<aside> 🧪

Run SEO changes as tests you can judge afterwards - not as hopes.

Each experiment names one change, one metric, a baseline, a success number, a read-out date and a stop condition before it starts. The verdict is computed: pass, fail, or unproven.

</aside>

Start here

Look at the seven sample experiments first. They show two passes, a fail, one still running, one stopped early, and two that are not ready to run.

  1. Open Metric Library and read the "Does Not Prove" column for the metric you plan to use. Add your own metrics if yours is missing.
  2. Add one row in Experiments: the change, the primary metric, the baseline with a link to where you saved it, the success threshold, the read-out date and the stop condition.
  3. Read Readiness Gate on that row. Your first useful output is one experiment showing ✅ Ready to run.

QualityGate

An experiment may start only when Readiness Gate is green and a person has ticked Human Approved. The gate needs: a change, exactly one primary metric, the "Metric Measures The Change" box ticked, a baseline with evidence, a direction, a success threshold, a read-out date and a stop condition.

A result counts only with a Result Value and a Result Evidence link. Without both, the verdict stays unproven - however good the change looks.

Working databases

Experiments

Metric Library

Decision rules

Verdict

Verdict Meaning What to do
🟢 Pass The result met the threshold in the stated direction Keep the change. Consider the next experiment on a bigger metric.
🔴 Fail The result did not meet the threshold Revert or revise. Write what you learned in the Hypothesis of the next test.
⚪ Unproven - no read-out yet Ready and maybe running, but no result with evidence Wait for the read-out date. Do not report it as a win.
⚫ Stopped early The stop condition was hit Revert, and note what was seen
⛔ Not ready Something needed to judge the test is missing Fill the gap before changing anything on the site

Does the metric measure the change?

Tick Metric Measures The Change only when the metric reads the exact thing you are changing. A title change is judged by a title metric. A heading change is judged by a heading metric. If you want to claim clicks, the primary metric must be clicks - and the baseline must come from the same source and the same length of window.

One change, one metric

One experiment changes one thing and names one primary metric. Two changes at once cannot be told apart afterwards.

Property guide