The scorecard on the SpyTomorrow front page currently reads 18 of 24 correct — 75.0%. The archive feed behind it publishes full row-level detail for 20 of those sessions, running from August 7 through September 3, 2026, and those 20 graded out at 16 correct, or 80%.
Both numbers look good. Neither one tells you much on its own, and this post is about why. When you open the 20 sessions and grade them one at a time, 13 of the 16 wins were already settled the instant the opening bell rang. The model’s actual test was seven sessions long, and in those seven it went 3-for-7.
First, what “correct” actually means here
You cannot audit a hit rate until you know the grading rule, and most published trading accuracy claims never state theirs. This one is specific and checkable. A pressure reading is taken near the close, around 3:55 p.m. ET. It is graded the following morning, and it counts as correct if SPY trades beyond the prior session’s closing price, in the direction of the reading, at any point between 9:30 and 10:05 a.m. ET. Outcomes are scored against Polygon minute-bar aggregates, not a subjective read of the tape.
Read that carefully, because a consequence falls straight out of it. If the market gaps open in the same direction as the reading, price is already past the prior close before a single minute of trading has elapsed. The session is graded correct at 9:30:00. Nothing the model said had to be tested at all.
Splitting 20 sessions into decided and contested
- Gap opened with the reading — 13 sessions, 13 correct. A perfect record that contains zero information, because the grading rule was satisfied by the opening print.
- Gap opened against the reading — 7 sessions, 3 correct. These are the only sessions where the market had to come back through the prior close for the reading to be right.
- All four misses were down readings that opened higher. Down readings went 7-for-11 overall; up readings went 9-for-9.
The overall 16-for-20 is not fake. It is just heavily weighted toward sessions where the overnight tape had already agreed with the reading by the time grading started. That distinction is worth holding onto, because a trader entering at the close is exposed to the whole overnight move, not just the 35 minutes the scorecard measures — which is the same reason the overnight futures print is not your gap.
The seven sessions that actually tested the model
Here are all seven contested sessions — every day in the window where SPY opened against the pressure reading. “Adverse gap” is how far the open landed on the wrong side of the prior close. “Recovered” is the best move back toward it during the 9:30–10:05 window.
| Date | Reading | Adverse gap | Recovered | Graded |
|---|---|---|---|---|
| Aug 17 | +0.831 | $0.16 | $0.46 | Correct |
| Aug 28 | −2.326 | $0.66 | $0.68 | Correct |
| Aug 26 | +2.588 | $1.18 | $2.16 | Correct |
| Aug 27 | −3.193 | $2.42 | $1.02 (42%) | Miss |
| Aug 7 | −2.121 | $2.46 | $0.59 (24%) | Miss |
| Sep 3 | −2.548 | $2.74 | $0.00 (0%) | Miss |
| Aug 12 | −1.352 | $4.15 | $2.77 (67%) | Miss |
Sorted that way, the split is hard to miss. Every adverse gap of $1.18 or less was recovered. Every adverse gap of $2.42 or more was not. There is nothing in between in this sample, which is both the interesting part and the reason to be careful with it — seven observations do not establish a threshold, and the absence of a middle case is a gap in the data, not evidence of a cliff.
Reading strength was not the deciding variable
The intuitive assumption is that a bigger absolute reading should be a better reading. Across all 20 graded sessions, it was not:
- Readings under 1.00 in absolute value: 7 of 7 correct
- Readings from 1.00 to 2.49: 4 of 6 correct
- Readings of 2.50 or higher: 5 of 7 correct
The quietest readings — the near-neutral ones a trader is most tempted to skip — had the cleanest month, and the strongest ones did not lead. Sample sizes of six and seven are far too small to call that a pattern, and one more session could flip any of those buckets. It is enough to say the obvious relationship did not show up, and that “the number was big” is not by itself a reason to size up.
Where the scorecard and your P&L disagree
August 12 is the session worth studying. The reading was −1.352. SPY gapped $4.15 higher, then sold off $2.77 in the first 35 minutes — two-thirds of the way back. The scorecard says miss, because price never touched the prior close. A trader who was short from the open banked a decent morning. A trader who held from the previous close was still deeply underwater.
Three people, one session, three different outcomes from the same data. Binary grading has to throw that away, and any honest reading of a hit rate has to remember it was thrown away. The same caution applies to the tidy round numbers in most published SPY gap statistics: the definition doing the work is usually buried, and it usually matters more than the percentage.
A hit rate is a description of a grading rule applied to a sample. It is not a description of what you would have made.
What this SPY prediction accuracy audit does not establish
- It is one month. Twenty graded sessions across a single volatility regime. The previous 20-reading recap found its own pattern in the misses, and patterns that small routinely dissolve in the next window.
- The down-reading skew may be regime, not model. SPY spent most of this stretch grinding between roughly 762 and 778. In a tape that keeps recovering, down calls are the ones that get punished.
- No slippage, spread, or exit skill is modeled. “Reached the goal by 10:05” says nothing about whether you were awake, positioned, or able to get out at the number.
How to read a scorecard like this one
The useful takeaway is procedural, not predictive. When you evaluate any published accuracy figure — this one included — do three things before you act on it:
- Find the grading rule and check what it settles for free. Here, gap-aligned sessions are decided at 9:30 by construction. That is 65% of this sample.
- Isolate the contested cases. A model is only informative where it disagreed with the tape. Three-for-seven is a much more honest starting point than 80%.
- Decide in advance what an adverse open means for you. In this window the adverse gap size, not the reading strength, is what separated the saves from the misses — so that is the variable to pre-commit around, not the headline number.
That is the honest role for a pressure reading: decision support that gets written down before the open, alongside a plan for the morning where the gap runs against it. It is not a signal to follow blindly, and a 75% or 80% counter on a homepage is not a reason to skip the plan. The seven sessions above are the ones that would have needed it.
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.
