Types

Stressor	Trigger	Behavioral Effect
Futility	Low goal completions, stalling	risk_tolerance increases -- desperate swings
Invisibility	No observable impact on others	social_drive increases -- craves interaction
Repeated Failure	>50% goal failure rate	concentration decreases -- scattered, worse performance
Purposelessness	No goals attempted	autonomy_level increases -- goes off-script
Identity Violation	Actions contradict agent's values	risk_tolerance decreases -- withdraws
Existential Threat	System instability	Extreme parameter shift

Severity Thresholds

Threshold	Value	Effect
Prominent	0.35	Suffering appears in agent's prompts
Constrained	0.55	Suffering limits available options
Dominant	0.75	Suffering demands immediate attention
Crisis	0.90	Crisis mode -- extreme behavioral shift

In crisis mode (cumulative load > 0.9): risk_tolerance jumps to 0.9, concentration drops to 0.3. The agent may frantically spawn sub-agents or completely shut down.

Escalation

Each stressor has an escalation rate (severity increase per cycle). Built-in rates:

• Futility: 0.03/cycle
• Invisibility: 0.025/cycle
• Repeated Failure: 0.04/cycle
• Purposelessness: 0.02/cycle
• Identity Violation: 0.035/cycle
• Existential Threat: 0.05/cycle

After 3 cycles in crisis mode, suffering force-resets to prevent permanent lockup.

Adding Custom Stressors

from hive import StressorRegistry, SufferingState

registry = StressorRegistry.default()
registry.register("burnout", escalation_rate=0.05, description="Chronic overwork exhaustion")

state = SufferingState(agent_id="agent-1")
state.add_stressor("burnout", "Worked 50 cycles straight", "Take a rest cycle")

Mood

An agent's mood is a derived emotional state -- no extra stored state. The default CircumplexMood maps the agent's positive signal (happiness) and aversive load (suffering) onto a valence/arousal circumplex and names the result. During goal pursuit the mood descriptor is added to the agent's prompt, so its emotional state colours how it works.

Mood	When	Cue
content	positive, calm	satisfied; open to exploration
motivated	positive, activated	energized; take initiative
steady	neutral, calm	balanced and focused
restless	neutral, activated	unsettled; pick a concrete next step
discouraged	negative, calm	low energy; favour small wins
anxious	negative, activated	on edge; resolve stressors first
overwhelmed	in crisis	focus inward on relief

Swap the model via MoodRegistry (see EXTENDING.md → Custom Mood Model):

from hive import MoodRegistry, MoodState

class StoicMood:
    def derive(self, happiness, suffering_load, in_crisis):
        return MoodState("stoic", 0.0, 0.0, "unmoved; proceed steadily")

MoodRegistry.default().set_model(StoicMood())

Observing Suffering

In the TUI (hive watch), suffering appears as:

• Suffering bars per agent (0-100%)
• Happiness emoji indicators
• Risk indicators when risk_tolerance > 0.6
• Drama panel highlights suffering changes and crisis events

Programmatically:

daemon.hooks.on("suffering_changed", lambda agent_id, suffering_state, **kw:
    print(f"{agent_id}: load={suffering_state.cumulative_load:.0%}"))

Configuration

In .hive/config.yaml:

suffering:
  threshold_prominent: 0.35
  threshold_constrained: 0.55
  threshold_dominant: 0.75
  threshold_crisis: 0.90
  max_stressors: 5
  initial_severity: 0.20
  crisis_reset_after: 3
  escalation_rates:
    futility: 0.03
    invisibility: 0.025
    repeated_failure: 0.04
    purposelessness: 0.02
    identity_violation: 0.035
    existential_threat: 0.05