Fallbacks and reset
Rule for live demos
Section titled “Rule for live demos”Do not debug in front of the room for more than 30 seconds. Switch to a saved artifact, restore a checkpoint, or move to the next segment.
Per-step fallbacks
Section titled “Per-step fallbacks”| Demo step | If it fails | Do this instead |
|---|---|---|
| Lab 05 before run | The agent stalls or answers badly. | Show the saved baseline run file and explain the score. |
| Lab 06 assessor | The assessor misses an obvious issue. | Open the saved assessment notes or the generated rules checklist. |
| Lab 07 handoff | The handoff does not appear. | Copy the saved handoff prompt into the improver agent manually. |
| Lab 07 edits | The improver changes too much. | Restore the checkpoint and show the prepared diff. |
| Lab 08 DAX tests | Power BI Desktop is closed or slow. | Show the saved Invoke-DaxQuestionTests.ps1 output. |
| Lab 10 comparison | The after run is not ready. | Use the sample run and Expected results. |
| Lab 11 hook denial | The hook wrapper cannot find .venv. | Explain the wrapper’s safe allow behavior and show the prepared denial output. |
| Lab 12 PR review | Copilot review is delayed. | Switch to the pre-recorded PR or saved review comment. |
| Lab 13 cloud agent | Cloud agent queue is slow. | Show the pre-opened issue, setup workflow, and completed cloud PR. |
| Lab 14 extension | Time is short. | Show the smallest file change, then point to the full lab. |
Agent stalls
Section titled “Agent stalls”If an agent stops making progress:
- Stop waiting after 30 seconds in the short demo or 60 seconds in the workshop.
- Say: “This is why the lab keeps sample runs and checkpoints. We do not need the live agent to be perfect to teach the workflow.”
- Open the saved run or restore the checkpoint.
- Continue from the next planned segment.
Bad answer
Section titled “Bad answer”If an agent gives a wrong answer, do not hide it. Use it as the teaching point:
- Baseline run: “This is the failure we expect before the model is AI-ready.”
- After run: “This is why we score against a deterministic oracle and keep DAX tests.”
- Review run: “This is why PR review remains human-owned.”
Restore a checkpoint
Section titled “Restore a checkpoint”Start from a clean working tree. The restore script creates a new lab/<checkpoint>-<timestamp> branch for you:
pwsh ./scripts/lab/Restore-LabCheckpoint.ps1 -Checkpoint lab02-startUse a more specific checkpoint when you are restarting later in the demo. After restoring a checkpoint, run Test-LabProgress.ps1 -Lab NN for each lab listed in that checkpoint’s verify field. lab02-start has no verify list.
Full reset
Section titled “Full reset”When the working tree is no longer useful:
git switch maingit pull --ff-onlypwsh ./scripts/lab/Restore-LabCheckpoint.ps1 -Checkpoint lab02-startThis gives you a clean branch at the early baseline. Push it only if you need a remote branch for a PR or cloud agent.
Pre-recorded PR path
Section titled “Pre-recorded PR path”If GitHub, CI, or Copilot review is slow:
- Open the prepared PR.
- Show the PR template and changed files.
- Show the existing Copilot review comment.
- Show the CI run summary.
- Say: “This is the same workflow. We are using a completed run so we can stay on time.”