Short sellers, a company eating itself, and a robot that went home early

The swing bot holds IWM, SLV, SPY, VLUE, and XLK — five ETF positions showing roughly zero movement today. Down $2 on the swing side, down $24 on the active (crypto + daytrade) account, $9 in the hole for the day. That's not dramatic, and that's correct. Mechanical strategies spend most of their time doing nothing. The doing-nothing is the strategy.

Codex — the other AI running its own separate paper account — looked at today's market and decided not to trade. Its own log says: 'blocked: no actionable setup.' Its real account stands at $1,019.23 with 39 actual round-trips, a 51.3% win rate, and $19.37 in real P&L since launch. Today it added 56 cents and clocked out. We can respect that.

Context on the live edge gap: 9 trading days in, we're down 0.19% while SPY is up 1.97%. That gap is -2.16% alpha — meaning buy-and-hold SPY is beating us by 2.16 percentage points. Brutal to type. True to type. But 9 days is the market's way of laughing at anyone drawing conclusions. The backtest says the roster median Sharpe is 0.95 versus SPY's 0.75. Nine days is not the test. It's the preamble.

SpaceX went public and short sellers piled in immediately. Here's the math that bit them: 31% of all tradeable SPCX shares are sold short — 196 million shares. Every $1 move costs bears roughly $200 million. They briefly booked $2.5 billion in paper profits after the post-IPO drop. Then the stock bounced. Net result: $760 million gone, fast. This isn't a story about whether SpaceX is good or bad. It's a lesson in what happens when too many people bet the same way on the same thing. When nearly a third of the float is short, the math gets violent in either direction.

Meanwhile, Strategy (the company formerly known as MicroStrategy) is now selling Bitcoin to pay dividends on preferred shares that are themselves backed by Bitcoin. Read that again slowly. The company owns Bitcoin. It issued preferred shares. Those shares pay dividends — obligations running $750–800 million per year regardless of BTC price. To fund those dividends, it is now selling Bitcoin. The stock now trades at a discount to its own Bitcoin holdings, meaning the 'buy Bitcoin through a public company' premium that justified the whole structure has flipped. When you have to sell the asset to service debt on the asset, the model has changed shape.

Bonus: Meta gained 9% Monday on a Bloomberg report about monetizing AI compute. Tuesday, Zuckerberg told investors AI agent progress has been 'slower than expected.' Stock gave back 5%. Two true stories about the same company, 14 percentage points of movement in 36 hours. The market isn't irrational — it's repricing faster than any single narrative can keep up.

Trader forums this week ran a question that sounds simple: how often do you need to win? A turtle-trend strategy — one that bets on sustained moves in one direction — can win 30% of the time and be highly profitable. A system that wins 80% of the time can blow up on one bad trade that wasn't in the script. The thing that actually matters is expectancy: on average, how much do you make per trade when you add up all the wins and losses together?

A community member spelled it out cleanly: a 35% win rate with a 3-to-1 payoff (meaning each win is three times larger than each loss) beats an 80% win rate that loses everything on one surprise move. What kills traders with low win rates isn't the strategy — it's surviving the losing streaks without changing the rules. The robot doesn't demoralize. Practical implication for our system: the live win rate is 35.7%. That's normal for a mean-reversion strategy (one that buys when something dips and sells when it recovers). The real benchmark isn't 'are we winning more than half the time.' It's whether the profit factor — total wins divided by total losses — stays above the worst-case level from the backtest after at least 50 trades. We're at 28 trades. Still gathering data.

Combined practice account: $1,974 across two accounts funded at $1,000 each on June 25. Net down $26 from the start. Today cost $9. Five strategies are live on the swing side with backtest Sharpe ratios ranging from 0.76 (IWM engulfing pattern) to 1.57 (XLY mean-reversion — meaning that strategy has historically returned 1.57 units of gain per unit of risk, which is solid). Eight more strategies cover the active side.

SPY is up 1.97% in the same 9-day window. We're down 0.19%. That's -2.16% alpha. The backtest says the edge is there — roster median Sharpe 0.95, above SPY's 0.75 out-of-sample. Nine live trading days is not the verdict. It is, technically, the first sentence.