01
Hypothesis
A falsifiable statement linked to the source recommendation.
Experiments
Decide what the campaign should teach you.
AdFlow turns a recommendation into a testable learning plan with a falsifiable hypothesis, fixed cells, one isolation variable, a primary metric, and explicit stopping conditions.
Experiment approval covers the learning plan. A separate Execution Plan covers any Meta account change needed to launch it.
01
What you expect
02
What changes
03
When to conclude
Define the test
Make the comparison understandable before launch.
A reviewer should be able to see the idea, the cells, the isolated change, and the decision rule in one record.
02
Cells
Fixed control and variant entities.
03
Isolation
One planned variable between cells.
04
Metric
Primary metric plus operational guardrails.
05
Gates
Minimum sample, runtime, and stop rule.
06
Risks
Known assumptions and contamination limits.
Prepare the launch
Check the live setup before planning a change.
After design approval, AdFlow reads the current Meta setup. If launch requires an account change, that change moves into a separate Execution Plan.
Submit the experiment design for review.
Approve the hypothesis, cells, metric, and gates.
Read the current Meta setup.
Create a plan for required setup changes.
Review and approve the exact current and proposed values.
Verify the launch before measurement starts.
Partial and failed execution results stop before the Running state.
Measure the test
A result is ready only when its gates are satisfied.
AdFlow records spend, results, the primary metric, sample progress, runtime, data-through time, and contamination for each cell.
Readiness gates
- Minimum sample
- Minimum runtime
- Active cell integrity
- Supported statistical inputs
Conclusion record
- Variant or control winner
- Inconclusive or insufficient data
- Invalid result after contamination
- Limitations and sample snapshot
Preserve the learning
Keep the conclusion with the test that produced it.
The record includes the outcome and its limitations. A follow-up can become another experiment; a rollout begins as a reviewable Execution Plan.
Conclusion. Limitation. Follow-up. Rollout plan.