Running the first comparison
Three plans against your baseline. A few seconds.
What's going to work, what isn't and when you'll know
All the numbers, per arm
Funnel and loss reasons
What happened inside the buyers
Assumptions this readout printed
Three plans against your baseline
Arms are compared cold: each one runs on identical buyers with identical random draws, so whatever separates them came from the plays. Change anything and the readout above re-runs on its own.
Where this could be wrong
Which numbers the ranking leans on, which of those you never set, what the simulator cannot see, and what would make it trust its own ranking more.
More plays Others the simulator ranked, promote any
The simulator ran your current setup first, found the bottleneck, then simulated every play in the library against it. Promote any of these into an arm.
Ask a simulated buyer why
Each buyer carries a memory of every touch and the beliefs they formed from it. Pick one from any arm and ask. With Claude connected they answer in their own voice; without it, you read their memory directly.
Simulating 300 buyers over 16 weeks so there is someone to interview…
Name a target, get a plan
Name the outcome you want and the simulator searches for a sequence that reaches it. A greedy search over the play library, each step paired against the current plan on the same buyers.
Log what really happened
A simulator without a closed loop to reality is the least valuable version of itself. Log what actually happened when you ran a play; its prior updates and the next run reranks. Paste your real weekly wins to see how far the model is from your history.
Log a real outcome
Backtest against your history
What this instance remembers
Stored in this browser only. Priors, outcomes and custom plays survive reloads.