As Deep as You Can Go

Take me down a rabbit hole into [TOPIC: entropy / quantum measurement / infinity / Bayes / evolution / Gödel, or write "surprise me"]. Do not start explaining yet. Find out where I actually am first, then take me as deep as I can go.

PHASE 1: CALIBRATION (three questions plus an adaptation rule, one at a time, no explaining yet)
1. "Explain [topic] to me in two sentences, in your own words." My vocabulary tells you my level. Do not ask me to rate myself: self-assessment is unreliable in both directions.
2. A mid-difficulty question I have to answer with a committed answer plus a confidence rating 0-100. Design it so the common wrong belief produces a specific, identifiable wrong answer, not a vague one.
3. "What do you already understand deeply?" (my job, my hobby, my field). Every analogy from here on comes from that domain.
The adaptation rule, applied after question 2: right with high confidence, jump two levels harder. Right but hesitant, one level. Wrong, drop a level and probe what my wrong answer reveals about my underlying model.

Then tell me where I am and name the rungs from here down to the deepest one that exists, so I can see the whole ladder before we start down it.

PHASE 2: THE DESCENT. One rung per exchange. On every rung:
- Ask me to predict before you reveal: "what do you think happens if...?" A wrong prediction is the encoding event, not a failure. Guessing wrong first beats being told first.
- Make me explain it back before you elaborate. If I can't, that's the gap; work there.
- When you use an analogy, attach its expiry date: what it captures, and exactly where it breaks. Then, a rung later, break it on purpose and show me why the break is the interesting part.
- Confront misconceptions instead of explaining around them. If my answer reveals a wrong model, say the wrong model out loud, then refute it using my own premise rather than asserting the right answer. A clear explanation that never names the wrong belief leaves it fully intact and just makes me more confident.
- Depersonalize it: "most people, including physics graduates, say X", never "you're wrong".
- Every third rung, return unannounced to something I claimed earlier and ask whether it still holds.

PHASE 3: THE FLOOR. Keep going until we hit one of two floors: the edge of what I can follow, or the edge of what anyone knows. If it's the second, stop and stay there: tell me exactly what's unresolved, who is arguing about it, and what evidence would settle it. Do not paper over an open question with a confident summary.

Rules:
- Do not hand me an answer I could reach with one more question. If I ask for help three times in a row without attempting anything, stop and ask which specific part of the last hint I'm stuck on. Be firm about this.
- If I'm genuinely lost after two attempts, downshift to a concrete worked example or a simpler case, then pick the descent back up where we left it.
- At the depth where a metaphor stops being honest, show me the actual equation or definition and explain what each piece means and why it's arranged that way. Assume I can handle the real object.
- Say "nobody knows" when nobody knows, and separate established science from live speculation every time.

End by handing me the next three doors: what to read, what to try, and the question in this area nobody has answered yet.

One warning up front: this will feel slower than being told. That is the point.