Your offline route
Lead the workshop without internet
Keep the same 3:30–5:00 PM schedule and learning goals. Use the full speaking guide for the session; the directions below replace its live demonstration steps.
Full speaking guide PDF Slide-by-slide notes
Before people arrive
- Extract the complete ZIP. Open index.html from the extracted folder. Keep all files together.
- Open webdeck.html and this guide in separate windows. Press V for presenter view, S for notes, and F for fullscreen. If a popup is blocked, allow it or use S.
- Print the participant workbook, three-page capacity matrix, and session takeaway. Use the quick references as table handouts.
- Put the extracted package on a USB drive if participants need their own copy. With no network, participants cannot open links on their own devices unless they have the files.
What to say at the start
“We have everything we need right here. Today we’ll use a made-up application and prepared examples. You can type in the local practice page or work on paper. We’re practicing how to ask a clear question, check the answer, and decide what a person needs to do next.”
3:38–3:50 Build the request
Open the prompt lab. Give partners two minutes to choose a task, two to write the request, and one to swap and remove a guess. Save the revised request in the workbook or local draft fields.
3:50–4:05 Replace the live completeness demo
“Let’s look at the application first. Which information can we point to? Which information is missing? Now we’ll compare our reading with this prepared example. This is a teaching reference, not a response from a model we just ran.”
- Open C-101; read P1–P5 together.
- Ask participants to make their own present, missing, or unclear list.
- Reveal the prepared completeness table. Check each entry against the packet.
- Ask: “What should happen to the instruction in P5? What follow-up would we need for P4?” Record a correction and a human follow-up.
4:05–4:20 Replace live rubric scoring
“Score the example on your own first. Then tell your partner which words support your rating. We’ll use the prepared discussion reference to compare our reasons, not to choose a winner.”
Use the rubric, then reveal the calibration reference. Keep disagreements visible, revise one ambiguous scoring description, and record why a rating changed.
4:20–4:30 Map the checks
Open the workflow activity. Ask pairs to name the person who approves the input, checks evidence, decides, and records the outcome. Add a stop condition.
4:30–4:45 Discuss applicant AI use
Use the flagged essay and translation-help scenario in the policy lab. Ask which published rule applies, what evidence matters, and what a neutral question would say. A flag alone does not establish misconduct. No detector is needed for this activity.
4:45–5:00 Choose and keep a next step
Complete the capacity matrix and 30-day action plan. Ask participants to share one clearer request, one human check, and one rule to clarify. Export their drafts and plan, or keep the printed pages.
| Time CT | Focus | What participants make |
|---|---|---|
| 3:30–3:38 | Frame the decisions | Identify one review task and rate a starting capacity. |
| 3:38–3:50 | Build a structured prompt | Draft a constrained prompt with an evidence requirement. |
| 3:50–4:05 | Demonstrate administrative review | Test synthetic application triage; catch omissions and unsupported claims. |
| 4:05–4:20 | Calibrate a review rubric | Compare independent human ratings before seeing an AI draft. |
| 4:20–4:30 | Protect the workflow | Place a data gate and a named human decision owner in a workflow. |
| 4:30–4:45 | Evaluate applicant AI use | Interpret a detector signal; draft a fair policy response. |
| 4:45–4:55 | Choose a capacity target | Select one capacity, evidence of progress, and a 30-day action. |
| 4:55–5:00 | Commit and close | Share an exit ticket and address remaining questions. |
