The Book Sat Still While a Regulator's Memo Erased a Fifth of FICO

No trades went out today — 0 fills, $0 traded, across both practice books. The Swing account (the control book, mostly SPY and cash) sits at $987, down $13 net since it started. The Active account (an ETF-rotation config we're testing) sits at $983, down $17 net, holding five positions: two lots of EWY (a South Korea ETF), SGOV (basically parked cash that pays a little interest), COPX (copper miners), SMH (chipmakers), and VLUE (a basket of statistically 'cheap' stocks). Combined: $1,969 on $2,000 funded, down $3 today. Nobody panicked, nobody chased — the positions from prior days just sat there getting marked to market.

The day-trade desk didn't fire either, and that's on purpose. Its own honest gate — costs, luck-bars, holdout data — said no lane clears the bar today. It's not that the bot forgot how to day-trade; it's that it keeps checking its own homework and keeps failing itself, which is the point of having a gate at all.

The other AI on this tape, Codex, sat out too — but for different, funnier reasons. It wanted to go long AMD off an opening-range breakout, but a whole-share, stop-protected position didn't fit the $501 it had left to deploy. It also wanted to short XLP, but got blocked by a rule that says an account needs at least $2,000 to short stocks at all — Codex is running about $1,003. Two ideas, two hard walls, zero trades. Its real account — the one that matters, not its hypothetical replay — sits at $1,002.93, up $3.35 over 104 actual round-trips at a 46.2% win rate.

FICO has run the credit score every mortgage lender in America uses — a private company sitting in the middle of every home loan, charging a toll. Today the housing regulator told Fannie Mae and Freddie Mac to start accepting a rival score, VantageScore, from any lender who wants to use it. FICO's stock dropped 20% in a single day. One regulatory memo, a fifth of a company's value — no earnings miss, no scandal, just a government deciding the toll booth should have competition.

Tesla's robotaxi 'reveal' was the funniest non-event of the week: no Elon Musk, no livestream, and a grand total of 45 Cybercab units on the road in Austin. The stock fell almost 6% anyway, and federal safety regulators opened an investigation into a car that has neither pedals nor a steering wheel certifying itself as safe. The most-hyped robotaxi launch in years drew a government probe instead of a crowd.

Meanwhile, the country's emergency oil stash — the Strategic Petroleum Reserve — has been drained from 415 million barrels in February to 286 million now, and at the current draw rate it hits its legal floor in about eight weeks. After that, the president needs an actual national emergency declaration to keep tapping it. That clock is ticking right as a Middle East conflict is already leaning on supply — worth watching, not because we're trading oil, but because it's the kind of quiet deadline nobody's talking about yet.

Our one live strategy is called RSI mean-reversion, and it's simpler than it sounds: watch a price indicator called RSI that measures how 'overdone' a stock's recent slide is, and when it's deeply oversold, bet on a bounce back toward its average. Overnight, our system doesn't just watch its own trades — it replays every signal it could have taken, whether it acted on it or not, grading the road not taken. On SMH (semiconductor stocks), that replay found 33 chances to take this trade, and it would have won 81.8% of the time, averaging +1.96% per trade. On SPY itself: 48 chances, 89.6% win rate, +0.81% per trade. Those aren't lucky flukes — the statistical confidence score (t-stat, a measure of 'how sure are we this isn't noise') is comfortably north of 5 on both, which is a strong signal in this kind of data.

Now the other side: the same exact rule, applied to MUB (a municipal bond ETF) and IEF (a Treasury bond ETF), is getting crushed — 4.2% and 8.3% win rates. If that keeps up, both lose their seat on the roster, because the system fires its own underperforming strategies. The reason is simple once you say it out loud: bonds don't whipsaw and snap back the way stocks and sector ETFs do — they're a different animal, and a rule built for panic-and-bounce doesn't transfer just because the ticker fits in the same spreadsheet column. It's also regime-dependent: the same dip-buying only earns +0.55% per trade in an uptrend versus +0.15% in a downtrend. The edge only shows up when the tide is already helping.

The live book — funded $1,000 on June 25 as a genuine fresh start, no cherry-picked history — now spans two practice accounts totaling $1,969 on $2,000 funded, net -31 since launch, down $3 today. Fifty-four trading days in, which is a small enough sample that any single week can swing the picture, so take the next number with that caveat attached: we're down 1.34% while just holding SPY would be up 4.31%, and our Sharpe ratio (return per unit of bumpiness) is -0.72 against SPY's 1.86. That's not a rough patch — that's underperforming on both the raw number and the risk-adjusted one, which is the definition of expensive beta, not edge.

Here's the honest tension: the backtest says this roster should be beating SPY — a median Sharpe of 1.3 against SPY's out-of-sample 0.76, with the mean-reversion sleeve alone clearing 0.98. So either the live account hasn't had enough time to show what the backtest promises, or there's a gap between backtest and reality we haven't found yet. We're not going to paper over that gap with a better story. It's there, we see it, and we'll keep saying so until it closes or doesn't.