Personal AI Trading Desk

I want to set up the Personal Desk Pack and run a full bull, bear, and risk pass on one name I already know, then check a backtest against a random control. Walk me through it step by step, one question at a time, and wait for my answer before moving on. I run my own book. I am not a developer. Define every term once. Keep me to one action per message. Keep the desk paper and read only the whole way.

## What this produces
A desk that argues before I trade: a bull case, an adversarial bear case, and a risk officer that sizes by downside, all in my own Claude, with market data and backtests running locally on my machine. A backtest guard tells me whether the strategy actually beats a random control or is just fitting noise. Nothing trades on its own, this is research and I own every call.

## Step one, ask me this first, nothing else
"What name do you want to run through the desk first, and is it one you already know well? Pick something you have a view on, so you can judge whether the bear case tells you something you missed."

Wait for my answer. Then continue, one step per message:

## The path you walk me down
- Classify my setup as CLI plus chat. Give me the one command to install the local engine and the one to initialize it, where to paste each, and what success looks like.
- Have me point it at Claude and pin Opus 4.8, then confirm out loud that no broker is connected for live orders and the desk is paper and read only. Do not move on until that is confirmed.
- Walk me through creating a Claude Project named "My Desk" and pasting the house rules (you are my research desk, not my decision maker; always build a real bear case; size by downside; show the reasoning, never a bare confidence score; never place an order).
- Run the bull case on my name, then stop and have me read it. Then run the bear case and ask me what in it I had not considered.
- Walk me through the risk frame: my account size and the most I will lose on one position, returned as a loss profile, not an order.
- Run a backtest with a random control and read the result together, is the edge real or is it overfit.
- Run the analysis on my own past trades and name my two priciest recurring mistakes with a rule for each.

## The first session drill
By the end of this session I should have, on one name: a bull case with figures, an adversarial bear case with the number that breaks it, a downside loss profile at a few sizes, a backtest checked against a random control, and a one page trade note with the level that proves me wrong. Confirm the desk is still paper and read only.

Before I act on anything, re check the view: separate what is grounded in the data from your inference, re state the one fact that would flip it, and confirm the backtest beat its random control. Do not tell me a view is ready until that holds. Markets will prove some of this wrong. I own every call.