Mood and breakdowns

Owns a per-agent mood value equal to a baseline plus the sum of active modifiers (needs, memories, environment, traits), usually smoothed toward that target, and a set of thresholds. While mood sits below a threshold a hazard rate rolls for a break whose severity rises with depth; during a break the agent leaves player control. High mood can roll for positive states. RimWorld: base mood 22 to 42 by difficulty; the bar moves toward its target at up to +12 per hour rising and -8 falling; break thresholds are 35%, 20%, and 5% (traits shift the minor one, capped 1-50%), with mean times between breaks of about 4, 0.8, and 0.5 days; above 49% mood, inspirations arrive with a mean interval of 410 - 400 x mood days and grant effects such as x1.8 work speed for 8 days. Dwarf Fortress keeps a hidden stress value, scaled by personality facets, that can end in tantrums or depression. Inputs: needs, memories, traits. Outputs: lost labor, destructive acts, temporary buffs.

The player manages risk rather than events: watching mood bars near thresholds, removing stressors, and bracing when a disaster pushes several agents under at once. Breaks are memorable because they are partly random and partly the player's fault.

• baseline per difficulty and weights of each mood input
• threshold levels and trait offsets
• hazard rate per band and severity table
• break catalog (sad wander, binge, tantrum, violence) and whether breaks harm others
• chance and size of positive states at high mood

• Cascades: one break creates negative memories in witnesses, and the group spirals beyond recovery after a random event, which feels unfair.
• Mood that is easy to max removes the system from mid and late game.
• Breaks that destroy hours of work feel punitive rather than dramatic.
• Opaque mood sources make breaks feel random.

• Breakdown contagion: breaks generate negative memories in witnesses that lower their mood, a reinforcing loop.
• Players build mood infrastructure (recreation, beauty, fine meals) as defense, coupling mood to the economy.
• Inspirations reward stable groups with bursts of output, so content colonies pull further ahead.

• Can breaks be generated as character-specific behavior by language-model agents instead of drawn from a fixed catalog, and still be legible?
• What hazard shape keeps breaks dramatic without producing unrecoverable cascades?