Mock article — Editorial demonstration, not a published research result.
Start with one question
An agent succeeds at a task once. What should we learn from that? Begin with a narrower question: which part of the setup made success possible? A useful experiment changes one condition and keeps the others visible.
Make the baseline explicit
Record the task, tools, instructions, and stopping rule before comparing approaches. Keep a simple baseline alongside the new idea. A comparison is easier to interpret when both runs share the same constraints.
Keep the failures
A failed run can expose an assumption that a successful run hides. Save the inputs and the trace, describe what happened, and separate the observation from its explanation. The next experiment starts there.