Legacy content experiments - Gauge Docs

This workflow applies to organizations that retain legacy experiment authoring. When Optimizations is enabled, experiment writes return 409 and the old experiments remain readable. Use Improve agent results for new work in those organizations.

A content experiment asks whether a coding session would have changed if a page it fetched had contained different information. Gauge substitutes a rewrite at an eligible fetch boundary and simulates the session. It does not publish the rewrite to your site.

Test a rewrite

Choose a completed session that fetched a page you want to improve.
Inspect the fetched content and the point of confusion. State one hypothesis, such as making an initialization requirement explicit.
Start an experiment from an eligible page-fetch boundary and describe the rewrite.
Review the generated alternative and launch its simulated sessions.
Compare the live-page baseline and rewrite using criterion verdicts and session evidence, not only the final response.

CLI workflow

gauge experiments boundaries --run <run-id>
gauge experiments create \
  --run <run-id> \
  --boundary <boundary-seq> \
  --mode direct \
  --change "Explain the required initialization step before the first API call."
gauge experiments watch <experiment-id>
gauge experiments launch <experiment-id>
gauge experiments watch <experiment-id>

The first watch follows rewrite generation. After launch, the second follows the simulated sessions and the resulting recommendation. Keep the original task and criteria stable. Review the rewrite for factual correctness, repeat important comparisons, then publish through your normal workflow and validate with fresh runs against the live page.