Google Simula's secret is mechanism design — planning the entire data set as a product before generating anything. Here's how it works.
Most AI data generation tools work one prompt at a time.
Like a factory worker making one shoe at a time without a plan.
Sometimes great shoes.
Sometimes a pile of left feet.
Simula is different.
It plans the whole data set top-down before making anything.
Google calls this mechanism design.
This post explains how it works.
Why Mechanism Design Matters
Three problems mechanism design solves:
1. Coverage gaps.
Random generation misses entire parts of a domain.
2. Quality variance.
Some prompts produce great data, others produce garbage.
3. Diversity collapse.
AI tends to repeat itself when generating in volume.
Mechanism design addresses all three.
The 3-Stage Mechanism Design
Simula breaks data generation into three stages.
Stage 1 — Global diversification
Map the entire domain first.
Stage 2 — Local diversification
Zoom into each spot on the map and generate variety.
Stage 3 — Dual critic filter
Quality control before saving.
Stage 1 — Global Diversification
This is the planning stage.
Simula uses a taxonomy.
Think of it as a giant menu of every possible topic and subtopic in the area you care about.
For cyber security data:
- Every type of attack.
- Every type of defender.
- Every type of system.
- Every corner of the space.
For legal data:
- Every type of case.
- Every type of legal question.
- Every jurisdiction relevant.
- Every category of practice area.
This taxonomy ensures coverage.
Without it, you'd miss whole topics.
Stage 2 — Local Diversification
Once the map is drawn, Simula zooms into each cell.
Two techniques.
One-of-N meta prompting
For each spot on the taxonomy, Simula generates many different versions.
"Many" not "one".
This prevents the data set from sounding repetitive.
Complexification
Then Simula takes simple examples and pushes them harder:
- Easy: basic version of the scenario.
- Medium: more nuanced version.
- Hard: edge case version.
- Boss fight: extreme edge case.
Like leveling up a video game.
The model trained on this learns the full range.
🔥 Want to apply mechanism design to your AI workflows? Inside the AI Profit Boardroom, I share how to apply Simula-style mechanism design to your own AI use, plus daily training and weekly live coaching. 2,800+ members. → Get the playbook
Stage 3 — Dual Critic Filter
Quality check before anything is saved.
Two different critic models look at each example.
They decide:
- Is this good enough to keep?
- Or should it be thrown out?
Numbers from Google's tests:
- Legal data set: 61% of generated data rejected.
That's a serious filter.
Most of what was generated wasn't good enough.
The output quality is high BECAUSE the filter is strict.
Why Two Critics, Not One
Single critic = single point of failure.
Two critics = checks and balances.
If one critic accepts something the other rejects, it's flagged.
If both reject, throw out.
If both accept, keep.
This matches how real research works — peer review involves multiple reviewers for a reason.
What Mechanism Design Means For You
Three lessons applicable beyond Simula.
1 — Plan before generating
Whatever you're building with AI, sketch the full scope first.
Don't just start prompting.
2 — Cover the full domain
Don't let AI default to the easy/common parts.
Push it to cover edge cases.
3 — Use a critic step
Always have a second AI (or human) review before publishing.
I apply this in Hermes Agent Swarm workflows.
Quality Vs Diversity Vs Complexity
This is one of Simula's biggest insights.
Most AI generation conflates these:
- Quality.
- Diversity.
- Complexity.
Simula treats them as three separate knobs.
You control them independently.
When to want high quality + low complexity
For training a chatbot — you want safe, simple examples.
When to want high complexity + narrow scope
For training specialist AI (legal, medical) — you want narrow but deep examples.
When to want high diversity + medium complexity
For training general-purpose models — broad coverage with depth.
Different use cases need different settings.
Simula gives you that control.
How Mechanism Design Compares To Other Data Generation
Simula:
- Top-down planning.
- Multi-stage refinement.
- Dual critic filter.
Most AI data generation tools:
- One-prompt-at-a-time.
- No taxonomy.
- Single critic at best.
Manual data generation:
- Slow.
- Expensive.
- Inconsistent.
Simula's approach is strictly better than most alternatives.
Real Numbers From Google's Tests
The math benchmark (GSMAT):
- Low complexity vs high complexity comparison.
- 64,000 data points each.
- High complexity gave 10% accuracy gain.
That's massive in AI terms.
But:
- Only worked with a strong teacher model.
- Weak teacher (57% accurate) saw performance drop with high-complexity data.
Lesson: complexity helps when the teacher can label correctly.
Why "Real Reference Data" Sometimes Loses
Real-world data covers what people happen to write online.
Simula covers what's needed on purpose.
Result: Simula data sets sometimes have better coverage than real data sets.
This is counter-intuitive but real.
Applying Mechanism Design Beyond Data Generation
This pattern applies to anything you're producing in volume with AI:
- Content (e.g. SEO posts).
- Customer responses.
- Code modules.
- Marketing assets.
Three steps:
1 — Define the taxonomy
What categories does your output need to cover?
2 — Generate diverse examples per category
Don't let AI default to one style.
Push for variety.
3 — Apply a critic step
Always review before publishing.
I apply this principle in Claude Code SEO Agent workflows.
Complexification As A Pattern
Simula's "complexification" is also broadly useful.
Take simple AI outputs and push them harder:
- "Make this more nuanced."
- "Add edge cases."
- "Challenge the obvious answer."
Output quality improves.
This is something you can apply today even without Simula.
What This Reveals About AI's Future
Three predictions.
1 — More products will use synthetic training data
Privacy + cost + access concerns make synthetic appealing.
2 — Mechanism design pattern spreads beyond data
Production-quality AI workflows will adopt similar planning + filtering.
3 — Quality bar rises industry-wide
When everyone uses better techniques, the floor rises.
What Solo Operators Can Take From Simula
Three lessons.
1 — Plan before generating
Don't just throw prompts at AI.
Map your domain first.
2 — Cover the full domain
Push AI to handle edge cases.
3 — Always use critics
Second-pair-of-eyes (AI or human) on everything.
Output quality jumps.
🚀 Want my full AI workflow design playbook? The AI Profit Boardroom has my AI workflow templates, OpenClaw 6-hour course, Hermes 2-hour course, daily training, weekly live coaching. 2,800+ members. → Join here
FAQ — Google Simula Mechanism Design
What is mechanism design?
Planning the full data set as a product before generating anything.
Top-down approach vs prompt-by-prompt.
Why two critics, not one?
Single critic is a single point of failure.
Two creates checks and balances.
Can I use Simula myself?
Not directly — it's a research framework.
But you can apply the pattern to your own AI workflows.
Will Simula become open source?
Possibly — Google often releases research papers.
Is mechanism design slow?
Initial planning takes time.
Then execution scales much faster than ad-hoc prompting.
Can Simula generate ANY type of data?
Best for structured domains.
Less suited for highly creative or stylistic data.
What's the biggest insight from Simula?
Quality, diversity, and complexity should be separate knobs — not lumped together.
Related Reading
- Google Simula Overview — what Simula does.
- Hermes Agent Swarm — multi-agent with critic pattern.
- Claude Code SEO Agent — applied AI workflow.
📺 Video notes + links to the tools 👉 https://www.skool.com/ai-profit-lab-7462/about
🎥 Learn how I make these videos 👉 https://aiprofitboardroom.com/
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉 https://www.skool.com/ai-seo-with-julian-goldie-1553/about
Google Simula's mechanism design is the smarter way to produce AI training data — and the same pattern can improve any AI workflow you run.