The whole track record.
Every slice, including misses.
Every prediction โ across all markets โ scored on the real close-to-close move, broken down by horizon, confidence, sector, region, model and regime. Calibration, risk-adjusted returns and platform discipline included. No cherry-picking.
95% confidence interval: 43.1%โ43.4% ยท All markets
Over these 90 days, each call's own direction happened 43.3% of the time in its market (an up call against how often prices rose, a down call against how often they fell). A forecaster with no skill would have scored about that, so our 43.2% is 0.1 points below that. When the market turns both numbers move together; the edge is what shows skill.
From 26 October 2026 OQ publishes and scores both up and down calls. Between 22 August 2026 and 26 October 2026 only up calls were published and scored, so the down calls made then are not counted; calls made before 22 August 2026 are scored in both directions, as they were published.
From 26 October 2026 OQ publishes and scores both up and down calls. Between 22 August 2026 and 26 October 2026 only up calls were published and scored, so the down calls made then are not counted; calls made before 22 August 2026 are scored in both directions, as they were published. Each call is compared with how often its own direction happened in its market over the same days, so a rise in accuracy when down calls enter a falling market reflects market direction, not skill.
By call direction. Each call is compared with how often each call's own direction happened in that market (same horizon, same days).
| Calls | Scored | Accuracy | Base | Lift |
|---|---|---|---|---|
| UP calls | 353,079 | 42.9% | 43.0% | โ0.1pp |
| DOWN calls | 70,309 | 44.9% | 44.7% | +0.2pp |
89,460 outcomes (17.4%) moved less than the symbol's noise band and are excluded as indeterminate โ not counted for or against.
Top-conviction calls ยท held 5d
Live (served) predictions only โ shadow-challenger calls, which users never saw, are excluded. No stop-loss and no price target: just the direction, held to the horizon. The base rate is how often each call's own direction happened in that market (same horizon, same days); lift is accuracy minus that rate, not minus 50%. This is a research signal, not investment advice.
The trust funnel
Every slot the system tried, through to the cohort the headline accuracy is computed on โ including the ones we declined to call. Hover a stage for its exact definition.
Accuracy over time
Rolling 7-day directional accuracy against the dashed line โ how often each call's own direction happened in that market on that day (an up call against how often prices rose, a down call against how often they fell). That line, not 50%, is where this book breaks even โ All markets.
Recent validation activity
Predictions scored per day, by the day the outcome resolved โ a throughput count, the honest answer to โis the scoring engine running right now?โ. It carries no accuracy: every accuracy chart below is by the day each call was made, and its recent days fill in as those calls mature.
Cumulative accuracy
Running accuracy across every scored prediction through each date โ All markets. It converges as the sample grows: early swings shrink as evidence accumulates, so a flat right-hand tail is the sample getting large, not performance freezing. Ends short of today by design, because recent predictions are still maturing.
Daily accuracy vs 7-day rolling
By the day each call was made. Daily values are noisy on thin days, so the rolling line is the one to read. Recent days are still maturing (the fainter line): only their shorter-horizon calls have been scored so far, and the figure fills in as the rest settle. A day's accuracy is drawn once at least 100 of its calls are scored; until then it shows only its scored and pending counts.
Scored predictions per day
Calls made each day and scored so far โ the sample behind each accuracy point. Recent days read low because their calls are still maturing, not because output stopped (837,542 calls in this window are still pending).
By horizon
Bullish vs bearish
Do the probabilities mean what they say?
The true reliability diagram: mean predicted P(up) against the realized up-rate in each probability bin. This is a different question from the confidence chart below it, which asks whether higher stated confidence means more calls come out directionally right. A model can rank direction usefully and still have probabilities that mean nothing โ these two charts are how you tell those apart.
overconfident, but that tag is only a direction test over the bin counts โ it has no notion of how large the errors are, so it reads the same whether the bins miss by one point or by ninety. With ECE 0.179 and MCE 0.898 the probabilities are not usable as literal odds, whichever direction the tag names.| Probability bin | Predicted P(up) | Realized up-rate | 95% CI | n | Realized โ predicted |
|---|---|---|---|---|---|
| 30โ40% | 37.4% | 58.0% | 57โ59% | 25,751 | +20.6pp |
| 40โ50% | 44.7% | 63.3% | 63โ64% | 13,570 | +18.6pp |
| 50โ60% | 53.0% | 58.9% | 58โ60% | 7,453 | +5.8pp |
| 60โ70% | 63.2% | 59.3% | 55โ64% | 442 | โ3.9pp |
| 70โ80% | 75.0% | 34.0% | 27โ42% | 144 | โ40.9pp |
| 80โ90% | 85.4% | 36.2% | 29โ44% | 141 | โ49.3pp |
| 90โ100% | 93.9% | 33.3% | 27โ40% | 195 | โ60.6pp |
Does confidence mean anything?
Actual directional hit-rate by the modelโs own stated confidence bucket โ higher confidence should mean more calls come out right. This is a different measurement from the reliability diagram above, which checks whether the calibrated probabilities match realized frequencies. Ranking well here and calibrating badly there is a real, common combination, so read them separately.
| Confidence band | Predictions | We said | We delivered | Delivered โ stated |
|---|---|---|---|---|
| Very Low (<50%) | 118,279 | 45.8% | 45.4% | โ0.4pp |
| Low (50-60%) | 144,331 | 54.8% | 40.9% | โ14.0pp |
| Medium (60-70%) | 68,826 | 64.0% | 43.5% | โ20.5pp |
| High (70-80%) | 53,689 | 73.7% | 44.1% | โ29.6pp |
| Very High (80-90%) | 25,261 | 85.4% | 44.4% | โ41.0pp |
| Extreme (90%+) | 13,002 | 90.3% | 42.6% | โ47.7pp |
Calibration by model
lower Brier / ECE is betterThe headline numbers above are the blend of every model that produced a scored call. Split out, they disagree โ which is the useful part. Delivered โ stated is realized accuracy minus stated confidence, the same convention as the band table above: negative means that model talks a better game than it plays.
| Model | Scored | Avg confidence | Accuracy | Delivered โ stated | Brier | ECE |
|---|---|---|---|---|---|---|
| modular | 45,189 | 42.4% | 59.9% | +17.5pp | 0.280 | 0.186 |
| v18 | 2,540 | 47.2% | 50.9% | +3.7pp | 0.236 | 0.118 |
Risk-adjusted returns
Measured on the disciplined book โ the signals we actually commit to after the confidence gate, not every raw signal. This is the money-weighted result.
Computed on all markets ยท all horizons ยท last 90 days, 265,298 closed trades.
For analysts โ Sharpe, Sortino, Calmar, information ratio, drawdown, risk/reward, VaR and CVaR
Tail risk & efficiency
How bad the bad trades get. VaR is the return the worst 1-in-20 (5%) and 1-in-100 (1%) trades beat; CVaR is the average of that tail โ the number that matters when the tail actually arrives.
Returns by horizon (holding-period)
How a user actually experiences it: hold from the call to its horizon. Sharpe here is annualized the correct way โ per-trade ร โ(252 / horizon-days), not a calendar-year compound.
This table covers all 425,719 scored signals in the window โ not only the 265,298 the gated book actually traded, which the cards above describe. Their profit factor is 0.74, against 0.81 for the traded book above.
| Horizon | n | Avg hold return | Sharpe (holding-period) | Trade win rate | Max DD |
|---|---|---|---|---|---|
| 1d | 28,461 | +0.199% | 0.69 | 53.4% | -7.58% |
| 5d | 80,602 | -0.550% | -0.21 | 46.0% | -33.40% |
| 7d | 73,879 | -0.811% | -0.32 | 44.0% | -47.04% |
| 10d | 81,606 | -1.259% | -0.36 | 42.2% | -59.47% |
| 15d | 70,019 | -1.662% | -0.73 | 41.1% | -67.17% |
| 20d | 13,755 | -2.379% | -0.75 | 39.9% | -74.58% |
| 25d | 10,907 | -1.401% | -0.46 | 41.6% | -70.09% |
| 30d | 27,086 | -2.261% | -0.51 | 39.4% | -82.77% |
| 45d | 16,363 | -0.396% | -0.06 | 44.3% | -73.46% |
| 60d | 16,066 | -2.521% | -0.29 | 42.6% | -61.91% |
| 90d | 6,975 | +0.960% | 0.06 | 48.2% | -35.21% |
Trade funnel
Strategy returns (gated book)
Cumulative return of the disciplined trade book โ the same daily equity series the Sharpe, Calmar and Max-drawdown figures above are built from, not the raw-signal firehose.
The 88 days behind the book return โ each counts once, whatever its trade count
| Exit day | Trades | Share of trades | Day return |
|---|---|---|---|
| 2026-08-19 | 4,505 | 1.7% | -0.11% |
| 2026-08-20 | 4,101 | 1.5% | +0.07% |
| 2026-08-21 | 4,657 | 1.8% | +0.20% |
| 2026-08-22 ยท thin day | 88 | 0.0% | +0.02% |
| 2026-08-23 ยท thin day | 40 | 0.0% | +0.00% |
| 2026-08-24 | 3,531 | 1.3% | +0.10% |
| 2026-08-25 | 3,195 | 1.2% | +0.04% |
| 2026-08-26 | 2,782 | 1.0% | +0.12% |
| 2026-08-27 | 2,586 | 1.0% | +0.09% |
| 2026-08-28 | 3,865 | 1.5% | +0.15% |
| 2026-08-29 ยท thin day | 36 | 0.0% | +0.00% |
| 2026-08-30 ยท thin day | 52 | 0.0% | -0.00% |
| 2026-08-31 | 3,501 | 1.3% | -0.02% |
| 2026-09-01 | 3,889 | 1.5% | +0.02% |
| 2026-09-02 | 3,701 | 1.4% | -0.08% |
| 2026-09-03 | 3,032 | 1.1% | +0.10% |
| 2026-09-04 | 3,631 | 1.4% | +0.12% |
| 2026-09-05 ยท thin day | 57 | 0.0% | +0.01% |
| 2026-09-06 ยท thin day | 53 | 0.0% | +0.01% |
| 2026-09-07 | 1,887 | 0.7% | +0.12% |
| 2026-09-08 | 3,384 | 1.3% | +0.02% |
| 2026-09-09 | 4,787 | 1.8% | -0.09% |
| 2026-09-10 | 4,348 | 1.6% | -0.21% |
| 2026-09-11 | 7,301 | 2.8% | -0.27% |
| 2026-09-12 ยท thin day | 64 | 0.0% | -0.00% |
| 2026-09-13 ยท thin day | 78 | 0.0% | -0.00% |
| 2026-09-14 | 5,868 | 2.2% | -0.28% |
| 2026-09-15 | 5,840 | 2.2% | -0.41% |
| 2026-09-16 | 5,290 | 2.0% | -0.34% |
| 2026-09-17 | 4,452 | 1.7% | -0.04% |
| 2026-09-18 | 8,023 | 3.0% | -0.05% |
| 2026-09-19 ยท thin day | 86 | 0.0% | +0.00% |
| 2026-09-20 ยท thin day | 236 | 0.1% | +0.00% |
| 2026-09-21 | 5,411 | 2.0% | +0.06% |
| 2026-09-22 | 5,909 | 2.2% | +0.11% |
| 2026-09-23 | 5,002 | 1.9% | -0.03% |
| 2026-09-24 | 5,720 | 2.2% | -0.23% |
| 2026-09-25 | 6,794 | 2.6% | -0.17% |
| 2026-09-26 ยท thin day | 85 | 0.0% | +0.00% |
| 2026-09-27 ยท thin day | 58 | 0.0% | -0.00% |
| 2026-09-28 | 6,447 | 2.4% | -0.33% |
| 2026-09-29 | 6,513 | 2.5% | -0.34% |
| 2026-09-30 | 5,666 | 2.1% | -0.14% |
| 2026-10-01 | 7,569 | 2.9% | -0.37% |
| 2026-10-02 | 6,263 | 2.4% | -0.08% |
| 2026-10-03 ยท thin day | 62 | 0.0% | -0.00% |
| 2026-10-04 ยท thin day | 90 | 0.0% | -0.00% |
| 2026-10-05 | 4,698 | 1.8% | +0.06% |
| 2026-10-06 | 3,526 | 1.3% | +0.10% |
| 2026-10-07 | 3,813 | 1.4% | -0.07% |
| 2026-10-08 | 4,444 | 1.7% | -0.27% |
| 2026-10-09 | 4,299 | 1.6% | -0.02% |
| 2026-10-10 ยท thin day | 44 | 0.0% | +0.00% |
| 2026-10-11 ยท thin day | 120 | 0.0% | -0.01% |
| 2026-10-12 | 3,177 | 1.2% | -0.06% |
| 2026-10-13 | 3,135 | 1.2% | -0.10% |
| 2026-10-14 | 2,548 | 1.0% | -0.04% |
| 2026-10-15 | 4,607 | 1.7% | -0.20% |
| 2026-10-16 | 4,041 | 1.5% | +0.02% |
| 2026-10-17 ยท thin day | 73 | 0.0% | -0.00% |
| 2026-10-18 ยท thin day | 70 | 0.0% | -0.00% |
| 2026-10-19 | 3,222 | 1.2% | -0.12% |
| 2026-10-20 | 2,928 | 1.1% | -0.07% |
| 2026-10-21 | 3,024 | 1.1% | -0.06% |
| 2026-10-22 | 4,407 | 1.7% | -0.16% |
| 2026-10-23 | 5,146 | 1.9% | -0.26% |
| 2026-10-24 ยท thin day | 64 | 0.0% | -0.00% |
| 2026-10-25 ยท thin day | 217 | 0.1% | -0.03% |
| 2026-10-26 | 3,367 | 1.3% | -0.09% |
| 2026-10-27 | 3,735 | 1.4% | +0.06% |
| 2026-10-28 | 3,586 | 1.4% | +0.07% |
| 2026-10-29 | 3,277 | 1.2% | -0.00% |
| 2026-10-30 | 4,767 | 1.8% | -0.09% |
| 2026-10-31 ยท thin day | 25 | 0.0% | -0.00% |
| 2026-11-01 ยท thin day | 10 | 0.0% | +0.00% |
| 2026-11-02 | 4,239 | 1.6% | +0.02% |
| 2026-11-03 | 3,581 | 1.3% | +0.01% |
| 2026-11-04 | 3,575 | 1.3% | +0.10% |
| 2026-11-05 | 3,302 | 1.2% | +0.17% |
| 2026-11-06 | 3,911 | 1.5% | -0.08% |
| 2026-11-07 ยท thin day | 21 | 0.0% | -0.00% |
| 2026-11-08 ยท thin day | 183 | 0.1% | +0.00% |
| 2026-11-09 | 3,173 | 1.2% | +0.06% |
| 2026-11-10 | 1,715 | 0.6% | +0.07% |
| 2026-11-11 ยท thin day | 1,388 | 0.5% | +0.05% |
| 2026-11-12 ยท thin day | 1,362 | 0.5% | +0.03% |
| 2026-11-13 | 1,580 | 0.6% | +0.04% |
| 2026-11-16 ยท thin day | 363 | 0.1% | -0.00% |
By asset class
By region
By sector
By model
Accuracy by market ร horizon
Directional accuracy for every market at every horizon, each compared with that market's own base rate at that horizon (how often each call's own direction happened there over the same window), never with a fixed 50% line. This is the joint cut, not the two marginal breakdowns above: a market can look strong overall and still trail its own base rate on the horizon you actually trade. Cohort: All markets ยท last 90 days.
| Market | 1d | 5d | 7d | 10d | 15d | 20d | 25d | 30d | 45d | 60d |
|---|---|---|---|---|---|---|---|---|---|---|
| United States | 47.4% n=8,725 | 47.7% n=16,009 | 47.3% n=13,022 | 45.1% n=15,410 | 41.1% n=12,504 | 45.1% n=1,521 | 39.9% n=1,781 | 34.8% n=1,712 | 39.9% n=992 | 34.6% n=101 |
| India | 58.4% n=7,918 | 41.3% n=15,631 | 37.8% n=16,171 | 37.8% n=15,404 | 38.9% n=8,665 | 42.0% n=2,526 | 28.5% n=1,308 | 25.3% n=4,029 | 38.8% n=539 | โ |
| Europe | 50.6% n=11,112 | 49.3% n=24,337 | 48.6% n=19,941 | 47.6% n=20,964 | 47.8% n=16,048 | 50.9% n=2,525 | 50.2% n=1,986 | 46.1% n=3,987 | 47.2% n=1,530 | 52.0% n=298 |
| Asia-Pacific | 52.0% n=14,908 | 43.0% n=34,035 | 41.8% n=28,398 | 39.0% n=28,333 | 34.0% n=19,633 | 33.9% n=5,070 | 37.5% n=2,227 | 27.3% n=4,729 | 37.6% n=2,317 | 26.1% n=494 |
| Crypto | 54.0% n=472 | 49.1% n=658 | 50.5% n=525 | 44.2% n=593 | 46.9% n=520 | 16.7% n=78 | 18.5% n=27 ยท thin | 62.7% n=201 | 51.9% n=104 | 63.6% n=11 ยท thin |
| Forex | 47.2% n=335 | 39.2% n=255 | 43.4% n=332 | 39.1% n=289 | 31.1% n=206 | 34.8% n=46 | 27.9% n=43 | 17.5% n=63 | 39.2% n=51 | 16.7% n=12 ยท thin |
| Commodities | 54.1% n=364 | 41.3% n=697 | 40.3% n=439 | 44.7% n=640 | 34.9% n=627 | 15.5% n=103 | 40.4% n=57 | 35.5% n=124 | 25.0% n=36 | 38.1% n=21 ยท thin |
Green is above that cell's own market base rate, red is below; intensity saturates at ยฑ10 points so a thin outlier can't shout down a real one. A cell without a base rate yet is shown uncoloured, with its accuracy and n. 4 cells are dimmed for a sample under 30 โ those numbers are shown as measured, but they are not yet evidence of anything. A cell with no scored predictions shows an em-dash rather than a zero.
Returns by market ร horizon
What following the calls actually returned, for every market at every horizon โ the same joint cut as the accuracy grid above, in the unit that pays. These come from the trade ledger, not the scored prediction cohort, so the sample sizes are their own and will not match cell for cell. Cohort: All markets - last 90 days.
| Market | 1d | 5d | 7d | 10d | 15d | 20d | 25d | 30d | 45d | 60d | 90d |
|---|---|---|---|---|---|---|---|---|---|---|---|
| United States | +0.10% n=3,296 ยท 46% won | -0.12% n=8,712 ยท 42% won | -0.19% n=6,846 ยท 40% won | -0.43% n=7,685 ยท 40% won | -1.04% n=7,711 ยท 35% won | -1.87% n=1,286 ยท 30% won | -1.72% n=1,742 ยท 32% won | -2.11% n=3,602 ยท 31% won | -0.99% n=1,717 ยท 38% won | -2.11% n=1,047 ยท 37% won | +4.24% n=315 ยท 68% won |
| India | +0.39% n=2,344 ยท 55% won | -0.35% n=6,374 ยท 37% won | -0.60% n=6,183 ยท 35% won | -0.99% n=5,755 ยท 33% won | -1.84% n=5,357 ยท 29% won | -2.86% n=1,786 ยท 25% won | -2.38% n=1,307 ยท 29% won | -2.56% n=3,690 ยท 30% won | -1.07% n=1,688 ยท 37% won | -1.35% n=1,100 ยท 40% won | -0.52% n=97 ยท 39% won |
| Europe | +0.14% n=5,616 ยท 47% won | +0.01% n=12,769 ยท 44% won | -0.08% n=11,807 ยท 44% won | -0.22% n=11,054 ยท 42% won | -0.46% n=10,098 ยท 41% won | -1.00% n=2,034 ยท 39% won | -0.82% n=2,134 ยท 43% won | -0.99% n=5,899 ยท 41% won | +0.08% n=2,427 ยท 48% won | +0.66% n=1,432 ยท 52% won | +1.53% n=237 ยท 57% won |
| Asia-Pacific | -0.03% n=4,865 ยท 40% won | -0.51% n=19,261 ยท 34% won | -0.57% n=17,026 ยท 36% won | -0.84% n=15,770 ยท 36% won | -1.11% n=13,065 ยท 34% won | -0.83% n=4,365 ยท 38% won | -0.40% n=2,809 ยท 42% won | -0.28% n=9,280 ยท 42% won | +1.42% n=3,531 ยท 52% won | +2.12% n=3,424 ยท 56% won | +5.08% n=1,132 ยท 65% won |
| Crypto | -0.44% n=221 ยท 35% won | -0.26% n=310 ยท 36% won | +1.09% n=294 ยท 49% won | -0.17% n=259 ยท 39% won | -0.56% n=242 ยท 34% won | +3.76% n=59 ยท 59% won | +9.86% n=37 ยท 89% won | +11.70% n=235 ยท 89% won | +16.21% n=103 ยท 96% won | +21.02% n=82 ยท 100% won | +24.67% n=9 ยท 100% won ยท thin |
| Forex | -0.05% n=207 ยท 49% won | -0.12% n=101 ยท 37% won | -0.10% n=173 ยท 46% won | +0.01% n=145 ยท 60% won | -0.33% n=99 ยท 47% won | -0.49% n=31 ยท 48% won | -0.44% n=60 ยท 48% won | -0.69% n=102 ยท 41% won | -1.51% n=47 ยท 32% won | -2.27% n=12 ยท 8% won ยท thin | -0.87% n=7 ยท 57% won ยท thin |
| Commodities | -0.03% n=147 ยท 48% won | -0.07% n=436 ยท 44% won | -0.25% n=284 ยท 40% won | +0.39% n=360 ยท 47% won | -0.33% n=389 ยท 42% won | +0.52% n=100 ยท 46% won | +3.43% n=66 ยท 68% won | +4.42% n=232 ยท 65% won | +6.39% n=95 ยท 75% won | +7.70% n=92 ยท 76% won | +13.68% n=11 ยท 91% won ยท thin |
| INDX | +1.08% n=10 ยท 80% won ยท thin | +0.00% n=43 ยท 60% won | +0.73% n=18 ยท 56% won ยท thin | -0.64% n=25 ยท 36% won ยท thin | -0.07% n=28 ยท 61% won ยท thin | +2.86% n=1 ยท 100% won ยท thin | +2.33% n=7 ยท 86% won ยท thin | +0.54% n=18 ยท 56% won ยท thin | -0.34% n=2 ยท 50% won ยท thin | +2.18% n=2 ยท 50% won ยท thin | +2.25% n=4 ยท 75% won ยท thin |
| Latin America | +0.15% n=982 ยท 41% won | +0.23% n=1,660 ยท 42% won | +0.29% n=1,883 ยท 44% won | +0.59% n=1,686 ยท 46% won | +0.66% n=1,592 ยท 46% won | +2.56% n=424 ยท 58% won | +3.52% n=386 ยท 62% won | +3.32% n=1,026 ยท 60% won | +2.94% n=346 ยท 62% won | +4.05% n=282 ยท 64% won | +2.38% n=45 ยท 56% won |
| Middle East & Africa | -1.39% n=7 ยท 0% won ยท thin | -0.43% n=531 ยท 36% won | -1.29% n=2,678 ยท 29% won | -1.26% n=2,683 ยท 32% won | -1.93% n=1,932 ยท 28% won | -2.56% n=609 ยท 28% won | -2.63% n=460 ยท 29% won | -2.78% n=840 ยท 31% won | -3.74% n=208 ยท 29% won | +0.44% n=153 ยท 46% won | +3.98% n=5 ยท 60% won ยท thin |
Green is a profit, red is a loss โ the pivot is zero, not the 50% of the accuracy grid, because this is a return and the null result is โmade nothingโ. Intensity saturates at ยฑ2% per trade, so a thin outlier can't shout down a real one. 16 cells are dimmed for a sample under 30 โ those numbers are shown as measured, but they are not yet evidence of anything, and the annualized view magnifies them hardest. A market/horizon pair with no trades shows an em-dash rather than a zero.
Net edge vs benchmark
Portfolio return minus what the benchmark did over the same window, per market, each against its own named local index. Positive means the calls beat simply holding that index; negative means they did not, and it is printed here either way.
Measuring net edge against each market's benchmarkโฆ
Performance by market regime
Not just accuracy per regime โ the return, Sharpe and win rate of the calls made in each market state. This is where you find out whether an edge is real or is one regime carrying the whole record.
Volatility x trend (ex-ante tags)
The market state recorded at the moment each call was made โ knowable in advance, so these numbers are the honest ones.
| Regime | n | Accuracy | Avg return | Sharpe | Trade win rate | Coverage |
|---|---|---|---|---|---|---|
| high trending | 84,062 | 46.9% | -0.89% | -1.06 | 46.9% | 42.0% |
| medium trending | 76,612 | 50.8% | +0.09% | 0.16 | 50.8% | 38.3% |
| medium mean reverting | 27,917 | 51.5% | +0.17% | 0.52 | 51.5% | 14.0% |
| high mean reverting | 10,971 | 48.8% | -0.78% | -1.29 | 48.8% | 5.5% |
Realized move size (post-hoc diagnostic)
| Move size | n | Accuracy | Avg return | Sharpe | Trade win rate | Coverage |
|---|---|---|---|---|---|---|
| small | 73 | 53.4% | +0.11% | 0.84 | 53.4% | 0.0% |
| mid | 57 | 57.9% | +0.49% | 1.87 | 57.9% | 0.0% |
| large | 45 | 40.0% | -2.58% | -4.02 | 40.0% | 0.0% |
Discipline & coverage
Most attempts never become a call at all โ the gate abstains upstream, before a prediction is issued, and those abstentions are counted against us in the two figures on the right. What remains is scored in full. Committed calls here are the published cohort, which applies no confidence floor โ every scored prediction is included.
Abstention by horizon
platform-wide 23.3%Share of attempted calls the engine declined to publish, per horizon. Higher means more was withheld.
| Horizon | Attempted | Abstained | Abstention rate |
|---|---|---|---|
| 1d | 251,121 | 128,400 | 51.1% |
| 5d | 271,602 | 50,423 | 18.6% |
| 7d | 266,328 | 33,300 | 12.5% |
| 10d | 263,544 | 38,882 | 14.8% |
| 15d | 311,836 | 58,943 | 18.9% |
| 20d | 259,318 | 51,457 | 19.8% |
| 25d | 254,787 | 47,243 | 18.5% |
| 30d | 351,390 | 106,678 | 30.4% |
| 45d | 290,637 | 70,289 | 24.2% |
| 60d | 292,389 | 71,439 | 24.4% |
| 90d | 284,072 | 65,890 | 23.2% |
The intelligence stack
Which models are serving the predictions scored above, and the rules that govern replacing them.
Currently serving
Cross-reference these against the โBy modelโ breakdown above โ a version with a small share of the book is a challenger being measured, not the champion.
DAILY_LIGHT_v20260622_1414_90d_AUTop forecasters
No per-agent forecast records scored yet. This leaderboard fills once the swarm attributes settled outcomes to individual agents โ until then there is nothing here to rank, and we would rather show that than an empty table.
Cohort definition & audit trail
canonical_trust_strictExactly which predictions the numbers above are computed over, and what was filtered out to get there. Published so the accuracy figure can be checked rather than taken on faith.
Strict directional accuracy โ a prediction is scored correct only when the symbol moved beyond its noise band and the model called that direction. Band-indeterminate outcomes (see indeterminate.band_description for the exact rule in force) are excluded from both the numerator and the denominator โ not counted for or against. Cohort also excludes archived, abstained, out-of-universe, and shadow-challenger predictions. No confidence floor is applied: every scored prediction is in the cohort, including the ones the model was least sure of. From 26 October 2026 OQ publishes and scores both up and down calls. Between 22 August 2026 and 26 October 2026 only up calls were published and scored, so the down calls made then are not counted; calls made before 22 August 2026 are scored in both directions, as they were published.
This window spans a change to the published cohort on 2026-10-09: it mixes calls published under the old rule and the new one, so a move in the figure can be the population changing, not the model. Calls below the former 0.52 confidence cut are now published and scored. Before this date about 31% of batch calls were withheld by that cut; the confidence tiers (TRADE_CANDIDATE / WATCH / BELOW_WATCH) are now display labels only.
walk-forward, daily by maturation. Sections backed by those are hidden rather than shown with partial numbers.Cohort invariants
Every prediction counted above satisfies all of these. Each one removes a population that would otherwise flatter the number.
- 1
archived != True (exclude pre-2026-05-07 broken-pipeline cohort) - 2
abstain != True (exclude model-declined predictions) - 3
in_universe != False (exclude out-of-universe symbols) - 4
is_neutral != True (exclude neutral predictions from directional accuracy) - 5
is_shadow != True (exclude shadow-challenger predictions โ the ~75% not shown to users) - 6
confidence >= 0 (no confidence floor โ every scored prediction is included, including the lowest-confidence ones) - 7
prediction_type not in ['causal', 'crowd', 'sentiment'] and source not in ['predict-any', 'realtime'] (published model calls only โ not non-model readings, not on-demand runs) - 8
direction as published: UP or DOWN, except DOWN calls made 2026-08-22 to 2026-10-26 (never published, never scored)
Filters applied
The query this snapshot was built with. null means unfiltered.
- days back
- 90
- exchange filter
- โ
- model version
- โ
- horizons
- โ
- direction
- โ
Excluded from the cohort
Printed exactly as the backend reports it โ where a count is not separately tracked it says so rather than showing a zero.
- archived count excluded
- see #202 archive script โ 11,153 docs archived 2026-05-16
- abstained count
- 722,944
- out of universe count
- not separately surfaced
- neutral count
- 89,460
- low confidence count
- not separately surfaced
Top-conviction ledger
The highest-conviction decile of settled calls, ranked on the calibrated probability โ not the stated confidence. Held to the horizon with no stop and no target, which is why these rows reconcile with the hero number above and not with the traded book.
This is the conviction slice itself. These are the individual predictions behind every aggregate number above โ one row per call, with the outcome it actually settled to. The aggregates stay public; the row-level record is for signed-in members.
Audit ledger
Every scored prediction in the canonical cohort, newest settlement first. This is the same set of rows the headline accuracy is computed over โ the six invariants noted beneath the table are applied identically โ so this is the panel to audit against. The recent-settled panel is a wider, non-canonical cohort and will not tie out.
This is the canonical audit trail. These are the individual predictions behind every aggregate number above โ one row per call, with the outcome it actually settled to. The aggregates stay public; the row-level record is for signed-in members.
Recent settled predictions
The latest calls that have already settled โ pending ones are pulled from the same response and filtered out here, so this is scored rows only. This cohort is not the canonical one. It excludes archived and shadow-challenger rows only โ it does not apply the six canonical invariants the headline accuracy above is computed on, so tallying these rows will not tie out to that number. The audit ledger panel is the cohort that does. Every market, because the scope above is set to Global.
This is the recent settled record. These are the individual predictions behind every aggregate number above โ one row per call, with the outcome it actually settled to. The aggregates stay public; the row-level record is for signed-in members.
How we measure
Directional
Right if the price moves the way we said by the horizon. Unambiguous, checkable.
Close-to-close
Official closes at the horizon date โ not intraday ticks we could pick to flatter.
No look-ahead
Locked before the outcome window opens. Out-of-sample, every time.
Against the base rate
Each call is compared with how often its own direction happened in its market and horizon โ not a fixed 50%.
The record is the proof. Now see the picks.
Todayโs high-conviction calls for India.