Bonus experiments
Finish early? Choose one bonus
Section titled “Finish early? Choose one bonus”Complete the finish-line verification first. These experiments are independent, optional, and designed to avoid pulling participants back into the main workshop path.
| Bonus | Time | Default verification | What you learn |
|---|---|---|---|
| On-demand workspace | 5–8 min | Live deploy | Expose Sandbox tools only after the model requests a workspace |
| AI Gateway + Guardrails | 8–10 min | Live deploy | Route Flue inference through a named gateway and observe a blocked prompt |
| Model-turn telemetry | 5 min | Live deploy | Export model-turn details and correlate them with AI Gateway |
| Content-free traces | 5–8 min | Live deploy | Keep trace shape and usage while removing prompts and tool payloads |
| Recover from tool failure | 3–5 min | Live deploy | Correct a failed forecast in the same conversation and compare outcomes |
| Compare two models | 5–8 min | Live deploy | Compare latency, tokens, and instruction following with the same agent |
Protect your completed workshop
Section titled “Protect your completed workshop”Create a branch before editing:
git switch -c bonus-experimentsEach page includes a reset section. Use a fresh conversation ID for every experiment so previous state does not affect the result. If a Cloudflare dashboard feature is unavailable in your account, choose another bonus rather than troubleshooting account entitlements.