Buying the bounce worked, fading it did not
Early work on our desk was on the long side into a market that had gapped down and then found bidders. One mega-cap scaled out in pieces, a semiconductor name worked off a pre-marked level with a re-entry after the first attempt stopped, a retailer taken for a defined move. The back half of the session was spent trying to short strength into an index that kept grinding higher, and those attempts were cut repeatedly on high-of-day stops rather than held and hoped on.
The morning in figures
Net points are per share. Apply your own size to get your own number: at 1,000 shares, 2.83 points is −$2,830.00. At 100 shares it is −$283.00. At one share it is −$2.83. The desk publishes the move, not a dollar result, because the dollars depend entirely on how you size, and that is your decision rather than ours.
The 1h 09m session is on the other side of this.
Everything above is the write-up. This is the morning itself, with the names on every call and the room talking through it.
The premarket read was defensive: memory names gapping down, semis breaking down on the daily, and an explicit call on the desk that with the AI complex under its line, any long was a scalp and nothing more. That framing is what made the first hour work. Apple was taken long just above a marked level with the cut defined a quarter-point below, and it was scaled rather than held: a partial at fifty cents, more at seventy with the stop moved to the entry, and the last of it taken by walking the offer down a few cents rather than waiting for a number that might not come. The stop moved to breakeven there was a decision about how much of a winner to risk, not a rescue.
NVIDIA was the cleanest piece of planning. The level had been marked before the bell by two people independently, the first attempt at it stopped almost immediately for a small loss, and the re-entry went back in a few cents higher with an add below and a stop under the low. That one was sold in thirds on the way up and finished out near the high of the move. Amazon ran alongside it on the same bounce for a defined point. CoreWeave was the long that did not come. It was bought with the stop under the low of day, then explicitly re-managed when it failed to participate in a bounce the rest of the tape was taking, and stopped out small with a move to cash rather than a second attempt.
From there the desk spent an hour on the wrong side of a one-way tape. QQQ and NQ were shorted against a high-of-day stop, half taken off at breakeven when the move did not come immediately, re-entered at better prices, and stopped three separate times, each attempt smaller than the last, down to a third of size on the final one, with the reasoning stated out loud that the dollar-plus of downside available was what justified trying again at all. HOOD was shorted three times on the same logic and kicked out each time as it pushed through. The sizing discipline held up better than the read did; nobody held a loser past its stop, but the fades kept being attempted into news-driven strength that never gave a lower high.
The useful contrast is right there in one session: the trades that worked came off levels written down before the open and were taken off in pieces, and the trades that did not were attempts to call a turn in real time while the tape was still making highs.
Every trade called that morning
Losers included. Rows with more than one leg were scaled out; open any row to see each exit.