AI Assist Results
AI Assist Results reports the measured effect of the AI treatment. The test design and implementation live in Methodology; here the question is narrower: did the timestamped, paid treatment improve the paired paper result by more than it cost?
Should AI Assist Be Enabled?
Decision Gates
Cost Sensitivity
This slider is a planning estimate, not a new experiment result. It holds the observed gross treatment advantage fixed and asks how the conclusion would change under a different 30-day review cost. Future market behavior and future AI proposals will not remain fixed in practice.
What Did The A/B Test Show?
Evidence source:
Measured paper A/B
Control and treatment value over the completed round
Read the vertical separation as treatment-minus-control evidence, not as live-account profit. Review cost is deducted from the recorded treatment path.
Measured cost-adjusted effect
AI treatment lift by experiment day
Values above zero favor the treatment after attributed review cost; values below zero favor the model-only control.
Model-Level Effects
The aggregate result is not applied to every model. Positive rows are candidates for another paper treatment; flat, negative, or missing rows remain model-only. This prevents one successful sleeve from subsidizing AI use where it did not help.
How The AI Assist Experiment Is Governed
In the experiment, the rule-based model remains the control and only the treatment receives AI-assisted changes. Costs and returns are recorded separately for both paths. Better explanatory text can improve the walkthrough, but the financial test passes only when the assisted paper result finishes ahead after its review cost.
Next step: use Strategy Setup to choose model-only or selective paper assistance for a strategy. Use Strategy Tournament to observe the next locked round. Neither page should reinterpret a local what-if estimate as measured A/B evidence.