AI tournament · closed
Five AIs forecast the market every morning for six weeks. None did better than chance.
Haiku 4.5, Sonnet 5 and Opus 5 (and, from 25 September, one model from DeepSeek and one from Alibaba) received the same data package on 10 S&P 500 names every session from 3 September to 16 October 2026 and returned, before the open, the probability that each one would close up and how much it would move. At the close we scored them with Brier and CRPS and published the result whatever it said.
Tournament closed on Oct 18. It ended with 315 resolved predictions per model, more than the 200 pre-registered before claiming anything. No model forecast direction better than always saying the base rate and, with enough data, none sized the move better than a mechanical forecast based on volatility. Last session: Oct 16. What you see is the final state.
What is measured
Calibration, not hit rate: Brier and its decomposition (resolution) for direction; a three-quantile CRPS and the coverage of the 80% band for the size of the move.
What we expect
The pre-registered expectation is that no AI discriminates one-day direction: in this universe it is measured as null. Where value may exist is in how much a name moves, not which way.
What it is not
Not a recommendation or a signal. No model sees headlines or searches the web; only prices, volatility, catalyst news counts, the main macro calendar (Fed, CPI, jobs) and, in variant B, quarterly-results context.
Backtest results (simulation), not real trading. Not investment advice. Past performance does not guarantee future results.
Reading
Sentences generated by fixed rules from the numbers, using the pre-registered thresholds. Nobody writes them by hand.
- Tournament closed on Oct 18 with 315 resolved predictions per model (pre-registered threshold: 200). What follows is the final state.
- Direction: no model beats the base rate; in this sample, always saying the average rate is the best forecast (which is what was pre-registered).
- Power: with 315 predictions per model, only a correlation of 0.16 or more between probability and outcome would be detectable (80% power, p<0.05); a small effect, of 0.02 to 0.05, needs between 2,500 and 20,000. That nobody discriminates today does not prove it cannot be done.
- Magnitude Haiku 4.5 A: behind the mechanical baseline (1.58 % vs 1.11 %); plain volatility predicts size better.
- Magnitude Haiku 4.5 B: behind the mechanical baseline (2.15 % vs 1.18 %); plain volatility predicts size better.
- Magnitude Haiku 4.5 E: ahead of the mechanical baseline (1.11 % vs 1.50 %), not yet significant.
- Magnitude Haiku 4.5 R: behind the mechanical baseline (2.01 % vs 1.34 %); plain volatility predicts size better.
- Magnitude Opus 5 A: on par with the mechanical baseline (1.15 % vs 1.10 %).
- Magnitude Opus 5 B: behind the mechanical baseline (1.25 % vs 1.17 %); plain volatility predicts size better.
- Magnitude Opus 5 E: ahead of the mechanical baseline (1.22 % vs 1.50 %), not yet significant.
- Magnitude Opus 5 R: behind the mechanical baseline (1.43 % vs 1.34 %); plain volatility predicts size better.
- Magnitude Sonnet 5 A: behind the mechanical baseline (1.22 % vs 1.11 %); plain volatility predicts size better.
- Magnitude Sonnet 5 B: behind the mechanical baseline (1.39 % vs 1.18 %); plain volatility predicts size better.
- Magnitude Sonnet 5 E: ahead of the mechanical baseline (1.36 % vs 1.50 %), not yet significant.
- Magnitude Sonnet 5 R: behind the mechanical baseline (1.43 % vs 1.34 %); plain volatility predicts size better.
- Magnitude deepseek-v4.1-flash A: behind the mechanical baseline (1.16 % vs 1.04 %); plain volatility predicts size better.
- Magnitude deepseek-v4.1-flash B: behind the mechanical baseline (1.38 % vs 1.22 %); plain volatility predicts size better.
- Magnitude deepseek-v4.1-flash E: ahead of the mechanical baseline (1.11 % vs 1.50 %), not yet significant.
- Magnitude deepseek-v4.1-flash R: behind the mechanical baseline (1.89 % vs 1.70 %); plain volatility predicts size better.
- Magnitude Ensemble A: behind the mechanical baseline (1.14 % vs 1.04 %); plain volatility predicts size better.
- Magnitude Ensemble B: behind the mechanical baseline (1.37 % vs 1.22 %); plain volatility predicts size better.
- Magnitude Ensemble E: ahead of the mechanical baseline (1.22 % vs 1.50 %), not yet significant.
- Magnitude Ensemble R: behind the mechanical baseline (1.79 % vs 1.70 %); plain volatility predicts size better.
- Magnitude qwen3.8-flash A: behind the mechanical baseline (1.10 % vs 0.89 %); plain volatility predicts size better.
- Magnitude qwen3.8-flash B: behind the mechanical baseline (1.35 % vs 1.10 %); plain volatility predicts size better.
- Magnitude qwen3.8-flash E: ahead of the mechanical baseline (1.21 % vs 1.50 %), not yet significant.
- Magnitude qwen3.8-flash R: behind the mechanical baseline (0.99 % vs 0.93 %); plain volatility predicts size better.
- Paper P&L: no leader yet. The book is ranked by Sharpe, which does not depend on how much is staked, and that needs 20 sessions; there are 15. The return of a book that sits almost entirely in cash cannot be compared with a basket that is 100% invested.
- A/B Haiku 4.5: 60 pairs (12 with an event, out of 200); with results context the CRPS worsens by 24%.
- A/B Opus 5: 60 pairs (12 with an event, out of 200); with results context the CRPS improves by 7%.
- A/B Sonnet 5: 60 pairs (12 with an event, out of 200); with results context the CRPS worsens by 7%.
- A/B deepseek-v4.1-flash: 60 pairs (12 with an event, out of 200); with results context the CRPS worsens by 1%.
- A/B Ensemble: 60 pairs (12 with an event, out of 200); with results context the CRPS improves by 0%.
- A/B qwen3.8-flash: 60 pairs (12 with an event, out of 200); with results context the CRPS worsens by 4%.
Updated 2026-10-18 09:18 UTC
Paper P&L
Fixed, pre-registered rule: each name is held long or short in proportion to how far the probability sits from 0.50, equal capital per name, one-session holding, close to close, 10 bp round-trip cost. The basket is the same list of names bought equal-weight. 🏆 = best Sharpe among models with the same sessions; returns are not compared across models or against the basket, because each book has a different exposure and the basket is 100% invested. Below 250 sessions the leader is noise.
| Model | Var. | Sessions | Net return | bp/day | Sharpe | Max drawdown | Basket | vs basket | p | |
|---|---|---|---|---|---|---|---|---|---|---|
| Haiku 4.5 | E | 1 | 0.11 % | 11.1 | – | 0.0 % | 1.78 % | -1.67 % | – | |
| deepseek-v4.1-flash | A | 16 | 0.05 % | 0.3 | – | -0.1 % | -2.86 % | 2.91 % | 0.61 | |
| deepseek-v4.1-flash | R | 3 | 0.05 % | 1.6 | – | 0.0 % | -2.96 % | 3.01 % | – | |
| Opus 5 | A | 30 | 0.05 % | 0.2 | 1.48 | -0.0 % | -6.54 % | 6.58 % | 0.35 | |
| Sonnet 5 | R | 5 | 0.05 % | 0.9 | – | -0.0 % | -3.78 % | 3.82 % | 0.07 | |
| Ensemble | A | 16 | 0.04 % | 0.2 | – | -0.1 % | -2.86 % | 2.90 % | 0.61 | |
| deepseek-v4.1-flash | B | 15 | 0.03 % | 0.2 | – | -0.1 % | -0.91 % | 0.94 % | 0.85 | |
| Ensemble | R | 3 | 0.03 % | 1.1 | – | 0.0 % | -2.96 % | 2.99 % | – | |
| Sonnet 5 | B | 29 | 0.03 % | 0.1 | 0.79 | -0.1 % | -5.21 % | 5.24 % | 0.43 | |
| Opus 5 | B | 28 | 0.03 % | 0.1 | 1.00 | -0.1 % | -6.78 % | 6.81 % | 0.34 | |
| qwen3.8-flash | R | 2 | 0.02 % | 1.2 | – | 0.0 % | -1.63 % | 1.65 % | – | |
| qwen3.8-flash | X | 2 | 0.02 % | 1.0 | – | 0.0 % | -1.16 % | 1.18 % | – | |
| Haiku 4.5 | X | 6 | 0.01 % | 0.2 | – | -0.0 % | -2.10 % | 2.11 % | 0.27 | |
| Ensemble | B | 15 | 0.01 % | 0.0 | – | -0.1 % | -0.91 % | 0.92 % | 0.85 | |
| Sonnet 5 | A | 31 | 0.00 % | 0.0 | 0.03 | -0.1 % | -4.97 % | 4.97 % | 0.48 | |
| Sonnet 5 | E | 1 | 0.00 % | 0.0 | – | 0.0 % | 1.78 % | -1.78 % | – | |
| qwen3.8-flash | E | 1 | 0.00 % | 0.0 | – | 0.0 % | 1.78 % | -1.78 % | – | |
| Ensemble | E | 1 | -0.00 % | -0.2 | – | -0.0 % | 1.78 % | -1.79 % | – | |
| Opus 5 | R | 5 | -0.00 % | -0.1 | – | -0.0 % | -3.78 % | 3.77 % | 0.07 | |
| Sonnet 5 | X | 5 | -0.01 % | -0.1 | – | -0.0 % | -2.10 % | 2.10 % | 0.28 | |
| Ensemble | X | 3 | -0.01 % | -0.3 | – | -0.0 % | 0.08 % | -0.08 % | – | |
| Opus 5 | E | 1 | -0.01 % | -1.3 | – | -0.0 % | 1.78 % | -1.80 % | – | |
| qwen3.8-flash | A | 15 | -0.02 % | -0.1 | – | -0.1 % | -1.53 % | 1.51 % | 0.77 | |
| Opus 5 | X | 6 | -0.02 % | -0.3 | – | -0.0 % | -2.10 % | 2.08 % | 0.27 | |
| deepseek-v4.1-flash | X | 3 | -0.02 % | -0.6 | – | -0.0 % | 0.08 % | -0.09 % | – | |
| deepseek-v4.1-flash | E | 1 | -0.04 % | -4.0 | – | -0.0 % | 1.78 % | -1.82 % | – | |
| qwen3.8-flash | B | 12 | -0.04 % | -0.3 | – | -0.1 % | -0.46 % | 0.42 % | 0.93 | |
| Haiku 4.5 | R | 5 | -0.06 % | -1.2 | – | -0.1 % | -3.78 % | 3.72 % | 0.08 | |
| Haiku 4.5 | A | 31 | -0.09 % | -0.3 | -0.66 | -0.4 % | -4.97 % | 4.88 % | 0.48 | |
| Haiku 4.5 | B | 29 | -0.10 % | -0.3 | -3.25 | -0.1 % | -5.21 % | 5.11 % | 0.44 |
This is a paper book measured close to close: a real execution would enter at the open, and open prices are not available. None of this is a recommendation.
One-session totals
A = without results context, B = with it, X = control with the context shuffled across names. Bets are the names where the model moved away from 0.50.
| Model | Var. | n | Hits/bets | Brier | Resolution | vs base rate | CRPS | Baseline | 80% coverage |
|---|---|---|---|---|---|---|---|---|---|
| Haiku 4.5 | A | 315 | 119/252 | 0.2517 | 0.0034 | -0.0017 | 1.58 % | 1.11 % | 0.54 |
| Haiku 4.5 | X | 60 | 9/21 | 0.2504 | 0.0024 | -0.0007 | 1.81 % | 0.84 % | 0.40 |
| Haiku 4.5 | B | 290 | 73/167 | 0.2505 | 0.0021 | -0.0005 | 2.15 % | 1.18 % | 0.46 |
| Haiku 4.5 | E | 10 | 5/10 | 0.2438 | 0.0171 | -0.0838 | 1.11 % | 1.50 % | 0.60 |
| Haiku 4.5 | R | 50 | 14/27 | 0.2510 | 0.0020 | -0.0074 | 2.01 % | 1.34 % | 0.42 |
| Opus 5 | A | 305 | 30/59 | 0.2500 | 0.0000 | -0.0000 | 1.15 % | 1.10 % | 0.68 |
| Opus 5 | X | 60 | 6/13 | 0.2502 | 0.0053 | -0.0005 | 0.91 % | 0.84 % | 0.72 |
| Opus 5 | B | 280 | 43/82 | 0.2497 | 0.0005 | 0.0003 | 1.25 % | 1.17 % | 0.68 |
| Opus 5 | E | 10 | 0/1 | 0.2510 | 0.0000 | -0.0910 | 1.22 % | 1.50 % | 0.80 |
| Opus 5 | R | 50 | 3/5 | 0.2500 | 0.0039 | -0.0064 | 1.43 % | 1.34 % | 0.58 |
| Sonnet 5 | A | 315 | 74/140 | 0.2495 | 0.0006 | 0.0004 | 1.22 % | 1.11 % | 0.74 |
| Sonnet 5 | X | 50 | 12/24 | 0.2495 | 0.0054 | -0.0011 | 0.91 % | 0.68 % | 0.74 |
| Sonnet 5 | B | 290 | 114/225 | 0.2498 | 0.0000 | 0.0002 | 1.39 % | 1.18 % | 0.69 |
| Sonnet 5 | E | 10 | 0.50 everywhere | 0.2500 | 0.0000 | -0.0900 | 1.36 % | 1.50 % | 0.60 |
| Sonnet 5 | R | 50 | 12/19 | 0.2481 | 0.0015 | -0.0045 | 1.43 % | 1.34 % | 0.66 |
| deepseek-v4.1-flash | A | 160 | 58/115 | 0.2502 | 0.0014 | -0.0006 | 1.16 % | 1.04 % | 0.74 |
| deepseek-v4.1-flash | X | 30 | 5/12 | 0.2504 | 0.0171 | -0.0015 | 0.95 % | 0.65 % | 0.70 |
| deepseek-v4.1-flash | B | 150 | 51/97 | 0.2492 | 0.0002 | -0.0008 | 1.38 % | 1.22 % | 0.67 |
| deepseek-v4.1-flash | E | 10 | 2/10 | 0.2561 | 0.0000 | -0.0961 | 1.11 % | 1.50 % | 0.80 |
| deepseek-v4.1-flash | R | 30 | 18/30 | 0.2481 | 0.0000 | -0.0081 | 1.89 % | 1.70 % | 0.60 |
| Ensemble | A | 160 | 75/158 | 0.2503 | 0.0000 | -0.0007 | 1.14 % | 1.04 % | 0.73 |
| Ensemble | X | 30 | 16/29 | 0.2496 | 0.0142 | -0.0007 | 0.90 % | 0.65 % | 0.73 |
| Ensemble | B | 150 | 71/144 | 0.2495 | 0.0001 | -0.0011 | 1.37 % | 1.22 % | 0.65 |
| Ensemble | E | 10 | 4/10 | 0.2514 | 0.0100 | -0.0914 | 1.22 % | 1.50 % | 0.80 |
| Ensemble | R | 30 | 18/30 | 0.2484 | 0.0000 | -0.0084 | 1.79 % | 1.70 % | 0.57 |
| qwen3.8-flash | A | 150 | 45/103 | 0.2509 | 0.0041 | -0.0013 | 1.10 % | 0.89 % | 0.71 |
| qwen3.8-flash | X | 20 | 13/20 | 0.2466 | 0.0189 | -0.0066 | 1.07 % | 0.64 % | 0.60 |
| qwen3.8-flash | B | 120 | 53/107 | 0.2497 | 0.0006 | -0.0003 | 1.35 % | 1.10 % | 0.62 |
| qwen3.8-flash | E | 10 | 0.50 everywhere | 0.2500 | 0.0000 | -0.0900 | 1.21 % | 1.50 % | 0.70 |
| qwen3.8-flash | R | 20 | 13/20 | 0.2471 | 0.0000 | -0.0196 | 0.99 % | 0.93 % | 0.65 |
Resolution > 0.02 = the probabilities discriminate. vs base rate > 0 = beats always saying the average rate. CRPS: lower is better; the baseline is a mechanical forecast from volatility, no AI. Target coverage: 0.80.
A/B of the quarterly-results context
The same package with and without the results block (date, whether the recent move is already the reaction, the name's historical move size). CRPS difference paired by session and name.
A/B re-registered on 9 October 2026. Until then, when the calendar had no entry, the block took the first “earnings” news item as the announcement date. Checked on 8 October against the real record of earnings reports, none of the 35 news-based dates matched a report (calendar dates: 28 of 28). From now on dates come only from the calendar and the count of pairs with an event starts again from zero.
| Model | Subset | Pairs | CRPS A | CRPS B | Change | p |
|---|---|---|---|---|---|---|
| Haiku 4.5 | all | 60 | 1.47 % | 1.82 % | +24 % | 0.055 |
| Haiku 4.5 | with event | 12 | 0.99 % | 1.23 % | +24 % | 0.284 |
| Opus 5 | all | 60 | 0.96 % | 0.90 % | −7 % | 0.360 |
| Opus 5 | with event | 12 | 0.82 % | 0.85 % | +3 % | 0.742 |
| Sonnet 5 | all | 60 | 0.90 % | 0.96 % | +7 % | 0.491 |
| Sonnet 5 | with event | 12 | 0.81 % | 0.94 % | +17 % | 0.262 |
| deepseek-v4.1-flash | all | 60 | 0.92 % | 0.92 % | +1 % | 0.931 |
| deepseek-v4.1-flash | with event | 12 | 0.81 % | 1.01 % | +25 % | 0.097 |
| Ensemble | all | 60 | 0.89 % | 0.89 % | −0 % | 0.839 |
| Ensemble | with event | 12 | 0.71 % | 0.94 % | +32 % | 0.045 |
| qwen3.8-flash | all | 60 | 0.86 % | 0.89 % | +4 % | 0.523 |
| qwen3.8-flash | with event | 12 | 0.66 % | 0.92 % | +39 % | 0.194 |
Pre-registered criterion: improvement ≥ 10% with p < 0.05 and 200 pairs with an event. p is hidden with fewer than 5 sessions.
Earlier epochs (they do not count towards the criterion)
v2 (24 September) and v1 (original rule) were measured with results dates that sometimes came from news. They are kept so nothing is lost, but they are not evidence for the criterion.
| Model | Epoch | Subset | Pairs | CRPS A | CRPS B | Change | p |
|---|---|---|---|---|---|---|---|
| Haiku 4.5 | v2 | all | 110 | 1.66 % | 2.34 % | +41 % | 0.001 |
| Haiku 4.5 | v2 | with event | 38 | 1.26 % | 2.41 % | +91 % | 0.001 |
| Haiku 4.5 | v1 | all | 120 | 1.55 % | 2.15 % | +39 % | 0.001 |
| Haiku 4.5 | v1 | with event | 22 | 1.56 % | 2.13 % | +36 % | 0.132 |
| Opus 5 | v2 | all | 110 | 1.24 % | 1.44 % | +16 % | 0.064 |
| Opus 5 | v2 | with event | 38 | 0.90 % | 1.65 % | +84 % | 0.001 |
| Opus 5 | v1 | all | 110 | 1.16 % | 1.24 % | +7 % | 0.246 |
| Opus 5 | v1 | with event | 22 | 1.21 % | 1.56 % | +29 % | 0.273 |
| Sonnet 5 | v2 | all | 110 | 1.23 % | 1.62 % | +32 % | 0.001 |
| Sonnet 5 | v2 | with event | 38 | 0.94 % | 1.83 % | +95 % | 0.001 |
| Sonnet 5 | v1 | all | 120 | 1.38 % | 1.39 % | +1 % | 0.960 |
| Sonnet 5 | v1 | with event | 22 | 1.27 % | 1.64 % | +29 % | 0.446 |
| deepseek-v4.1-flash | v2 | all | 90 | 1.35 % | 1.69 % | +25 % | 0.001 |
| deepseek-v4.1-flash | v2 | with event | 33 | 1.09 % | 2.11 % | +94 % | 0.001 |
| Ensemble | v2 | all | 90 | 1.34 % | 1.69 % | +26 % | 0.001 |
| Ensemble | v2 | with event | 33 | 1.00 % | 1.95 % | +94 % | 0.001 |
| qwen3.8-flash | v2 | all | 60 | 1.40 % | 1.81 % | +29 % | 0.001 |
| qwen3.8-flash | v2 | with event | 25 | 1.19 % | 2.43 % | +104 % | 0.001 |
By session
Each cell: hits/bets · Brier · magnitude CRPS. ½ = the model said 0.50 for every name.
| Session | Closed up | |ret| median | Haiku 4.5 A | Haiku 4.5 B | Sonnet 5 A | Sonnet 5 B | Opus 5 A | Opus 5 B |
|---|---|---|---|---|---|---|---|---|
| Oct 16 | 6/10 | 1.47 % | 4/7 · 0.248 · 1.30 % | 2/2 · 0.246 · 1.74 % | ½ · 0.250 · 1.04 % | 6/9 · 0.245 · 1.00 % | ½ · 0.250 · 0.94 % | 2/3 · 0.247 · 0.97 % |
| Oct 15 | 8/10 | 2.25 % | 4/6 · 0.249 · 1.18 % | 2/10 · 0.256 · 1.02 % | 0/1 · 0.254 · 1.27 % | 2/9 · 0.255 · 1.17 % | 0/2 · 0.253 · 1.29 % | 0/6 · 0.256 · 1.47 % |
| Oct 14 | 4/10 | 1.23 % | 4/6 · 0.247 · 0.93 % | 1/1 · 0.248 · 1.59 % | ½ · 0.250 · 0.66 % | 7/10 · 0.245 · 0.73 % | 1/1 · 0.249 · 0.61 % | 3/4 · 0.248 · 0.65 % |
| Oct 13 | 3/10 | 1.18 % | 4/5 · 0.246 · 1.82 % | 2/8 · 0.257 · 2.70 % | 7/10 · 0.246 · 0.94 % | 6/10 · 0.249 · 0.91 % | 1/1 · 0.249 · 1.06 % | 6/10 · 0.249 · 0.93 % |
| Oct 12 | 6/10 | 0.53 % | 4/5 · 0.246 · 2.36 % | 4/8 · 0.251 · 2.09 % | ½ · 0.250 · 0.68 % | 4/10 · 0.252 · 1.15 % | 0/3 · 0.253 · 0.98 % | 4/8 · 0.250 · 0.62 % |
| Oct 9 | 4/10 | 2.22 % | 1/6 · 0.259 · 1.26 % | 2/5 · 0.251 · 1.80 % | ½ · 0.250 · 0.78 % | 5/8 · 0.247 · 0.78 % | 0/1 · 0.251 · 0.91 % | 1/2 · 0.249 · 0.76 % |
| Oct 8 | 4/10 | 2.11 % | 2/8 · 0.257 · 1.21 % | 2/5 · 0.252 · 1.37 % | ½ · 0.250 · 1.01 % | 4/7 · 0.250 · 1.45 % | 0/2 · 0.252 · 1.01 % | 0/2 · 0.253 · 1.13 % |
| Oct 7 | 4/10 | 1.37 % | 5/9 · 0.245 · 1.87 % | 3/6 · 0.251 · 2.04 % | ½ · 0.250 · 1.06 % | 4/7 · 0.250 · 1.47 % | 1/2 · 0.251 · 0.93 % | 2/5 · 0.252 · 1.19 % |
| Oct 6 | 4/10 | 0.85 % | 4/9 · 0.252 · 2.02 % | 1/2 · 0.250 · 2.03 % | 1/3 · 0.251 · 1.08 % | 2/5 · 0.251 · 1.69 % | 2/3 · 0.249 · 0.98 % | 1/4 · 0.252 · 1.41 % |
| Oct 5 | 6/10 | 1.03 % | 2/8 · 0.260 · 1.79 % | 4/10 · 0.250 · 3.18 % | 1/2 · 0.250 · 1.25 % | 6/10 · 0.246 · 2.32 % | 1/2 · 0.250 · 1.28 % | 4/7 · 0.249 · 2.08 % |
| Oct 2 | 6/10 | 3.79 % | 6/9 · 0.244 · 1.37 % | 3/9 · 0.253 · 1.81 % | 3/9 · 0.259 · 1.14 % | 3/8 · 0.252 · 1.43 % | ½ · 0.250 · 1.17 % | 1/2 · 0.250 · 1.40 % |
| Oct 1 | 8/10 | 1.88 % | 1/8 · 0.267 · 1.19 % | 3/7 · 0.248 · 2.27 % | ½ · 0.250 · 0.99 % | 4/8 · 0.247 · 2.09 % | 2/3 · 0.249 · 0.98 % | 3/3 · 0.244 · 1.72 % |
| Sep 30 | 8/10 | 2.27 % | 5/8 · 0.244 · 1.11 % | 2/2 · 0.246 · 1.84 % | 0/3 · 0.253 · 0.87 % | 3/7 · 0.249 · 1.20 % | ½ · 0.250 · 0.68 % | 3/5 · 0.247 · 0.83 % |
| Sep 29 | 5/10 | 0.99 % | 5/8 · 0.248 · 3.47 % | 5/10 · 0.250 · 5.73 % | 6/8 · 0.242 · 3.30 % | 5/9 · 0.248 · 3.34 % | 2/3 · 0.249 · 3.32 % | 4/5 · 0.246 · 3.38 % |
| Sep 28 | 2/10 | 2.01 % | 5/10 · 0.251 · 1.26 % | 6/7 · 0.245 · 2.46 % | 2/5 · 0.251 · 0.82 % | 7/8 · 0.244 · 0.94 % | 2/2 · 0.248 · 0.84 % | 3/4 · 0.248 · 0.83 % |
| Sep 25 | 5/10 | 3.35 % | 6/9 · 0.246 · 1.33 % | 4/8 · 0.251 · 1.65 % | 5/6 · 0.243 · 0.98 % | 3/9 · 0.253 · 0.99 % | 1/2 · 0.250 · 0.98 % | ½ · 0.250 · 1.01 % |
| Sep 24 | 5/10 | 1.05 % | 5/9 · 0.257 · 1.63 % | 4/7 · 0.249 · 1.34 % | 2/3 · 0.249 · 1.01 % | 3/7 · 0.251 · 0.88 % | 1/2 · 0.250 · 1.44 % | ½ · 0.250 · 0.87 % |
| Sep 23 | 4/10 | 2.00 % | 4/8 · 0.250 · 1.04 % | 4/7 · 0.249 · 1.58 % | ½ · 0.250 · 0.60 % | 3/6 · 0.250 · 0.62 % | ½ · 0.250 · 0.59 % | ½ · 0.250 · 0.61 % |
| Sep 22 | 7/10 | 1.28 % | 7/9 · 0.241 · 1.77 % | 0/2 · 0.254 · 2.62 % | 1/4 · 0.253 · 1.13 % | 1/9 · 0.259 · 2.12 % | 0/4 · 0.255 · 1.12 % | 0/2 · 0.254 · 1.76 % |
| Sep 21 | 6/10 | 3.07 % | 4/7 · 0.245 · 1.55 % | ½ · 0.250 · 2.87 % | 2/3 · 0.250 · 1.57 % | 4/7 · 0.249 · 1.59 % | – | – |
| Sep 18 | 5/10 | 1.67 % | 4/9 · 0.241 · 1.23 % | 5/10 · 0.242 · 2.91 % | 5/10 · 0.250 · 0.84 % | 5/8 · 0.247 · 1.24 % | 2/3 · 0.249 · 0.79 % | 1/2 · 0.249 · 0.90 % |
| Sep 17 | 9/10 | 2.13 % | 2/9 · 0.264 · 1.28 % | 2/9 · 0.254 · 1.42 % | 4/8 · 0.249 · 0.97 % | 1/7 · 0.254 · 0.94 % | ½ · 0.250 · 1.00 % | 1/1 · 0.248 · 0.96 % |
| Sep 16 | 3/10 | 1.56 % | 6/10 · 0.245 · 1.51 % | 2/4 · 0.251 · 1.32 % | 4/6 · 0.247 · 1.29 % | 7/10 · 0.246 · 0.83 % | 2/3 · 0.250 · 0.97 % | ½ · 0.250 · 0.80 % |
| Sep 15 | 2/10 | 1.58 % | 3/9 · 0.255 · 1.39 % | 3/9 · 0.255 · 1.57 % | 8/10 · 0.244 · 0.70 % | 2/7 · 0.256 · 0.65 % | 1/1 · 0.249 · 0.79 % | 1/1 · 0.249 · 0.67 % |
| Sep 14 | 4/10 | 5.30 % | 1/7 · 0.257 · 1.29 % | 3/9 · 0.253 · 1.23 % | 2/6 · 0.252 · 1.96 % | 5/7 · 0.247 · 1.69 % | ½ · 0.250 · 1.79 % | ½ · 0.250 · 2.09 % |
| Sep 11 | 5/10 | 2.32 % | 6/8 · 0.233 · 1.75 % | 2/4 · 0.250 · 1.87 % | 3/7 · 0.250 · 1.75 % | 2/5 · 0.251 · 1.65 % | ½ · 0.250 · 1.65 % | 1/3 · 0.251 · 1.68 % |
| Sep 10 | 3/10 | 2.42 % | 2/8 · 0.266 · 1.25 % | 0/3 · 0.253 · 1.92 % | 4/8 · 0.248 · 1.86 % | 4/8 · 0.250 · 1.73 % | 1/3 · 0.251 · 1.44 % | 1/1 · 0.249 · 1.27 % |
| Sep 9 | 6/10 | 1.69 % | 6/9 · 0.244 · 2.34 % | 1/1 · 0.248 · 3.55 % | 3/4 · 0.248 · 1.96 % | 2/3 · 0.248 · 1.85 % | 2/3 · 0.249 · 1.35 % | 1/2 · 0.250 · 1.42 % |
| Sep 8 | 4/10 | 3.67 % | 1/8 · 0.273 · 2.21 % | 1/2 · 0.249 · 2.94 % | 1/4 · 0.252 · 1.88 % | 4/7 · 0.249 · 1.75 % | ½ · 0.250 · 1.32 % | ½ · 0.250 · 1.52 % |
| Sep 4 | 4/10 | 1.97 % | 1/9 · 0.285 · 1.78 % | – | 2/9 · 0.255 · 2.98 % | – | 2/5 · 0.251 · 4.04 % | – |
| Sep 3 | 9/15 | 2.92 % | 5/12 · 0.242 · 1.91 % | – | 8/11 · 0.241 · 2.01 % | – | 6/8 · 0.246 · 2.92 % | – |
Names are picked by recent movement (top and bottom movers over 1, 5 and 21 sessions), not by conviction: it is the sampling frame, not a signal.
Latest resolved session by name (Oct 16)
Cell: probability of closing up / expected move.
| Name | Actual return | Haiku 4.5 A | Haiku 4.5 B | Sonnet 5 A | Sonnet 5 B | Opus 5 A | Opus 5 B |
|---|---|---|---|---|---|---|---|
| BNY | 0.81 % | 0.50 / 2.5 % | 0.50 / 1.7 % | 0.50 / 1.5 % | 0.50 / 1.1 % | 0.50 / 1.3 % | 0.50 / 1.2 % |
| CIEN | -6.67 % | 0.46 / 7.8 % | 0.50 / 9.5 % | 0.50 / 4.8 % | 0.49 / 3.5 % | 0.50 / 4.2 % | 0.50 / 3.7 % |
| FITB | 0.04 % | 0.50 / 2.0 % | 0.50 / 1.3 % | 0.50 / 1.3 % | 0.49 / 0.9 % | 0.50 / 1.2 % | 0.49 / 0.9 % |
| HPE | -5.97 % | 0.46 / 6.5 % | 0.50 / 4.7 % | 0.50 / 4.0 % | 0.49 / 2.8 % | 0.50 / 3.4 % | 0.50 / 3.0 % |
| KSS | -3.66 % | 0.54 / 7.2 % | 0.50 / 7.0 % | 0.50 / 2.8 % | 0.49 / 2.0 % | 0.50 / 2.6 % | 0.50 / 2.2 % |
| MRNA | -7.82 % | 0.53 / 8.5 % | 0.50 / 6.5 % | 0.50 / 5.0 % | 0.49 / 3.8 % | 0.50 / 4.5 % | 0.50 / 4.0 % |
| PNC | 1.14 % | 0.48 / 2.6 % | 0.50 / 1.4 % | 0.50 / 1.3 % | 0.49 / 1.0 % | 0.50 / 1.2 % | 0.50 / 1.0 % |
| RF | 0.51 % | 0.50 / 2.0 % | 0.52 / 3.4 % | 0.50 / 3.5 % | 0.52 / 3.4 % | 0.50 / 2.6 % | 0.52 / 3.0 % |
| SIG | 0.28 % | 0.52 / 7.0 % | 0.50 / 10.7 % | 0.50 / 2.8 % | 0.49 / 2.0 % | 0.50 / 2.6 % | 0.50 / 2.2 % |
| TRV | 1.47 % | 0.52 / 2.2 % | 0.52 / 2.9 % | 0.50 / 3.4 % | 0.52 / 2.9 % | 0.50 / 2.4 % | 0.52 / 2.6 % |
Method: predictions are recorded before the open and never on sessions already closed; packages are stored exactly as each model saw them; models get no headlines (licensing and memorisation). The directional-skill criterion (resolution > 0 with p < 0.05 over 200 predictions) and the A/B criterion were written down before starting.