On a broken or silent output the agent must recover to a known-good state before adding new ideas

When a live output has errored, gone silent, or gone black (a name typo, a missing output call, feedback runaway), the correct first action is to recover — revert to the last-known-good state or a safe fallback — rather than stack new ideas on top of a broken state. Layering a new idea onto broken output compounds the failure and makes the cause harder to isolate, and it wastes the audience’s attention on a state that is not producing sound or image. Recovery takes priority over every arc-shaping move: it is the highest row of the autonomous decision table and fires immediately and once. This is the performance analogue of fixing the build before adding a feature.

Examples

The visuals go black after a save because the sketch lost its output call. The agent reverts that one edit to restore output before considering any new visual idea — it does not add a new layer on top of the black canvas.

Assessment

Why must a broken or silent output be recovered before new ideas are added? Contrast recovering to last-known-good with stacking a new idea onto a broken state.