Write down what you can see
Describe the action in one sentence. For example, the arm puts one block on top of another on a clear table. Then check the source for who controlled the robot and when a person might have intervened. If the source leaves that out, treat autonomy as an open question. A cut can also hide the time between actions, though it does not establish that an attempt failed.
Look for the full set of attempts
If the team reports a success rate, check how many attempts it covers and what counted as success. For block stacking, changing the starting positions can help test whether the robot repeats a trajectory or adjusts its movement. Videos recorded under different conditions tell you different things.
Check whether the test was in simulation
Ai2 introduced ToyBlocks on November 6, 2026. It lets researchers vary block tasks and evaluate models on a simulated tabletop. Ai2 describes changing the training demonstrations and evaluating on tasks set aside for testing. That setup helps researchers study what a model learned in that environment. Claims about performance on a physical robot would need results that describe the hardware and test conditions. Ai2 announcement · ToyBlocks documentation
When sharing the demonstration, say what the robot did and under what main condition. If you saw one attempt, describe what happened in that attempt.