Every forecast, graded
Before every open we commit each symbol to a projected range, an expected-volatility call and touch odds on each gamma level. After the close we grade it against what actually happened. Both halves are timestamped and published - the commitment before the market can settle it, the receipt after. We never forecast direction. This page is the whole record, misses included.
The SPX projected range has contained the day in 81 of 94 graded sessions (86%) against an 80% target - which means the band is currently wider than it needs to be, not that the forecast is better than advertised.
5 of those 13 misses fell in 2 runs of back-to-back sessions, so they are volatility events rather than independent days. Confidence intervals assume independence, which makes the ones below narrower than the truth.
Range coverage, all graded sessions
The projected range is the one claim with a published target, so it is the one that can be scored against something other than itself. The target is a coverage rate, not an accuracy score: a band advertised to contain the day 80% of the time should contain it about that often - no more. Sitting well above target means the band is wider than advertised and is carrying less information, which is why it is reported as a fault below rather than a win.
- SPY84%75-90%79 of 94 held · target 80% - on target - the band is containing the day about as often as it claims to
- QQQ80%71-87%75 of 94 held · target 80% - on target - the band is containing the day about as often as it claims to
- SPX86%78-92%81 of 94 held · target 80% - above target - the band is holding more often than advertised, which means it is currently wider than it needs to be
- NDX31%22-42%25 of 80 held · target 80% - below target - the band is too narrow on this sample and is missing more days than it should
- ES86%78-92%81 of 94 held · target 75% - above target - the band is holding more often than advertised, which means it is currently wider than it needs to be
- NQ31%22-42%25 of 80 held · target 75% - below target - the band is too narrow on this sample and is missing more days than it should
- NVDA—0 graded · fewer than 5 graded sessions - no rate published
- MSFT—0 graded · fewer than 5 graded sessions - no rate published
| Symbol | Graded | Held | Coverage | 95% interval | Against target |
|---|---|---|---|---|---|
| SPY | 94 | 79 | 84% | 75-90% | 80% - on target - the band is containing the day about as often as it claims to |
| QQQ | 94 | 75 | 80% | 71-87% | 80% - on target - the band is containing the day about as often as it claims to |
| SPX | 94 | 81 | 86% | 78-92% | 80% - above target - the band is holding more often than advertised, which means it is currently wider than it needs to be |
| NDX | 80 | 25 | 31% | 22-42% | 80% - below target - the band is too narrow on this sample and is missing more days than it should |
| ES | 94 | 81 | 86% | 78-92% | 75% - above target - the band is holding more often than advertised, which means it is currently wider than it needs to be |
| NQ | 80 | 25 | 31% | 22-42% | 75% - below target - the band is too narrow on this sample and is missing more days than it should |
| NVDA | 0 | — | — | — | fewer than 5 graded sessions - no rate published |
| MSFT | 0 | — | — | — | fewer than 5 graded sessions - no rate published |
Intervals are 95% Wilson across 536 graded sessions in total. They are wide because the samples are small, and a point estimate would hide that. A rate is only published once a symbol has 5 graded sessions; below that a rate can only read 0% or 100% and means nothing.
The days we missed
Every session where the day traded outside the band we published that morning. They are listed because a track record without its failures is advertising, and because each one links to the original commitment and its receipt - you can check that we are not describing them charitably.
- Nov 12, 2026
- Nov 11, 2026
- Nov 9, 2026
- Nov 5, 2026
- Oct 30, 2026
- Oct 16, 2026
- Oct 14, 2026
- Sep 21, 2026
- Aug 13, 2026
- Aug 4, 2026
- Aug 3, 2026
- Jul 29, 2026
- Jul 27, 2026
- Jul 23, 2026
- Jul 7, 2026
4 of these arrived in 2 runs of back-to-back sessions - the band fails on consecutive days, not at random.
- Nov 13, 2026
- Nov 9, 2026
- Nov 5, 2026
- Oct 22, 2026
- Oct 16, 2026
- Oct 14, 2026
- Oct 8, 2026
- Sep 21, 2026
- Aug 13, 2026
- Aug 4, 2026
- Aug 3, 2026
- Jul 30, 2026
- Jul 29, 2026
- Jul 27, 2026
- Jul 24, 2026
- Jul 23, 2026
- Jul 16, 2026
- Jul 14, 2026
- Jul 7, 2026
7 of these arrived in 3 runs of back-to-back sessions - the band fails on consecutive days, not at random.
- Nov 5, 2026
- Oct 30, 2026
- Oct 16, 2026
- Oct 14, 2026
- Oct 5, 2026
- Sep 23, 2026
- Aug 4, 2026
- Aug 3, 2026
- Jul 23, 2026
- Jul 15, 2026
- Jul 8, 2026
- Jul 7, 2026
- Jul 6, 2026
5 of these arrived in 2 runs of back-to-back sessions - the band fails on consecutive days, not at random.
- Nov 5, 2026
- Oct 30, 2026
- Oct 16, 2026
- Oct 15, 2026
- Oct 14, 2026
- Oct 13, 2026
- Oct 12, 2026
- Oct 9, 2026
- Oct 8, 2026
- Oct 7, 2026
- Oct 6, 2026
- Oct 5, 2026
- Oct 2, 2026
- Oct 1, 2026
- Sep 30, 2026
- Sep 29, 2026
- Sep 28, 2026
- Sep 25, 2026
- Sep 24, 2026
- Sep 22, 2026
- Sep 21, 2026
- Sep 18, 2026
- Sep 17, 2026
- Sep 14, 2026
- Sep 11, 2026
- Sep 10, 2026
- Sep 9, 2026
- Sep 8, 2026
- Sep 4, 2026
- Sep 3, 2026
- Sep 1, 2026
- Aug 31, 2026
- Aug 28, 2026
- Aug 27, 2026
- Aug 26, 2026
- Aug 24, 2026
- Aug 20, 2026
- Aug 19, 2026
- Aug 18, 2026
- Aug 17, 2026
- Aug 14, 2026
- Aug 13, 2026
- Aug 12, 2026
- Aug 11, 2026
- Aug 10, 2026
- Aug 7, 2026
- Aug 6, 2026
- Aug 5, 2026
- Aug 4, 2026
- Aug 3, 2026
- Jul 30, 2026
- Jul 29, 2026
- Jul 28, 2026
- Jul 27, 2026
- Jul 24, 2026
52 of these arrived in 6 runs of back-to-back sessions - the band fails on consecutive days, not at random.
- Nov 5, 2026
- Oct 30, 2026
- Oct 16, 2026
- Oct 14, 2026
- Oct 5, 2026
- Sep 23, 2026
- Aug 4, 2026
- Aug 3, 2026
- Jul 23, 2026
- Jul 15, 2026
- Jul 8, 2026
- Jul 7, 2026
- Jul 6, 2026
5 of these arrived in 2 runs of back-to-back sessions - the band fails on consecutive days, not at random.
- Nov 5, 2026
- Oct 30, 2026
- Oct 16, 2026
- Oct 15, 2026
- Oct 14, 2026
- Oct 13, 2026
- Oct 12, 2026
- Oct 9, 2026
- Oct 8, 2026
- Oct 7, 2026
- Oct 6, 2026
- Oct 5, 2026
- Oct 2, 2026
- Oct 1, 2026
- Sep 30, 2026
- Sep 29, 2026
- Sep 28, 2026
- Sep 25, 2026
- Sep 24, 2026
- Sep 22, 2026
- Sep 21, 2026
- Sep 18, 2026
- Sep 17, 2026
- Sep 14, 2026
- Sep 11, 2026
- Sep 10, 2026
- Sep 9, 2026
- Sep 8, 2026
- Sep 4, 2026
- Sep 3, 2026
- Sep 1, 2026
- Aug 31, 2026
- Aug 28, 2026
- Aug 27, 2026
- Aug 26, 2026
- Aug 24, 2026
- Aug 20, 2026
- Aug 19, 2026
- Aug 18, 2026
- Aug 17, 2026
- Aug 14, 2026
- Aug 13, 2026
- Aug 12, 2026
- Aug 11, 2026
- Aug 10, 2026
- Aug 7, 2026
- Aug 6, 2026
- Aug 5, 2026
- Aug 4, 2026
- Aug 3, 2026
- Jul 30, 2026
- Jul 29, 2026
- Jul 28, 2026
- Jul 27, 2026
- Jul 24, 2026
52 of these arrived in 6 runs of back-to-back sessions - the band fails on consecutive days, not at random.
Calibration of the other two claims
The range is scored above, over the whole record. These two cannot be: the touch-odds Brier score and the volatility call are computed over a rolling 30-session window, which is the only place they exist. Deliberately no range number here - every claim is reported once, from the deepest sample that can score it, and a rolling window flatters a record whose misses are all older than the window.
- Touch-odds Brier(30)
- 0.161
- Vol call
- 43%vs 43%
Brier better calibrated than guessing - 0.25 is what you score by guessing the base rate every time, and lower is better.
Vol call no better than answering "expansion" every day - on this sample the call is not adding anything.
- Touch-odds Brier(30)
- 0.134
- Vol call
- 43%vs 47%
Brier better calibrated than guessing - 0.25 is what you score by guessing the base rate every time, and lower is better.
Vol call WORSE than answering "expansion" every day. On this sample the volatility call is subtracting value, and we would rather say that than leave you to notice it.
- Touch-odds Brier(30)
- 0.153
- Vol call
- 20%vs 43%
Brier better calibrated than guessing - 0.25 is what you score by guessing the base rate every time, and lower is better.
Vol call WORSE than answering "expansion" every day. On this sample the volatility call is subtracting value, and we would rather say that than leave you to notice it.
- Touch-odds Brier(30)
- 0.248
- Vol call
- 17%vs 53%
Brier no better than guessing - 0.25 is what you score by guessing the base rate every time, and lower is better.
Vol call WORSE than answering "expansion" every day. On this sample the volatility call is subtracting value, and we would rather say that than leave you to notice it.
- Touch-odds Brier(30)
- 0.153
- Vol call
- 20%vs 43%
Brier better calibrated than guessing - 0.25 is what you score by guessing the base rate every time, and lower is better.
Vol call WORSE than answering "expansion" every day. On this sample the volatility call is subtracting value, and we would rather say that than leave you to notice it.
- Touch-odds Brier(30)
- 0.248
- Vol call
- 17%vs 53%
Brier no better than guessing - 0.25 is what you score by guessing the base rate every time, and lower is better.
Vol call WORSE than answering "expansion" every day. On this sample the volatility call is subtracting value, and we would rather say that than leave you to notice it.
What this record does not claim
- It is not a direction call. Nothing here predicts whether price goes up or down, and no coverage rate should be read as though it did. The range says how far, never which way.
- High coverage is not high accuracy. A band containing the day more often than an 80% target is a band that is too wide. On the current record ours is - by roughly 20% on the live range model, measured by asking how far each day’s band could have been narrowed and still contained it.
- The intervals are optimistic. Wilson intervals assume independent trials. Our misses cluster on consecutive sessions, so the effective sample is smaller than the session count and the true intervals are wider than the ones above.
- The sample is short and the model has changed. The range model has been revised twice over this history. Sessions graded under a retired version are still counted here - removing them would be grading ourselves only on the version that is working.
- Past grading is not a promise. Nothing on this page is a prediction about future sessions, and none of it is investment advice.
SPX record covers Jul 6, 2026 through Nov 13, 2026. Rebuilt every 30 minutes from the same graded receipts published under /forecast.
