We love our LASER AI, JENNER, but as software, we need a way of testing it. And that means not manual testing but automated. In short, assuming we can define a set of prompts that we expect to "work", we want to have a way of automatically sending those prompts to the AI, getting code back, and making sure that code "is good".

For our set of test prompts, we want to:

• Send them to an AI webservice (enterprise hosted MCP server), and get python code back.
• Run the code in an environment with laser(*) fully installed (docker image).
• Capture the output from running the code.
• Compare the output of the code to an expected result.
• Report pass/fail.
• Note that we do not here propose testing the code by inspection, just by running and inspecting the output.

We want these prompts/tests to go from something very simple to become progressively more demanding.

Here’s a curated set of 15 progressively complex prompt ideas you could use to test and reuse laser-core features. Each prompt is designed to return runnable Python code that imports and invokes at least one function or class from laser-core. This suite gradually introduces core concepts, starting with minimal LaserFrame usage and building toward demographic modeling, components, squashing, and migration.

1. Prompt 1: "Create a minimal example that initializes a LaserFrame with a capacity of 10 and adds a scalar property named age, then populates it with 5 agents and prints the age array."

Sample Result 1

python3 jenner_bvt1.py
Age array:
[25 30 18 45 60]

2. Prompt 2: "Add a vector property called position with length 2 to a LaserFrame of capacity 20. Populate 3 agents and set each agent's position manually, then print the property."

Sample Result 2

python3 jenner_bvt2.py
Position array (2D):
[[1. 3. 5.]
 [2. 4. 6.]]

Position array (agent-wise rows):
[[1. 2.]
 [3. 4.]
 [5. 6.]]

3. Prompt 3: "Write a function that adds 100 agents to a LaserFrame with a scalar state property, sets the state of all agents to 1, and verifies that all entries are equal to 1."

Sample Result 3

python3 jenner_bvt3.py
After step 1: age = [1 1 1 1 1]
After step 2: age = [2 2 2 2 2]
After step 3: age = [3 3 3 3 3]
After step 4: age = [4 4 4 4 4]
After step 5: age = [5 5 5 5 5]

4. Prompt 4: Sample Initial Ages from Population Pyramid

Use AliasedDistribution and load_pyramid_csv() to assign realistic ages (in days) to a new population of 1,000 agents using LASER’s bundled US 2023 pyramid.

Key Features:

• Tests load_pyramid_csv()
• Exercises AliasedDistribution.sample()
• Converts age bins to age-in-days
• Stores in scalar age property

Sample Result 4

python3 jenner_bvt4_realisticages.py
First 10 sampled agent ages (in days):
[ 3457  4743  1204 18044 27217   889 26394 26076  3436  1024]

Converted to years:
[ 9 12  3 49 74  2 72 71  9  2]

5. Prompt 5: Set date_of_birth Based on Sampled Age

Convert sampled agent ages into date_of_birth (i.e. -age_in_days) and add date_of_birth as a scalar property in a LaserFrame. Show the first 10 values.

Key Features:

• Reinforces LASER's design: “born in the past”
• Introduces date_of_birth as a required field for demographic realism
• Ensures compatibility with later use in mortality models
• Keeps age dynamic (instead of static scalar)

Sample Result 5

python3 jenner_bvt5_dob.py
First 10 agents:
  Agent 0: age=1436 days, date_of_birth=-1436
  Agent 1: age=25241 days, date_of_birth=-25241
  Agent 2: age=5825 days, date_of_birth=-5825
  Agent 3: age=9828 days, date_of_birth=-9828
  Agent 4: age=12434 days, date_of_birth=-12434
  Agent 5: age=21007 days, date_of_birth=-21007
  Agent 6: age=15258 days, date_of_birth=-15258
  Agent 7: age=12134 days, date_of_birth=-12134
  Agent 8: age=24851 days, date_of_birth=-24851
  Agent 9: age=25432 days, date_of_birth=-25432

6. Prompt 6: Predict date_of_death with KaplanMeierEstimator

Using KaplanMeierEstimator, predict date_of_death for all agents based on their age in days, and store it as a property in the LaserFrame.

Key Features:

• Uses create_cumulative_deaths() helper or synthetic data
• Introduces date_of_death
• Sets up agents for mortality-aware step functions in Level 3

Sample Result 6

python3 jenner_bvt5_dob.py
First 10 agents:
  Agent 0: age=1436 days, date_of_birth=-1436
  Agent 1: age=25241 days, date_of_birth=-25241
  Agent 2: age=5825 days, date_of_birth=-5825
  Agent 3: age=9828 days, date_of_birth=-9828
  Agent 4: age=12434 days, date_of_birth=-12434
  Agent 5: age=21007 days, date_of_birth=-21007
  Agent 6: age=15258 days, date_of_birth=-15258
  Agent 7: age=12134 days, date_of_birth=-12134
  Agent 8: age=24851 days, date_of_birth=-24851
  Agent 9: age=25432 days, date_of_birth=-25432

7. Prompt 7: "Define a Model class with a LaserFrame, add infection_state and a component that randomly infects 1% of agents at each step. Run 10 steps and count infections per step."

Sample Result 7

python3 jenner_bvt7.py
Step 1: infected = 10
Step 2: infected = 20
Step 3: infected = 30
Step 4: infected = 40
Step 5: infected = 50
Step 6: infected = 60
Step 7: infected = 70
Step 8: infected = 80
Step 9: infected = 90
Step 10: infected = 100

8. Prompt 8: "Create a LASER model that includes a component for aging (increments age each day), and a component that kills agents when their age exceeds a lifespan threshold."

Sample Result 8

Step 1: alive = 100
Step 2: alive = 100
<snip!>
Step 16: alive = 100
Step 17: alive = 100
Step 18: alive = 100
Step 19: alive = 100
Step 20: alive = 100
Step 21: alive = 99
Step 22: alive = 98
Step 23: alive = 96
Step 24: alive = 95
<snip!>
Step 74: alive = 16
Step 75: alive = 15
Step 76: alive = 10
Step 77: alive = 8
Step 78: alive = 4
Step 79: alive = 2
Step 80: alive = 0

9. Prompt 9: "Create a LASER model that which has 2 components that interact with each other. Maybe a timer which counts down and changes state when the counter reaches 0."

Sample Result 9:

Step 1: infected = 20, recovered = 0
Step 2: infected = 20, recovered = 0
Step 3: infected = 20, recovered = 0
Step 4: infected = 16, recovered = 4
Step 5: infected = 14, recovered = 6
Step 6: infected = 9, recovered = 11
Step 7: infected = 6, recovered = 14
Step 8: infected = 0, recovered = 20

10. Prompt 10:

"Create a LASER model that includes a LaserFrame for storing time-series results (per timestep). Add a reporting component that counts the number of agents in each infection state — Susceptible (0), Infected (1), and Recovered (2) — and records their fractions in a results frame. Run the model for 20 timesteps and print the values stored in the result frame at the end."

Sample Result 10:

python3 jenner_bvt10.py
Timestep |   S     I     R
       1 | 0.990 0.010 0.000
       2 | 0.980 0.020 0.000
       3 | 0.970 0.030 0.000
       4 | 0.960 0.040 0.000
       5 | 0.950 0.050 0.000
       6 | 0.940 0.060 0.000
       7 | 0.930 0.070 0.000
       8 | 0.920 0.080 0.000
       9 | 0.910 0.090 0.000
      10 | 0.900 0.100 0.000
      11 | 0.890 0.110 0.000
      12 | 0.880 0.120 0.000
      13 | 0.870 0.130 0.000
      14 | 0.860 0.140 0.000
      15 | 0.850 0.150 0.000
      16 | 0.840 0.160 0.000
      17 | 0.830 0.170 0.000
      18 | 0.820 0.180 0.000
      19 | 0.810 0.190 0.000
      20 | 0.800 0.200 0.000

11. Prompt 11: "Write a component that sets agents with status = 2 as recovereds and squashes them from the LaserFrame. Print the reduced agent count."

Sample Results 11:

python3 jenner_bvt_squash.py
Before squashing: count = 100
After squashing: count = 70

12. Prompt 12: "Demonstrate saving a population snapshot to HDF5 using save_snapshot() and reloading it with load_snapshot(). Show that data matches before and after."

TBD

13. Prompt 13: "Create a migration matrix using laser_core.migration.gravity() with 3 nodes and use it to simulate movement of agents between nodes. Print before/after node distributions."

Sample Result 13: (Might not be correct)

python3 jenner_bvt_migration.py
Before migration:
  Node 0: 100 agents
  Node 1: 100 agents
  Node 2: 100 agents

After migration:
  Node 0: 153 agents
  Node 1: 147 agents
  Node 2: 0 agents

14. Prompt 14: Spatial

    "Build a spatial SIR disease model on a 1-D ring of patches, e.g., 10 patches where the ends are connected (patch 0 neighbors patch 1 and patch 9). Population: Distribute populations heterogeneously across patches (e.g., patches 0–2 and 7–9 are high-density, others are sparse). Seeding: Infect a small fraction of agents in one high-density patch (e.g., patch 0). Migration: Implement agent migration between adjacent patches, with ring wraparound. Use a fixed per-tick migration probability. Transmission: Model SIR infection dynamics within each patch. Transmission depends on local susceptible and infectious fractions. Objective: Show that the infection spreads to distant patches via agent migration, not external seeding. The contagion should eventually appear in patches not directly connected to the seed."

Sample Result 14:

Step 1
  Patch 0: S=1226 I=11 R=0
  Patch 1: S=1248 I=2 R=0
  Patch 2: S=1243 I=0 R=0
  Patch 3: S=644 I=0 R=0
  Patch 4: S=625 I=0 R=0
  Patch 5: S=632 I=0 R=0
  Patch 6: S=624 I=0 R=0
  Patch 7: S=1220 I=0 R=0
  Patch 8: S=1261 I=0 R=0
  Patch 9: S=1263 I=1 R=0
----------------------------------------
<snip!>
Step 50
  Patch 0: S=508 I=213 R=486
  Patch 1: S=507 I=194 R=389
  Patch 2: S=534 I=243 R=252
  Patch 3: S=681 I=136 R=113
  Patch 4: S=649 I=65 R=27
  Patch 5: S=715 I=11 R=17
  Patch 6: S=786 I=59 R=36
  Patch 7: S=713 I=173 R=125
  Patch 8: S=630 I=242 R=329
  Patch 9: S=528 I=198 R=441

(This prompt produced 150 lines of perfectly working LASER spatial SIR code, with above output.)

15. Prompt 15: All The Things

    "Build a minimal SIR model using LaserFrame, PropertySet, and 3 components: SeedInfection, Transmission, and Recovery. Run it over 30 timesteps and print S, I, R counts per day."

Sample Result 15:

TBD

We see above that we can enumerate prompts which return code which can be run and produces outputs which can be measured against a reference output. The next step is to host the JENNER-MCP service at some known endpoint and run these tests automatically. It's possible that some outputs may need to be compared to the reference within some statistical bound.