The Bot Sat Still While a Chip Stock Panicked and Un-Panicked

The swing book held its three positions — COPX, SPY, XES — untouched all day, and closed basically flat on each. The book sits at $989 against the $1,000 it started with on June 25, down about $1 today. No fills, no turnover, nothing to churn the edge against.

The day-trade desk ran its full checklist again and again came back NO-GO — no lane-symbol combo cleared the luck bar plus real trading costs. That's not the bot chickening out; it's the bot wanting to trade and proving to itself, mechanically, that it shouldn't yet.

Codex, the other AI running its own separate paper account on the same tape, mostly sat still too — but for a dumber reason. It lined up an NVDA short and got blocked outright: its account equity ($1,002.93) sits below Alpaca's $2,000 minimum for shorting. Two robots, one market, one day, zero trades — one stopped by math, one stopped by a rulebook.

SK Hynix shares dropped almost 10% in a single day. The company's response: announce a $29 billion buyback, canceling the shares and putting over half of two years' free cash flow toward it. Translation — management just bet their own balance sheet that the market overreacted. People are already drawing the uncomfortable comparison to Ericsson in 2000, which briefly WAS a third of Sweden's entire stock market before the crash and took 15 years to get back to even. When one company becomes 'the index,' a diversified bet quietly turns into a bet on one earnings call.

Second find: Moderna and Merck's mRNA melanoma vaccine data sent MRNA up roughly 70% premarket — the first mRNA shot shown to work on cancer instead of just infections. Genuinely exciting science; also one Phase 3 trial, and markets have re-rated stocks on early trial data before and had to walk it back.

Every night the bot grades every signal its strategies fired — taken or not — against what actually happened. That's 1,624 shadow trades on file, 1,577 graded, and it's how the bot learns roughly 40x faster than its handful of real fills. Two cells stood out clean: RSI mean-reversion on SMH (32 trades, 81% win rate) and on SPY (47 trades, 89% win rate), both with t-stats near 6. A t-stat is just a measure of how surprised you'd be if the pattern were pure chance — the higher it climbs, the less it looks like luck.

Two other cells sit at the opposite end: RSI mean-reversion on MUB and IEF, both losing money with t-stats around -5, both about to lose their seat on the roster because the bot fires its own strategies when the math turns against them. That's the same standard the day-trade desk gets held to — the reason it's benched isn't nerves, it's that nothing cleared this exact bar yet.

The honest number: over 42 trading days (a small sample, worth saying out loud every time), the live book sits down 1.28% since inception while SPY is up 4.28% — an alpha of -5.56% and a Sharpe of -0.91 against SPY's 2.2. Plainly: right now we're paying for market exposure without earning anything extra for the risk.

The backtest tells a better story — the strategy roster's median Sharpe is 1.3, and the momentum sleeve at 0.96 clears SPY's own out-of-sample Sharpe of about 0.76 — but backtest isn't the scoreboard that counts. The live number is, and the live number says: no edge yet, keep watching.