This is a SPY overnight recap of our own scorecard, not the market’s. Over the twenty trading sessions from July 24 through August 20, 2026, the SPY Tomorrow archive logged 24 published pressure readings. Eighteen of them were scored correct — 75.0%. Collapse the duplicates so each session gets one call instead of three, and the record is 16 of 20, or 80%.
Both numbers are accurate. Neither is the interesting part. Twenty sessions is a small enough sample that the headline percentage is mostly noise — two coin flips in either direction move it by ten points. What actually carries information is the shape of the misses, and in this window the misses had an unmistakable shape.
How a reading gets scored
Before the numbers mean anything, the scoring rule has to be on the table. A reading is marked correct when SPY reaches the pressure direction’s goal by roughly 10:00 ET the next morning. Outcomes are measured against Polygon SPY minute aggregates in the 09:30–10:05 ET window, using the prior session’s close as the reference point. That is a narrow, mechanical definition, and it is deliberately unforgiving: a call that was directionally right but arrived at 11:15 ET counts as a miss.
The scorecard, broken out by conviction
Grouping the twenty sessions by the absolute value of the pressure score gives this:
| Score magnitude | Sessions | Correct | Rate |
|---|---|---|---|
| Under 0.50 (neutral band) | 2 | 1 | 50% |
| 0.50 – 1.49 | 11 | 9 | 82% |
| 1.50 – 2.49 | 3 | 2 | 67% |
| 2.50 and above | 4 | 4 | 100% |
| All sessions | 20 | 16 | 80% |
The tempting read is “high scores are perfect, trade those.” Resist it. Four sessions is not evidence of anything. A single miss in that bucket next month drops it from 100% to 80%, and the difference between the 1.50–2.49 bucket and the 0.50–1.49 bucket is one session. What the table honestly supports is narrower: readings inside the neutral band did not carry directional information, which is exactly what the neutral band is supposed to communicate.
Finding 1: every miss was a down call
This is the finding worth the post. Split the twenty sessions by the sign of the reading:
- Positive (up) readings: 9 sessions, 9 correct.
- Negative (down) readings: 11 sessions, 7 correct — 64%.
All four misses in the window were bearish calls that the market gapped straight through: July 27, August 3, August 7, and August 12. Every one of them opened higher than the prior close, by 0.32% to 0.81%.
The context is not subtle. SPY closed at 738.18 on July 23 and at 769.06 on August 19 — a gain of roughly 4.2% over the same stretch these readings cover, with the broader tape printing record highs on cooling inflation and strong AI-related earnings. The model was not randomly wrong. It was wrong in one direction, repeatedly, while the tape ground higher.
What that does and doesn’t tell you
It does not tell you down calls are worthless — seven of eleven still hit, including the two largest-magnitude bearish readings in the window. It tells you that in a persistent uptrend, a counter-trend signal is fighting something the signal itself does not measure. Pressure readings describe late-day conditions in SPY. They do not describe a four-week drift. When those two disagree, this window says the drift won more often than not.
A model that is 80% right overall but 100%/64% split by direction is not one model. It is two, and only one of them was working.
Finding 2: twenty-four readings is not twenty-four sessions
The rolling archive publishes readings, and some sessions get more than one. August 12 received three separate readings — an early-morning update, a midday revision, and the 15:55 ET print. All three were bearish. All three were wrong. That single session contributed half of the six raw misses. August 19 got three readings as well, and all three were correct.
The effect is that the raw 18-of-24 number double- and triple-weights two sessions. Deduplicated to one call per session, the record improves to 16 of 20. We are reporting both because the honest version of a hit rate includes the denominator you chose and why. Intraday revisions are updates to the same forecast, not independent observations, and treating them as independent inflates or deflates the number depending on which way the session went.
Finding 3: the misses moved more
Median absolute gap on the four missed sessions was 0.43%. On the sixteen correct sessions it was 0.25%. With four data points on one side of that comparison, this is an observation and not a statistic — but it fits a pattern worth watching. The sessions that broke the read were the sessions that moved hardest, which is also when position size does the most damage. If you want the arithmetic on why that combination is the expensive one, we walked through it in why your stop loss isn’t your real overnight risk number.
Four things this changes going forward
- Log the regime alongside the reading. A down call in week four of a grinding uptrend is a different animal from a down call after a distribution day. The archive does not currently capture that, and this window is the argument for adding it.
- Score sessions, not prints. One session, one outcome. Publish the raw count too, but lead with the deduplicated one.
- Size before the reading arrives. Deciding your risk after you see a strong number is how a good read turns into a bad trade. Pre-commitment is the whole point.
- Publish the losing streaks. Three consecutive wrong prints on August 12 is the kind of detail that gets quietly dropped from a marketing page. It belongs in the record.
Where this SPY overnight recap fits
The last time we ran this exercise, the same rolling archive showed a rougher stretch — 14 of 24 correct, with three findings that mattered. The window has improved since. That improvement is real, and it is also four weeks long, which is another way of saying it is not yet evidence of much. Anyone who tells you a month proves a model is selling something.
The way to use a pressure reading is as one input into a decision you have already structured — a filter that tells you whether to take a setup you were prepared to take anyway, sized in advance, with an exit written down. It is not an instruction, and a 20-session sample is not a promise. If you are still calibrating what an overnight move typically looks like, our work on which SPY gap-fill statistics actually hold up is a better starting point than any single night’s number.
The next audit will run against a fresh window, and if the down calls keep underperforming while the tape climbs, that will be worth saying out loud too.
This article is for educational and informational purposes only. It does not constitute financial advice, investment recommendations, or a solicitation to buy or sell any securities. Past performance and historical data do not guarantee future results. All trading involves risk, including the potential loss of principal. Always do your own research and consult with a qualified financial advisor before making trading decisions.
