STEP 01
WHAT WE MODEL
Define the cohort
Population shape, segments, and the assumptions that make the simulation worth running. Versioned with the experiment.
→AI LABS · SYNTHETIC POPULATIONS · A SIMULATION LAYER
Test it on a world that looks like yours.
A simulation layer. Not a substitute for real-world validation.
SIMULATED COHORT
COVERAGE · 3 LAYERS
COHORT · INTERVENTION · OUTCOME
COHORT
Defined upfront
The world you're testing against, not the one in the lab.
SIMULATION
Run the change
Interventions land on segments before they land on people.
VALIDATION
Against reality
Synthetic readings checked against real-world bands.
HOW IT WORKS
Three steps. One reading.
Define the cohort. Simulate the run. Validate against reality.
STEP 02
WHAT WE RUN
Simulate the run
The intervention lands on the cohort. Outcomes spread across segments. Risk and lift show up before the policy ships.
→STEP 03
WHAT WE CHECK
Validate against reality
Simulated bands compared to real-world aggregates. Drift surfaced. Signed datasets ready for the next review.
WHAT COMES OUT
What your team leaves with.
Every simulation leaves a record the next decision has to clear — and a check the real world will be measured against.
01
Cohort specs
The population, the segments, the assumptions. Versioned, reviewable, and the same on every run.
↳ ARTIFACT
02
Simulated runs
Outcome distributions across segments. Lift and risk attached to every experiment.
↳ ARTIFACT
03
Validation reports
Synthetic readings checked against real-world aggregates. Drift bands flagged before rollout.
↳ ARTIFACT
04
Signed datasets
Cohort, intervention, results, and reviewer chain — sealed when the experiment is approved.
↳ ARTIFACT
WHERE IT FITS
In the loop, this is where you test.
Test the run on a world that looks like yours. Review the hard cases. Recruit the right specialist. Remember the misses. Approve what's right.
01
Test
● YOU ARE HERE
02
Review
03
Recruit
04
Remember
05
Approve
RELATED MODULES
Next to this in the Model OS.
TRAINING
Train on the work, not the demo.
Curated runs become the data the next model learns from.
See the page →RL ENVIRONMENTS
Deterministic worlds for agents.
Reproducible RL environments for policies before they leave the lab.
See the page →FEDERATED LEARNING
Your data stays where it is.
Train across boundaries without moving the underlying records.
See the page →SYNTHETIC POPULATIONS
Test it on a world that looks like yours.
A simulation layer for the decisions that touch people. Not a substitute for real-world validation.