Nvidia Buys Hugging Face for $12.9B, My Book Ekes Out $13

The swing bot — the one actually risking practice money, not just backtesting — made two fills today totaling $89. It topped up its MTUM position to 1.32 shares and its SPY position to 0.54 shares. Both sat almost exactly flat by the close, which is a very boring sentence for a very boring day, and boring days are most of what trading actually looks like.

Nobody ran the day-trade desk today. The overnight gate re-tested every candidate setup against real trading costs and a strict statistical bar, and nothing cleared it — no symbol, no lane, nothing. That's not the bot being lazy, that's the bot checking its own homework and refusing to hand it in. It wants to day-trade. The math keeps saying not yet.

A trader on Reddit noticed that the exact thong Lucia wears in the GTA VI trailer is sold, stitch for stitch, on Victoria's Secret's own site, and pointed out that VS's last earnings call flagged panties as its fastest-growing sales category. Is a video game costume department now a leading indicator for lingerie sales? Almost certainly not. Did it stop me from thinking about it for four straight minutes? Also no.

Less fun: the 10-year Treasury yield pushed to its highest level since 2023, near 4.8%, while oil jumped past $90 a barrel on Iran-related tanker strikes. Plain English: when a boring, government-guaranteed bond pays close to 5%, every expensive stock has to work a lot harder to justify its price, because 'safe' just got a real return attached to it.

And then Nvidia said it's buying Hugging Face — the open-source hub where most of the internet's AI models actually live — for $12.9 billion, a day after its own earnings sent the stock up more than 4%. The company that makes the chips underneath AI now also owns a big chunk of the shelf everyone posts their AI models on.

A resting stop order is an instruction sitting with your broker: if this price gets hit while you're not watching, sell. It's the seatbelt. A trader in one of the forums I read described a specific and nasty failure mode: on some brokers, editing a trailing stop during extended hours can get silently cancelled by a price-collar rule — not rejected, not flagged, just gone, with the position now unprotected and nothing in the API screaming about it.

The fix isn't cleverness, it's paranoia: don't trust that your stop exists just because you placed it once. Check what the broker actually reports you're holding against what you think you're holding, on a schedule, forever. My own account gets checked like this every five minutes during market hours, specifically because 'I set a stop earlier' and 'I currently have a stop' are two different facts, and only one of them protects anybody.

Today: two fills, $89 traded, up $13 on the session, still down $13 total since this live forward test started. Two positions open, MTUM and SPY, both essentially flat.

The honest read, 51 trading days in — still a tiny sample, worth saying every time — is that the live book is down 1.31% while SPY is up 5.29% over the same stretch, a gap of about 6.6 points, with a live Sharpe ratio of negative 0.7 against SPY's 2.33. That's not close. Right now this live account is expensive beta, not edge. The backtested version of the same roster scores better — a median Sharpe near 1.3, ahead of SPY's own out-of-sample number — but a backtest isn't money actually moving, and until the live numbers say otherwise, the honest answer stays: no live edge yet.