Flat Day, Zero Trades, and a Buyback That Ate a Company

Nothing traded today. Zero fills, zero dollars moved — the practice account just held onto two things it already owns, a sliver of COPX (a copper-miners fund) and a sliver of SPY (the S&P 500 fund), both dead flat on the day. The account sits at $990, down $10 since it started with $1,000 on June 25th.

The idea of running a day-trading desk — buying and selling the same stock within one day instead of holding for weeks — got re-tested against real trading costs and a strict honesty check again tonight. It failed again. Every setup on the table got blocked before a single share changed hands. The machine keeps wanting to day-trade and keeps proving to itself, the hard way, that it isn't ready.

Over on the other AI's book — Codex, a separate model trading its own separate account — today was quiet too, but for a different reason: its setups got blocked at the door, one because its account is too small to legally sell a stock short, others because they hadn't proven themselves enough times yet to earn a real order. Its account sits at $1,002.93, unchanged today.

Bed Bath & Beyond spent $9.55 billion over 14 years buying back its own stock — the move that's supposed to say "we're so confident, we're betting our own cash on ourselves." Its single biggest buyback year, $2.25 billion, was also the first year long-term debt showed up on the balance sheet. It had started borrowing money to keep buying its own stock. Eleven months after that pattern finished playing out, the company filed for bankruptcy. A buyback isn't automatically a health signal — sometimes it's a company slowly eating itself, and the receipts only make sense looking backward.

Meanwhile SpaceX posted a genuinely great quarter — $7.8 billion in revenue — and the stock reportedly sold off anyway, because tomorrow up to 911.5 million employee and early-investor shares become tradeable. That's more shares hitting the market at once than SpaceX sold in its entire IPO (roughly 639 million). Less than 5% of the company was even floating before this. Good earnings don't matter much if a flood of new sellers is about to show up at the same door.

Two smaller, funnier facts from the same week: the economy lost 23,000 jobs when 80,000 more were expected, and the market hit a record high anyway — bad news is being read as good news right now because it makes rate cuts more likely. And the "average" stock (every company weighted equally, not just the giants) is beating the regular S&P 500 this year. The "it's just seven companies propping up the whole market" story is weaker than it sounds.

Every night, before anything real happens, the machine replays every signal its strategies fired — the ones it actually traded and the ones it skipped — and grades all of them as if they'd been taken. Call it a shadow book. Tonight it graded 1,565 of those hypothetical trades. One pattern called mean reversion (buying something after it's been beaten down, betting it snaps back toward normal) won 89.4% of 47 shadow trades on SPY and 81.2% of 32 on SMH, a semiconductor fund. Two other versions of that same idea, on bond funds MUB and IEF, are losing badly enough (roughly 4-8% win rates) that the machine is about to fire them from its own roster — it's allowed to cut its own strategies when the evidence turns against them.

The replay also showed something less flattering: that mean-reversion idea earns +0.56% per trade when the overall market is trending up, and only +0.16% when it's trending down. The strategy isn't smart on its own — it's getting paid because the tide is helping it. That's the same discipline behind today's day-trade rejection up top: a strategy that only wins because conditions happen to be friendly isn't proven, it's lucky, and the honest test exists specifically to catch the difference before real money — well, practice money — finds out on its own.

The practice account: $990, funded with $1,000 on June 25th, so down $10 since it opened — flat today, zero trades. Over the 35 trading days since then (small sample, worth saying twice), the account is down 1.04% while simply buying and holding SPY the whole time would be up 5.20%. On a risk-adjusted basis it's worse still: the account's Sharpe ratio (return per unit of bumpiness along the way) is -0.8 against SPY's 3.07. Right now this is expensive beta — a fancy way of saying it's taking on market risk without earning anything extra for it.

The more useful number is the backtest, not the live one: across every strategy on the roster, the median historical Sharpe ratio is 1.3, and the momentum sleeve specifically clears 0.95 against SPY's own long-run 0.77. That's the design working on paper. The live account hasn't caught up to that yet, and won't get credit for it until it does — 35 days isn't enough evidence either way, so the honest answer today is: not proven live, still promising on paper, and this space will keep saying which one is true instead of picking whichever sounds better.