The Morning Bell No. 8: A Beat Nobody Had Priced
Updated: Oct 4
Pricing update, September 10, 2026: expert working sessions are $250 for 60 minutes. Build in a Day starts at $1,500; 30-Day Implementation starts at $6,000. Earlier package inclusions, credits, and hourly support terms below are historical. See current services and pricing for scope and terms.
Snowflake printed 62 cents against a 26 cent estimate after the bell last night and it is trading up about 24 percent this morning. That is the single biggest overnight fact on my screen. My desk still might not touch it, and the reason why is the whole edition.
I am not a financial advisor and nothing here is financial advice. This is education and entertainment built from public filings and my own research notes. Always do your own research and size like you might be wrong.
The tape before the bell
Every number here is my own fetch at 8:40 a.m. ET, premarket, against the official prior close:
SPY 766.69, up 0.20 percent against a 765.16 close.
QQQ 709.72, up 0.07 percent. Flat, honestly.
IWM 294.95, up 0.32 percent.
TLT 82.22, up 0.33 percent. Long bonds bid, which argues against the hawkish read one of my sources is running.
GLD 410.14, up 1.83 percent. The biggest move on the board and nobody is talking about it.
Bitcoin 78,306, up 0.81 percent against a 77,680 prior close.
The regime call from the 7:46 run is AMBER. SPY sits below its 20 day average at 768.98 and above its 50 day at 755.34, so it fails the green test on the first clause and never gets near the red one. Real VIX was 15.37 against a 15.20 prior close, up 1.12 percent, and the red trigger would have needed 18.24.
AMBER is not a shrug. It cuts the size multiplier on anything the system takes to 0.6 and it means the tape is not paying me to be brave.
What the system flagged overnight
Two prints, both semis adjacent, both beats, and the tape sent them in opposite directions.
Snowflake. Beat and raise, which is the full credit class for post-earnings drift. Product revenue growth guidance for the full year went up to 36 percent year over year in the company release. An estimate at 26 cents against a print at 62 cents is not a small miss by the analysts, it is a repricing that takes days to finish.
Broadcom. Record quarter, revenue 29.6 billion, up 86 percent, and AI semiconductor revenue of 16.7 billion, up 221 percent year over year. Then Q4 revenue guidance came in around 34.8 billion against consensus near 35.0 billion, and the stock is down about 3 percent premarket. A beat with a soft guide is not the drift class, so it does not become a card.
My panel of tracked analysts was no help. Only 5 of 13 sources published inside the window, which is 38 percent and under my own confidence floor, so convergence is suppressed rather than reported.
And the blind spot is worth saying out loud. Not one of those five mentioned Snowflake or Broadcom. They were positioned on macro, gold and oil while the two prints that actually moved money went off behind them.
The setups on my screen
Research notes only. These get graded tomorrow, win or lose, and nothing here is an instruction.
Snowflake long, one card. Fill rule is market on the open, because on an earnings gap a limit sitting in the hole is void by my own rules. Reference fill 377.42 from the 7:50 a.m. pull, stop 361.64, target 408.98. That is an earnable R of exactly 2.00.
Here is the part I would rather not print. At a five session horizon the card auto-fails my own test: it needs 31.56 points of move and 0.8 times expected range is 23.52. It only passes at ten sessions, where expected range is 41.58. So the horizon is ten, and the five session failure goes on the page instead of in a drawer.
Crowding is high, the base rate is UNPROVEN at n equals zero because my grading file does not exist yet, and the falsifier is simple. Opens under plus 2 percent, or loses VWAP inside the first thirty minutes, and it never triggers.
Near misses, with what would flip each:
Broadcom: flips long on a reclaim of 367.24 on above average volume, flips short on a second day of lower highs.
USO: a level from one of my tracked analysts at 142.50. Flips on a close above it.
NTAP and HPE: both down hard, no catalyst parsed to schema, so no card. Flip on a confirmed earnings or guidance document.
One honesty line on correlation. The Snowflake long and the Broadcom pass sit on the same AI capex story. They point opposite ways only because one raised and one trimmed. If that story cracks, both go the same direction.
The automated side, which is a small automated test account, entirely separate from my personal investing, holds about 101 dollars and is flat with nothing open. A 377 dollar share position does not fit, and its rules keep it to single leg long options only, so today it takes nothing. Behind pace, and reported as such.
Today on the calendar, all times ET
The week has one event that matters and it is not today:
8:30 a.m. weekly jobless claims.
10:00 a.m. the services activity survey.
After the close: LULU, ZS, DOCU, PATH, GWRE, AMBA and IOT.
Friday 8:30 a.m. the August employment report. That is the week.
Monday: US markets closed for Labor Day.
Thursday the 10th, after the close: Oracle and Adobe.
I left this month’s inflation release date off deliberately. I could not verify it this morning, and a date I have not checked is worse than a gap.
Why the desk publishes a setup it might never take
A card is not a trade. It is a written claim with a fill rule, a stop, a target, a horizon and a falsifier, filed before the open so it can be graded later whether or not any money moves.
The version of me that trades off a feeling never has to be wrong, because nothing was written down. NOT TRIGGERED is a real grade here, with its own tally next to the wins and the losses.
That is the same discipline behind the free Claude stock research agent I published, which logs every price call it hears with a probability and a horizon, so an analyst moves up or down on evidence instead of on how convincing they sounded that week.
It also runs unattended, which is its own problem. I wrote up what I learned the hard way about what a Claude scheduled task actually does when it wakes up late. This edition is an example of the same category of failure, and the next section says how.
The report card
This is where yesterday’s setups get graded, and today I owe you a gap instead of grades.
No resolution rows were written for the claims filed on September 2. The register shows them still open with horizons running out to September 16 and 17, and the grading run that resolves them fires Monday. I am not going to hand grade them here, because a writer marking his own homework is exactly what the register exists to prevent.
The second thing I owe you. This edition shipped without its four painted data boards. The render engine lives in a folder on my Mac that this run could not reach, so the boards could not be built at all. A text edition on time beats a painted one that is late, so it went out plain, and the failure gets named rather than papered over.
Where this leaves me
An AMBER tape, one crowded card that only passes its own math at the long end of its horizon, and a jobs report on Friday that can make all of it moot. The macro condition that would change my read is simple. A hot enough employment print to put a rate hike back on the table, which is what one of my tracked sources has argued for two days while the bond tape quietly disagrees with him.
If you want something like this running on a schedule for your own business, that is what I do for a living. Consulting is $250 for a 60-minute expert working session. Share your project here.
One question I keep chewing on. When a system tells you its own setup fails at five sessions and passes at ten, is that honesty, or is it the setup talking me into a longer leash? I do not know yet, and the grades will settle it.
Again, not financial advice, and not a recommendation to buy or sell anything. Education and entertainment only. I publish my losses too.
Sources
Opened and read this morning:
Price and index levels are my own broker fetches, timestamped in the tape section. Only sources I actually opened this morning are listed.
Comments