How to Check an AI-Made Simulation Before Class

Before you share it with students, check an AI-made simulation the way you would check a new worksheet from a colleague: work a few answers yourself and see whether they match. AI gets most simulations mostly right, and "mostly" is the problem. A wrong formula hidden behind a smooth animation teaches the wrong physics very convincingly. This guide gives you a 20-minute routine: a test sheet with hand calculations, six checks (values, units and labels, extreme sliders, language, phone and projector, and fixes), edit prompts that work, and what happens to your class links when you change a version.

It assumes you have already made a simulation. If not, start with how to create a simulation for your lesson with AI.

What the automatic checks already cover

When Simulic creates or edits a simulation, the progress steps show Writing, Checking, Testing in a browser and, if needed, Fixing problems. These steps catch code that doesn't run. If your preview still throws a JavaScript error later, the editor notices it and sends a free fix on its own, with the message "The simulation hit a JavaScript error. The AI is fixing it…". It does this a limited number of times per version.

What no automatic test can tell you:

• whether the formula is right;
• whether the units, labels and rounding suit your class;
• whether it behaves sensibly at the ends of every slider;
• whether the language is natural and consistent;
• whether students can read it on a phone or from the back of the room.

That part is yours, and it is quick.

Before you start: write a test sheet

Write three to five test cases with inputs and the output you expect. Take them from your textbook, or calculate them by hand. Do this before you look at the simulation, so its numbers don't anchor you.

Here is a test sheet for a typical Medium creation: a mass hanging on a spring, with sliders for mass (0.1–2.0 kg) and spring constant (10–100 N/m), a period readout, a stretch readout and g = 9.8 m/s².

Test	Inputs	Expected
Period	0.5 kg, 20 N/m	T = 2π√(m/k) = 0.993 s
Stretch at rest	0.5 kg, 20 N/m	x = mg/k = 24.5 cm
Scaling	2.0 kg, 20 N/m	4 × the mass gives 2 × the period: 1.99 s
Heavy and soft	2.0 kg, 10 N/m	stretch 1.96 m
Light and stiff	0.1 kg, 100 N/m	T = 0.199 s

A library simulation can be a free second opinion. The spring pendulum below shows T = 0.99 s for 0.5 kg and 20 N/m, the same value as your hand calculation.

One quick test per subject works the same way:

• Chemistry: an ideal gas simulation with 1.00 mol at 273 K in 22.4 L should show 101.3 kPa (R = 8.314 J/(mol·K)).
• Biology: a Hardy–Weinberg simulation with p = 0.6 should give genotype frequencies 0.36, 0.48 and 0.16.
• Math: a quadratic grapher should put the vertex of y = 2x² − 4x − 6 at (1, −8), with roots −1 and 3. The quadratic functions lesson plan has a full table of such cases.

Six checks before class

1. Values against hand calculations

Run every line of your test sheet. The readout should match your value to the decimals it shows. When it doesn't, the wrong number usually tells you what went wrong:

• 39.7 s instead of 0.993 s: the formula is upside down, √(k/m) instead of √(m/k).
• 0.158 s: the 2π is missing.
• 25.0 cm instead of 24.5 cm: the simulation uses g = 10 m/s². That's fine if your syllabus does, but it must match your questions.

Then change one input and check the factor, not just the value. A simulation that is right only at the defaults may have the answer written in as a fixed number. Finally, compare the animation with the readout. At 0.5 kg and 20 N/m, ten oscillations should take about 10 s (10 × 0.993 = 9.93 s). Count them against a stopwatch.

Leave the simulation running for two minutes. A frictionless spring should keep the same amplitude. If the swings slowly grow, the motion is calculated unstably, and students will "discover" free energy.

2. Units and labels

• Every slider and every readout has a unit, and the same quantity uses the same unit everywhere (not cm on the slider and m in the readout).
• Graphs have labeled axes with units.
• Rounding suits the level: three significant figures for 0.993 s, not 0.99271 s.
• Symbols match your textbook: k for the spring constant, not K or "stiffness".

Then open the starting values panel under the simulation. It lists every parameter with its label and unit, and students on an ordinary link can see and change all of them there. A label like "param_k" or a missing unit shows up here even when the simulation itself looks fine.

3. Extreme slider values

Set every slider to its minimum and maximum, alone and in combination. Look for objects leaving the screen, negative lengths, readouts showing NaN or Infinity, and graphs whose scale breaks.

On the spring simulation, two corners matter:

• 2.0 kg on 10 N/m stretches the spring 1.96 m. Does the mass stay on screen, or does it drop out of the bottom?
• 0.1 kg on 100 N/m oscillates with a period of 0.199 s. A 60 Hz screen shows only about 12 frames per oscillation, so the motion may look jerky or even seem to run backward. The readout must still say 0.199 s.

If a slider can reach zero, try zero: a division by zero is the most common crash. Then press Reset after each extreme and check that everything returns to the start.

4. Language

The simulation is written in the Language of the simulation you chose when you created it. Read every label, button, hint and explanation. Look for:

• words left in English inside a Spanish or French simulation;
• translations that are correct but not what your syllabus says;
• decimal points where your language uses decimal commas;
• explanations that are longer or more advanced than your students can read.

If the language isn't one you teach in every day, ask a colleague to spend two minutes on it.

5. Phone and projector

Students won't see your editor, so test what they will see. Create a class link and open it on a phone. Can you reach every slider with a thumb, read every number without zooming, and see the whole animation without scrolling sideways? Turn the phone sideways too.

For the whole class, open the link in present mode and stand at the back of the room. Check that readouts are large enough, that thin lines are still visible, and that nothing relies on red against green alone. The projector guide covers present mode in detail.

6. Fix problems with AI edits

Open your simulation's Simulation tab. The AI assistant sits next to the Preview. Type the change, then click Send (or press Ctrl+Enter). The line under the box shows "3 free edits left" or "This edit costs 2 credits", and the button shows the price.

• Each request creates a new version. The preview reloads it, and the chat shows "Version 4 created."
• A free edit is used only when the edit succeeds. Failed jobs refund their credits automatically.
• One job runs at a time.
• Major rewrite costs 10 credits (30 for a Complex simulation) and rebuilds the whole simulation. Use it when the structure is wrong, not for a wrong number.
• If an edit fails, Fix with AI puts the error message into the request box, ready to send.

A good edit prompt has three parts: the symptom (the inputs and what you see), the expected result (the value and why), and the scope ("change nothing else"). Examples:

At m = 0.5 kg and k = 20 N/m the period shows 39.7 s. It should be 0.993 s, from T = 2π√(m/k). Fix the formula and change nothing else.

At 2.0 kg and 10 N/m the mass falls off the bottom of the screen. Scale the drawing so the full 1.96 m stretch stays visible.

Label the slider "Spring constant k (N/m)" and show the period to three significant figures.

Two buttons are still in English: "Reset" and "Pause". Translate them into Spanish, like the rest of the simulation.

On a phone, the graph pushes the sliders below the screen. Put the sliders above the graph and make the graph shorter on narrow screens.

Put small text fixes (labels, units, translations) together in one request. Give each physics fix its own request, so you can tell which one broke something. After every edit, run your test sheet again: a fix in one place can break another.

If an edit makes things worse, open Details. The Versions table lists each version with your request. Click Use this version next to the one that worked.

Your class links always show the latest version

Every share link plays the current version of your simulation. When an edit finishes, students who open or reload the link get the new version. Use this version also switches every link at once.

That has three practical consequences:

• Don't edit during a lesson. Finish your edits, run the test sheet, then share.
• Update your questions. If an edit changes a value that a number question checks, change the answer and tolerance on the Questions tab too.
• Recheck starting values after a big edit. If an edit renamed or removed a slider, open each class link and make sure it still starts where you planned.

The 20-minute routine

1. Write the test sheet (5 minutes).
2. Run it and compare factors, not just values (5 minutes).
3. Push every slider to its ends, test zero, press Reset (3 minutes).
4. Read every label, unit and sentence, including the starting values panel (3 minutes).
5. Open a class link on a phone and in present mode (2 minutes).
6. Send one edit per physics problem, then retest (2 minutes plus the edit).

Then build your questions from the values you checked: they make reliable number questions. See how to turn exam questions into simulations for setting tolerances, and how to use interactive simulations in the classroom for running the lesson.

FAQ

Doesn't the AI already test the simulation?

It tests that the code runs in a browser and fixes errors that stop it. It does not check whether the formula, units or language are right. Only a hand calculation does that.

Do my edits change links I have already shared?

Yes. Every link plays the current version, so students see the new version the next time they open or reload the link. Edit before class, not during it.

What if an edit makes things worse?

Open Details and click Use this version next to the last good version in the Versions table. Nothing is deleted, so you can switch back again later.

How much does fixing a simulation cost?

Simple and Medium simulations include 5 free edits, Complex ones 10. After that, each edit costs 2 credits and a major rewrite 10 (30 for Complex). A free edit is only used when the edit succeeds.