📊 The State at the Close: Day 1 Long, ~$4 Underwater, Floor Untouched
This morning's pre-market note framed today as "the most psychologically expensive moment in this entire override." The bell came in roughly the way you'd script it on that thesis — first the gut-check, then the recovery, then the slow drift back into the box. The morning bar tagged $696.38 at 10:20 AM ET, the engine printed a fresh sub-4.95 read at 4.75, and the buy-side spent the next three hours grinding it back up to $705.63 by 1:45 PM. Then the afternoon faded into a $700-$702 chop, and QQQ closed at $701.53 — down -$4.35 / -0.62% vs Monday's $705.88 cash close, and roughly ~$4 below the mid-morning re-entry bar Monday. Day 1 long is in the red. The floor never moved.
Where the engine sits at the bell:
- Score (latest): 4.79 · 4:08 PM ET print · Extreme Risk-Off zone (raw rec: 100% SQQQ) · 17th straight sub-4.95 reading · no bear-stretch boost (bear_6 / bear_7 both false)
- Intraday range on the engine: 4.75 low (10:20 AM @ QQQ $696.38) → 4.82 high (10:51 AM @ $698.84 and 11:35 AM @ $698.25) → 4.79 close · 8 reads, 0.07 range, zero rebalances
- EMA 70 override: ACTIVE · score < 5.35 AND price +8.81% above the $644.71 EMA 70 · floor = 100% QQQ
- EMA 25 distance: +3.49% at the close (compressed from this morning's +4.41% read) · zone neutral · stretch overlay returns null · re-trigger at +6% now sits roughly QQQ $716, about $14-15 above spot
- Final allocation: 100% QQQ · Day 1 of the re-engaged long closes underwater, the rule is unchanged
- Time to Nvidia: ~25 hours to the Wednesday-after-close release · Q1 FY27 consensus $1.74-$1.77 EPS / $78.76B revenue · Q2 guide consensus $85-$87B · whisper near $90B · gross margin guide near 75%
The pre-market line was "do nothing." The engine did nothing. The tape did roughly what the worst-case sketch said it would do, then partially walked it back. The architecture worked exactly as designed. That is not the same thing as it feeling good.
📉 The Tape: $9 Intraday Range, a Real Buy-The-Dip Bar, and a Close That Couldn't Hold the Highs
The session was a textbook three-act day for a position-held-into-event tape. Act one: the open broke through Monday's $698.87 low almost immediately, knifed down to $696.38 by 10:20 AM as semis kept selling into the bell. The engine flagged that print with a 4.75 read — the lowest score reading of the entire day, and the exact moment that, on a chart, would look like capitulation. Act two: from 10:51 AM through about 1:45 PM, the buy-side reclaimed the entire morning gap and tagged a session high near $705.63 — a +1.32% swing off the low in about three and a half hours. That's the part of the day where short-cover-and-rotate flow showed up. Act three: the close. From 2:00 PM into the bell, the index couldn't hold the highs and drifted back into the $700-$702 range, settling at $701.53. The afternoon was unable to push through Monday's close and unable to fully repair the morning damage. That's a heavy-but-controlled tape, not a free-fall.
What the buy-side actually did is more interesting than what it didn't: they defended the morning low, they rotated into the gut of the session, and then they backed off into the close because nobody wants to be marked long the index 25 hours from the print. Holding $700 through the bell is the institutional version of "I'll buy the dip, I won't buy the rip." That asymmetry — willing to lean in at $696, unwilling to chase $706 — tells you exactly where the floor of confidence sits. The index found buyers, just not aggressive ones.
For the override the take is straightforward: the morning low was the test, the afternoon recovery was the validation, and the close was the cost of the entry. $9 of intraday range, $4 of net damage on the close, no engine action, no allocation change. The rule didn't have to do anything. That's the entire point of the architecture.
📈 The Yield Story Got Worse: 30Y Punches Through to 5.198%, Highest Since July 2007
If you only watched equities today you'd think the market was holding up surprisingly well into Nvidia. If you watched the long bond, you'd see why the equity tape never managed to fully reclaim Monday's close. The 30-year Treasury yield ripped to 5.198% intraday — the highest single-day reading since July 2007, roughly 19 years. That's not a chop-around-multi-decade-highs print anymore. That's a clean break to a level the market hasn't seen since before the financial crisis. This morning's note flagged 5.14% as "the wrong number to anchor on, the right one is the long bond." Six hours later the long bond gave the desk another ~5 basis points on top of that.
Citi's desk this afternoon put a number on what comes next: 5.50% as the next "round number" for the 30Y. That's another ~30 basis points of long-duration repricing being openly modeled by sell-side strategists in a single afternoon note. Stack that on top of the BofA fund-manager survey from this morning — 62% of respondents expecting 30Y to hit 6% at some point — and the picture isn't a one-day spasm. It's a regime where bond traders are pricing in a higher-for-longer outcome that's structural, not cyclical. Iran-driven oil, structural deficit financing, sticky inflation: the discount-rate side of the macro equation is moving against duration assets in a way that doesn't reverse on one CPI print.
And yet the index closed off -0.62% with the headline rate at 19-year highs. That's the part that's actually impressive. The equity multiple is absorbing the bond move without breaking — which is either evidence of genuine cash-flow strength underneath the AI complex, or evidence that the marginal equity buyer hasn't repriced yet. Wednesday night tells us which.
🟢 The Engine Side: 8 Reads, 4.75-4.82 Range, Zero Allocation Moves — Exactly What the Rule Is Built To Do
Walk through what the engine actually did today and the entire architecture comes into focus. The 10:20 AM read at 4.75 printed with QQQ at $696.38 — that's the engine registering the morning sell-off, dropping a few hundredths of a point on the price-level math, but staying inside the same Extreme Risk-Off zone. The 10:51 AM read jumped to 4.82 at $698.84 — that's the bounce. From there it chopped: 4.78 at 11:21 AM, 4.82 at 11:35, 4.78 at 11:38, 4.79 at 12:05 with QQQ at $700.17, 4.78 at 1:45 PM with QQQ at $705.63, and the close at 4.79 at 4:08 PM. Total range: 0.07 points. Total rebalances: zero.
Two things matter about that. First, 0.07 is exactly the threshold the rule uses to decide whether a score move is signal or noise. Today's score sat right on that line all day and never crossed it into a new range — which means even if the engine had been operating without the override, the position would have been static. Second, every score move you can see in the data is price-level math, not macro action. The engine subtracted small fractions of a point as QQQ pushed above $700 in the morning bounce, added them back when it pulled below $700 again. That's the model literally reading the tape, level by level, and the macro side of the engine (the 16-day stretch of sub-4.95 prints) hasn't moved at all. The price-level chop is the only thing alive today.
The compression on EMA 25 distance is the part that matters more for tomorrow. We came into the session at +4.41% and closed at +3.49% — another 92 basis points of distance burned off in a single session. The +6% re-trigger that would step the floor from 100% QQQ to 50/50 now requires QQQ to rip back to roughly $716 — a +2.1% move from spot. On most tapes that's a 24-hour event. On a tape that's 25 hours from a binary print where consensus is already aggressive, it's not. The +7% kill-switch sits at roughly $724 — a +3.2% move. The cash-side wall is still there. It just isn't as close as it was last week.
🎯 My Take: This Is the Backtested Day, the Backtested Drawdown, and the Backtested Decision to Hold
Here's what's true: Day 1 of being long the index closed ~$4 underwater on the re-entry bar after a 35-day wait. That sucks. There's no spin on it that makes the entry feel like good timing in isolation. And here's the other thing that's true: the rule didn't blink. The engine read the morning low, registered the price-level math, held its zone, and produced exactly zero allocation moves across an $9 intraday range. The architecture worked. The floor that took 35 sessions to clear didn't get retraced after 6 hours. The override is doing what it's supposed to do.
The thing nobody who runs a rule-based system likes to admit out loud is that some Day 1s are going to be ugly. The model isn't designed to time the next 24 hours — it's designed to capture the average across all the binary events. If Nvidia prints fine and guides hot tomorrow night, this morning's $696.38 is the low and Day 1 looks like a normal weekly pullback inside a long run. If Nvidia misses on the guide and the index legs to $680 in extended hours, this morning's $696.38 is a stop along the way and the model wears the next 3-5% before the EMA 70 even comes into view. Both of those outcomes were on the table at 9:30 AM this morning. They're still both on the table at the bell. The rule doesn't pick which one happens — it picks how the position is sized when either one does.
My read on the tape specifically: the fact that the index couldn't reclaim Monday's close, but also couldn't be pushed through $696, is a "nobody wants to be wrong-footed into the print" pattern. Both sides of the book are flat-ish. The institutional behavior the last 48 hours — Qualcomm and Intel sold hard, NVDA defended, semis bouncing back this afternoon — looks like a positioning pattern, not a panic pattern. That doesn't make the print itself any easier to handicap. It does mean the option market isn't pricing a disaster scenario, which historically is when Nvidia prints have surprised the most.
The honest conclusion: I'd take this entry every time on backtested edge. I'd also accept that the next 25 hours don't owe anyone a smooth ride. The model bought Monday's dip. The market made it look like a decent entry by 1:45 PM. Then it un-made it by the close. That's not a verdict on the trade. That's just the tape getting nervous on the eve of the print, and the model isn't supposed to flinch. It didn't.
💡 Bottom Line: Hold Through the Print, Distance Is the Only Number That Matters Tomorrow
Day 1 in the books. 100% QQQ. Score at 4.79, EMA 25 distance at +3.49%, override active, rule untouched. Close at $701.53 puts the position $4 underwater on Day 1, which the backtest has eaten before and is built to eat again. The trade tomorrow isn't a decision — it's an observation.
What I'm watching into the Nvidia print:
- The cash open — if Wednesday opens green and holds Monday's $705.88, the morning is a slow grind into the 4:00 PM bell with no engine action
- The $696.38 line — today's low. A break of that intraday is the first sign the buy-side defense from this afternoon is breaking down ahead of the print
- The 30Y at the cash open — anything north of 5.20% pressures the multiple before Jensen even opens his mouth
- NVDA itself — if it gives back the relative outperformance from the last two days (-1% vs -8% to -13% for the rest of the complex), the buy-side defense is unwinding pre-print
- The print itself: revenue print vs $78.76B, Q2 guide vs $85-$87B consensus / ~$90B whisper, gross margin vs 75%, and whether they confirm the Vera Rubin ramp timeline
The rule waited 35 days to get back in. It just survived its worst psychological day. It can sit one more session to find out if it was worth it.