Mini-Checklist: Run Your Own Before/After Test

We do not publish win-rate charts. The real benchmark is your own goal, run on your own host. Why this matters: part 01.

☐ Same goal, written once. Use one real task you would have done anyway.

☐ Same host model. Do not switch assistants between the two runs.

☐ Before: prompt the host the way you normally do, then save the output.

☐ After: call Edge with the same goal, feed the brief to the host, then save that output too.

☐ Score coherence. Which version holds together from start to end?

☐ Score specificity. Which one says concrete things instead of filler?

☐ Count your repairs. How many edits did each output need before you could use it?

☐ Write down the verdict. One line on which one won and why.

If the brief does not reduce repair work for that kind of goal, say so and move on. If it does, you have evidence instead of hype. Either way, you made the call with the lights on.

Soft door: run your first before/after pair with one Edge call from Members.