High-stakes control

Before the session

• Skim the three outline lessons (what counts as high stakes, monitoring and mitigation, environment construction) and the readings under each, so you can route questions to sources.
• Open the defer-to-resample demo and try each readout once; you will drive it on a shared screen.
• If you run a classroom, check the dashboard for who reached the two guided walkthroughs (collusion prevention, legibility protocols)

Agenda

Times are a running clock for a 60-minute session

Warm-up: what is high stakes? (0:00–0:10)

Draw two empty columns (high-stakes / low-stakes) on the whiteboard. The room supplies failures from earlier modules; sort them live into the columns using the definition from the win-continue-lose reading.

Prompts:

• "Which column does this failure go in?"
• "Are there failures that are ambiguous, what makes it this way?"

Rogue deployments on the whiteboard (0:10–0:25)

One participant draws the threat model from the rogue deployments reading from memory; the room corrects out loud as they go. Then add where internal deployments via external APIs fit on the drawing, and finish with the monitoring discussion (prompts below).

Prompts:

• "Which arrow on this drawing would you monitor first, and what would that monitor actually see?"
• If the drawing stalls: hand the marker to a second participant and let the first narrate instead.

Resampling, live (0:25–0:40)

Recap blocking as the baseline response, from the monitoring section. Then step through the defer-to-resample demo on the shared screen — before every readout, the room predicts win, continue, or lose, and says why, before you click. Connect the mechanics back to the Ctrl-Z reading.

Prompt:

• "When would you rather block than resample?"

Walkthrough debrief (0:40–0:55)

Split into two groups, one per guided walkthrough: collusion prevention and legibility protocols. Each group reconstructs the decision its walkthrough forced and whether the walkthrough's evidence changed its members' initial choice. Groups then report to each other; compare the two head-on.

Prompts:

• "What was the exact moment in the walkthrough where your choice flipped, if it did?"
• "What additional experiment would have settled your walkthrough's question?"

Wrap-up (0:55–1:00)

Each participant names the mitigation from this module they would implement first in a real lab. Collect open questions.

Prompt: "What is the cheapest thing this module described that a lab could ship this quarter?"

After the session

• On the classroom dashboard, check who completed the two guided walkthroughs after the meeting; the debrief usually motivates the holdouts.
• The module's readings carry spaced-repetition recall cards; remind participants that the review tab resurfaces what this session touched.