Sixteen days ago, our September 9 accuracy audit ended with a tentative observation: when SPY opened against a pressure reading by about a dollar or less, price came back through the prior close every time, and when the adverse gap was $2.42 or more, it never did. We said at the time that seven sessions could not establish a threshold. This SPY overnight recap is where we find out whether that caution was warranted.
It was. The small-gap half of the pattern failed both of its new tests. Here is the full scorecard, the rows that broke the idea, and what we are keeping from the month.
SPY overnight recap: the headline number slipped from 18 of 24 to 17 of 24
The rolling scorecard on the SpyTomorrow homepage now reads 70.8% — 17 of 24 correct, covering readings graded from August 20 through September 23, 2026. On September 9 it read 75.0%. The grading rule has not changed: a reading counts as correct if SPY trades beyond the prior session’s close, in the reading’s direction, between 9:30 and 10:05 a.m. ET, scored against Polygon minute-bar aggregates.
Split the window in half and the drift is clearer. Sessions from August 20 through September 3 went 9 of 11. Sessions from September 4 through September 23 went 8 of 13. The second stretch is the one this SPY overnight recap is about.
Every reading graded since September 3
“Open vs. prior close” is where SPY’s first print landed relative to the previous session’s close. A positive number means a gap up.
| Session | Reading | Gamma regime | Open vs. prior close | Graded |
|---|---|---|---|---|
| Sep 4 | −1.39 | Long | −$1.16 | Correct |
| Sep 8 | −1.66 | Neutral | −$1.12 | Correct |
| Sep 8 (two pre-market readings) | +1.34 / +2.45 | Long | −$1.12 | Miss ×2 |
| Sep 11 | +0.14 | Short | +$6.89 | Correct |
| Sep 14 | +0.80 | Short | −$5.29 | Miss |
| Sep 15 | −0.27 | Short | −$0.76 | Correct |
| Sep 16 | +1.39 | Short | +$2.11 | Correct |
| Sep 17 | +5.21 | Short | +$9.10 | Correct |
| Sep 18 | −1.33 | Short | −$1.29 | Correct |
| Sep 21 | −0.18 | Neutral | +$4.56 | Miss |
| Sep 22 | −1.45 | Long | +$0.53 | Miss |
| Sep 23 | −4.17 | Long | −$0.59 | Correct |
Two things stand out before any analysis. Every miss in the table opened on the wrong side of the prior close. And September 21 and 22 were back-to-back misses — two down readings into two gap-up mornings — the only consecutive losing sessions in the 24-row window.
Retesting the $1.18 line
Only sessions where the open went against the reading can test the earlier idea, because gap-aligned sessions are settled at 9:30 by the grading rule itself. Four new sessions qualified. “Best retrace” is the furthest SPY moved back toward the prior close, measured on minute-bar closes inside the window.
| Session | Adverse gap | Best retrace | Prediction from Sep 9 pattern | Outcome |
|---|---|---|---|---|
| Sep 22 | $0.53 | $0.33 (61%) | Recover | Miss |
| Sep 8 (pre-market readings) | $1.12 | $0.23 (21%) | Recover | Miss |
| Sep 21 | $4.56 | $0.09 (2%) | Fail | Miss |
| Sep 14 | $5.29 | $1.56 (29%) | Fail | Miss |
The large-gap half held: both big adverse gaps stayed unrecovered. The small-gap half did not. September 22 opened just 53 cents on the wrong side and still never got back. Put the two recaps together and sessions that opened against the reading from August 7 through September 23 recovered 3 times out of 11.
A pattern found in seven observations is a question, not a rule. This month answered the question, and the answer was mostly no.
That lines up with what the broader data says about opening gaps: fill odds depend heavily on gap size, and even small gaps are far from a sure thing — see our breakdown of gap-fill rates by gap size. A 35-minute window is not much time for any gap to reverse.
Three splits worth watching, none worth trusting yet
Gamma regime
Across all 24 rows, readings taken in a short-gamma regime went 13 of 15. Long-gamma readings went 3 of 6, or 3 of 4 if you set aside the two pre-market readings. Neutral-gamma readings went 1 of 3. That is the most lopsided split in the data, and also the one built on the smallest groups. It is a reason to keep tracking the regime column, not a reason to act on it.
Readings the homepage labels neutral
The archive shows a reading as NEUTRAL when its score sits within ±0.50 or the gamma regime is neutral. Seven of the 24 graded rows carried that label, and the scorecard still graded them by the sign of the score. They went 5 of 7. Both misses — September 3 and September 21 — were neutral-labeled down readings that met gap-ups. If you skip neutral readings in your own process, your personal hit rate on this window is not the homepage number.
Which readings count
The 24 rows cover 22 trading sessions. Two extra readings logged around 5:27 a.m. ET on September 4 were graded against the September 8 open, next to that session’s regular closing reading. Both missed. Count only the regular 3:55 p.m. readings and the window scores 17 of 22, or 77.3%. Neither figure is the “real” one. They answer different questions, which is why a scorecard’s denominator deserves as much attention as its percentage.
What this SPY overnight recap is keeping from September
- The open is the verdict more often than not. In this window, 15 of 15 gap-aligned rows were correct and 2 of 9 adverse-gap rows were. Plan the adverse-open morning before you hold, not after.
- Do not pre-commit to a “small gap will come back” exit. The data offered that idea once and took it away the next time. Decide your exit on an adverse open from your own risk limits, not from a pattern in a handful of sessions.
- Log the regime and the label, not just the direction. Gamma regime and neutral status are free columns in the archive. Recording them next to your fills in a trading journal built around pre-committed rules is how you find out whether these splits hold for you.
- Read the denominator. 70.8% and 77.3% describe the same month. Know which one matches the readings you would actually trade.
Still ungraded: the −7.12 reading
The September 24 reading of −7.12 is the largest in absolute terms anywhere in the archive going back to early July, and it had not been added to the scorecard at the time of writing. September 25’s −1.71 is also pending. A big number is not a promise; the September 9 audit found that stronger readings did not lead the pack, and one session will not settle that either way. Both will be in the next SPY overnight recap, graded by the same rule, whichever way they went.
That is the point of publishing the misses. A pressure reading is decision support you write down before the close, next to a plan for the morning it is wrong. September gave us five fresh examples of that morning.
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.
