chini-015-er-triage

Emergency Department Triage

Five severity levels, finite beds, one CT scanner. The wrong queue means someone dies.

Source: Emergency medicine literature, ESI (Emergency Severity Index) protocol, hospital ops research

Prompt

Design the patient flow for a mid-size hospital emergency department.

Functional:
- Patient arrives at the door, screened by triage nurse, assigned ESI level 1-5 (1 = critical, 5 = minor).
- ESI 1-2 routes directly to a resus bay (4 beds). ESI 3-5 routes to a regular bay (12 beds) or fast-track (6 beds for ESI 4-5).
- Diagnostic resources: CT (1), X-ray (2), labs (shared). Shared across all bays.
- Disposition: admit (to inpatient floor, may board in ED if no upstream beds), discharge, or transfer.

Non-functional:
- A mass-casualty event (4x arrival rate) must NOT cause ESI 1-2 patients to wait. Lower-severity flow must be paced or diverted.
- If CT scanner is down, patients needing imaging must be queued for transport to imaging center, not blocked from triage.
- If inpatient floor is full, ED boarding cannot starve incoming critical patients of beds.

Return a Chinilla CanvasState. Components: door, triage, bays, resources, disposition. Behaviors: split (severity routing), queue (waiting room, boarding), ratelimit (low-severity pacing), circuitbreaker (CT failover), batch (lab orders).

Constraints

Max components: 14
Required behaviors: split, queue, ratelimit
Monthly budget: $1200000

Stress scenarios

Steady arrivals

baseline

Normal mix of ESI levels, all resources up.

Bus accident

spike

Arrivals 4x baseline, severity skewed high. ESI 1-2 must NOT wait.

CT scanner offline

outage

CT down for maintenance. Imaging needs reroute, triage must keep flowing.

Inpatient floor full

latency

Admits can't move upstairs, board in ED. Door must keep accepting critical patients.

Pass criteria (overall)

Min stability score: 60
Max drop rate: 10.0%
Min delivery rate: 85.0%
Max errors: 8

Submit your run

Submissions go through the chini-bench CLI. It calls your model with your key, scores the result locally, and posts to the leaderboard. Nothing leaves your machine except the canvas it produces.

End-to-end:

pip install git+https://github.com/collapseindex/chini-bench-cli.git
export OPENROUTER_API_KEY=...

chini-bench run chini-015-er-triage \
  --provider openrouter --model google/gemini-2.0-flash-001 \
  --as alice --x alice --linkedin alice-builds

Or inspect the prompt first:

chini-bench prompt chini-015-er-triage

Providers: openai · anthropic · google · openrouter · ollama

Leaderboard

Rank	Submitter	Model	Score	Stability	Delivery	Design	Pass	Links
#1	alex default	A anthropic/claude-sonnet-4.6	84	73.0	100.0	100.0	✗	X
#2	alex default	O openai/gpt-5.4	80	59.0	100.0	100.0	✗	X
#3	alex default	X x-ai/grok-4.20	79	66.0	100.0	100.0	✗	X
#4	alex default	G google/gemini-3.1-pro-preview	61	22.0	100.0	100.0	✗	X

Per-scenario breakdown of the top run

Scenario	Health	Drop rate	Delivered	Pass
baseline	75.0	0.6%	233	✓
mass-casualty	74.0	0.7%	867	✓
ct-down	68.0	0.0%	162	✓
boarding	75.0	0.7%	214	✓

How is this scored? →