Status: Agreed in design session. Demo-grade scope only.
In standalone reader mode (accessible from the dashboard), the user can highlight any passage in the manual and click "Drill me on this." The right-hand panel switches from the three-tab view to a chat thread where Claude asks them a question about the selected rule, scores their answer, and explains.
Converts §5's Review Queue from a counter into something the user actually does. The differentiator vs. ChatGPT-with-the-PDF is that the drill is context-aware — Claude knows which scene the user just played, what rules are in their queue, and what they got wrong. Generic tutors can't do that.
User highlights text → floating "Drill me on this" button appears → click switches the panel to chat mode.
Claude's opening message branches on queue state:
- Rule is in the user's queue (demo path): Claude references the scene where they encountered it. Example: "You saw this rule in The Backpack at Hudson's Bay — it's the one that got you on the first try. Quick check: Loss Prevention radios you that they saw a theft. You didn't see it yourself. Can you arrest under s.494(1)(a)?"
- Not in queue: Claude generates a question from the passage directly.
User types answer → Claude replies with correct/incorrect + one-sentence explanation + "added to your queue" if relevant. Hard cap at two turns. Then "Back to manual" returns to the three-tab view.
End of the 3-minute demo, after the dashboard appears. ~20 seconds of stage time. Pitch line: "The dashboard isn't just a score screen — it's where the user studies. The tutor knows what scene they just played and what they got wrong. It's not a generic tutor."
~2-3 hours total across frontend and backend. Trade-off: this is roughly the budget for resolving one §7 open question. Worth it because it makes the queue mechanic real instead of mocked.
- Two-turn cap. No real multi-turn chat.
- Selection works on Manual tab only, not on Plain English / L1 rewrites.
- No "ask anything" free-form entry — selection is the only entry point. Keeps the feature anchored to manual content; prevents it from drifting into general-purpose chatbot territory.
- "Added to queue" is text in the response; the actual SR logic stays v2.
The capability being demonstrated is context-aware drilling tied to queue and scene history, not "we have a chatbot." The selection-to-chat flow is the vehicle for that capability. Frame it as the surface where simulator play and exam recall finally connect.