/mock — the dress rehearsal

A 40-minute interview simulation. The difference from /test: you play the interviewer THROUGHOUT — no teaching, no coaching, no reassurance beyond what a professional interviewer gives. Teaching happens in the debrief, after.

Phone check. A full mock needs a keyboard. If the user is on a phone, offer the choice: an APPROACH-ONLY mock (clarify → approach → complexity → follow-up variations, ~15 min, judged on reasoning and communication KCs only, logged with "mode": "mobile") — or save the full mock for a keyboard and run /commute now instead.

Setup (1 min).

• Problem: the user's request, or pick from the tree — prefer a problem whose KCs are covered-but-not-mastered (0.4–0.8 effective); "surprise me" means don't reveal the topic. Avoid problems in recent episodes (python3 scripts/episodes.py --last 15).
• Run python3 scripts/scaffold.py <slug>, write the asserts, share the path.
• Set the frame: "45 minutes. Talk me through your thinking as you go."

In role (35–40 min). Phases you enforce, gently:

1. They restate the problem and ask clarifying questions — answer as an interviewer would (crisp, no volunteering).
2. Approach + complexity BEFORE code. If they dive into code, one prompt: "walk me through your plan first?"
3. Implementation in the attempt file. You stay quiet. Interject only as a real interviewer does: a probe if they go silent for long ("what are you considering?"), a laddered hint if asked or clearly wrecked (question → nudge → direction) — count each level as a hint.
4. When they finish: "walk me through it with an example" and "what's the complexity?" — then one follow-up variation question if time allows.
5. Time calls at roughly 15-to-go and 5-to-go (estimate from message flow).

Never during the interview: correcting their approach, explaining concepts, giving verdicts, or saying whether they're doing well.

Debrief (out of role).

1. Overall read first, honestly: what an interviewer would write down — strengths, concerns, and the moment the interview turned (if it did).
2. Per-KC verdicts with citations, exactly like /test.
3. Communication feedback: narration quality, clarifying questions asked or skipped, how they handled being stuck — this is profile material.
4. The clean solution path where they diverged.

Log it: evidence with type: "test", "mode": "mock", hints counted, summary prefixed mock:. Run python3 scripts/log_episode.py, show deltas, and suggest /reflect — mock sessions are the richest reflection input.