- Log test idea in brain with:
- Hypothesis: "We believe [change] will produce [outcome] because [reason]"
- Variable being tested (single variable only — no multivariate)
- Success metric: specific, measurable (e.g., CTR, CPQBC, lead volume)
- Minimum run duration (see Gap Flags for threshold TBD)
- Control vs. variant definition
hypothesis, success metric, and minimum run duration are logged and approved
Branch
Test disrupted mid-run (accidental edit, platform issue):
Invalidate test. Log disruption reason in brain. Restart from Backlog with same hypothesis if disruption was external.
Branch
Test reveals a new hypothesis:
Log new hypothesis as a separate Backlog entry — do not extend the current test.
2
In Progress (SLA: 168h to launch after approval)
- EA sets up test per protocol:
- Duplicate control asset or workflow
- Apply single variable change to variant
- Confirm tracking is in place before launch
- Log start date and baseline metrics in brain project record
- Do not touch test mid-run — no optimizations, no budget changes, no creative edits
test has run for full minimum duration with no interruption
3
Completed (SLA: 72h post-run)
- Pull results: winner, loser, delta on success metric
- Log in brain project record:
- Winner (control or variant)
- Delta (absolute + percentage)
- Sample size and run duration
- Confidence level if calculable
result logged in brain with all fields above
- Write findings doc — include:
- Hypothesis tested
- Result summary
- Recommended action (scale winner, update SOP, run follow-up test)
- Update relevant SOP if result changes best practice
- Link findings doc in `brain.resource_docs` with tag `split-test`
findings doc exists, SOP updated if applicable — WIN terminal
↓
Hands off to - **Winning variant (creative)** → Meta Ads Creative Pipeline at Not Started. Include: asset link, test summary, delta achieved. - **SOP-impacting findings** → relevant pipeline SOP file updated in same session as Documented stage. Akash approval required before update. - **Findings docs** → linked in `brain.resource_docs` with tag `split-test` for quarterly review.