# Build a weekly social media experiment log

Write the decision, variation and comparison rule before publishing. Record the exact assets and consistent observation windows. Use the log to understand what changed; a small uncontrolled comparison is a clue for the next test, not proof of a universal best practice.

Experiment name and owner:
Question that changes a production decision:
Hypothesis:
Variation being changed:
Conditions intended to remain similar:
Asset references and publication times:
Metric definition and source:
Observation window:
Comparison rule chosen in advance:
Actual results and denominators:
Unexpected changes or recording gaps:
Alternative explanations:
Conclusion supported by the evidence:
Next decision or follow-up test:


## Complete the record

"Log entry","Fictional example","Interpretation"
"Question","Does a close-up opening help viewers understand the product?","One production decision"
"Variation","Close-up first versus packaging first","Keep the later explanation comparable"
"Measure","Completion under one consistent platform definition","Use the same observation window"
"Conditions","Different topics and audiences may still affect results","Record differences instead of concealing them"
"Next step","Repeat with another suitable product explanation","Check whether the finding holds"


## Check before sharing

- [ ] Change one main feature when the question allows it.

- [ ] Record both the numerator and denominator for ratios.

- [ ] Avoid choosing the winning metric after seeing the results.

- [ ] State when the evidence is too sparse or inconsistent for a decision.

## Use the worksheet in practice

A post with eight saves from 200 reached accounts has a 4% saves-per-reach ratio; another with twelve saves from 600 has 2%. The second has more saves in total. Which matters depends on the decision and consistent metric definitions, not whichever number looks better.

An experiment log preserves the context for a follow-up. It should include changed offers, paid distribution, unusual events and delayed publication when those affect the comparison. Do not remove an inconvenient result from the record.