Example 07: Hospital ICU Ward¶
The deepest type hierarchy in the library. 6 patients with organ-level physiology, 4 nurses with workload/fatigue dynamics, bureaucratic state (insurance, staffing, bed pressure), and family units. Declares the causal structure of a hospital ward as a type hierarchy.
Source: examples/07_hospital_icu.py
Results¶
5×4 command center figure: Per-patient vitals (HR, BP, SpO2, RR), organ system heatmaps, deterioration risk trajectories, nurse workload/fatigue, bureaucratic pressure indicators, and training curves — dark command center aesthetic.
Animation: Ward monitor dashboard with patient vital signs, nurse status panels, alert system, bed pressure and staffing gauges, shift handoff indicators.
Type hierarchy¶
@dataclass
class CardiovascularSystem:
heart_rate: Field = Field(1, 2, period=1)
blood_pressure: Field = Field(1, 4, period=2)
cardiac_output: Field = Field(1, 2, period=5)
@dataclass
class RespiratorySystem:
spo2: Field = Field(1, 2, period=1)
respiratory_rate: Field = Field(1, 2, period=1)
ventilator_settings: Field = Field(1, 4, period=2, is_output=False)
@dataclass
class RenalSystem:
urine_output: Field = Field(1, 2, period=12)
creatinine: Field = Field(1, 1, period=24)
electrolytes: Field = Field(1, 4, period=24)
@dataclass
class NeurologicalSystem:
consciousness: Field = Field(1, 4, period=6)
sedation_level: Field = Field(1, 2, period=4)
pain: Field = Field(1, 2, period=2)
delirium_risk: Field = Field(1, 2, period=12, loss_weight=3.0)
@dataclass
class PsychologicalState:
anxiety: Field = Field(1, 2, period=4)
sleep_quality: Field = Field(1, 2, period=24)
will_to_recover: Field = Field(1, 2, period=24, loss_weight=2.0)
@dataclass
class Patient:
cardiovascular: CardiovascularSystem
respiratory: RespiratorySystem
renal: RenalSystem
neurological: NeurologicalSystem
psychological: PsychologicalState
deterioration_risk: Field = Field(2, 4, loss_weight=8.0)
organ_failure_risk: Field = Field(1, 6, loss_weight=5.0)
@dataclass
class Nurse:
workload: Field = Field(1, 2)
fatigue: Field = Field(1, 2, loss_weight=2.0)
stress: Field = Field(1, 2, loss_weight=2.0)
competence: Field = Field(1, 2, is_output=False)
rapport: Field = Field(1, 2)
@dataclass
class BureaucraticState:
insurance_auth: Field = Field(1, 2, is_output=False, period=24)
bed_pressure: Field = Field(1, 2, period=12)
staffing_ratio: Field = Field(1, 2, period=8)
discharge_pressure: Field = Field(1, 2, loss_weight=2.0)
@dataclass
class FamilyUnit:
presence: Field = Field(1, 2, is_output=False, period=24)
emotional_state: Field = Field(1, 2)
communication_quality: Field = Field(1, 2, loss_weight=1.5)
@dataclass
class ICUWard:
global_acuity: Field = Field(2, 4, loss_weight=3.0)
resource_state: Field = Field(1, 4, is_output=False)
patients: list # 6 patients
nurses: list # 4 nurses
bureaucratic: BureaucraticState
families: list # 6 family units
Connectivity¶
bound = compile_schema(
ward, T=1, H=24, W=24, d_model=32,
connectivity=ConnectivityPolicy(
intra="dense",
parent_child="hub_spoke",
array_element="ring",
temporal="dense",
),
)
# 152 fields, 576 positions, 1,466 connections
The deterioration pathway¶
Sepsis trajectory: renal.creatinine_rise → cardiovascular.mean_arterial_pressure_drop → neurological.altered_consciousness → deterioration_risk. A flat model learns the correlation. This model is forced to route through the causal pathway — making it interpretable and robust to distribution shift.
This is the flagship example
Every field has a physiological reason for existing. Every connection has a known biological pathway. The type system is not decoration — it is the domain knowledge.