Last 30 Days
Sample 329 · Actual up 52 / non-up 277 · majority class Non-up
Direction Accuracy
Model direction
33.4%
Majority-class rule
84.2%
Price Proximity
Model reference price
90.5%
Assume no price change
93.6%
Warming up the engine
Evidence limit: stock outcome observations have not been shown to use the same price-adjustment basis as the original records. These stored statistics need further checking and do not establish an advantage for a default method or future performance.
Evaluation period: 30 days (ending 2026-10-26T15:33:43.414Z). Metric version: terminal_direction_v1.
All-time verified samples: Direction consistency N=14,658, SMAPE N=14,662; Direction consistency: 51.0%; Model reference price proximity: 90.5%.
Direction consistency compares whether the model reference price and the later observed price are on the same side of the recorded current price. Its sample also requires a valid recorded current price and model reference price. Price proximity is 100% minus mean SMAPE between the model reference price and observed price; it is not a return, and its separate sample requires valid predicted and actual prices with a nonzero denominator. Both metrics use records with status correct or incorrect and a verified actual price; pending records are excluded. The public endpoint does not separately classify overdue records from other pending records.
A dash means a metric has no eligible samples or its sample count is unavailable; it is not a zero score. This is a retrospective aggregate, not a forecast of future performance. Model stability is a reported score, not a calibrated probability. Sample size, market, horizon, model version, and snapshot time are part of the result.
Public data source: GET /api/v1/public/accuracy-dashboard?days=30
Direction consistency, model reference price proximity, and stability across verified research records
Historical aggregate for research review. Each result should be read with its verified sample size, market, horizon, and model version.
Evidence limit: stock outcome observations have not been shown to use the same price-adjustment basis as the original records. These stored statistics need further checking and do not establish an advantage for a default method or future performance.
329
Matured records with verified market outcomes
33.4%
Share of verified records with matching model and observed direction
90.5%
How close model reference prices are to observed prices (100 - mean SMAPE)
408
Research records created in the selected period
79%
Average model stability score
36
Users who made predictions in this period
Model metrics are shown beside simple references calculated on the same matured, verified sample, so no percentage stands alone.
Sample 329 · Actual up 52 / non-up 277 · majority class Non-up
Direction Accuracy
Model direction
33.4%
Majority-class rule
84.2%
Price Proximity
Model reference price
90.5%
Assume no price change
93.6%
Sample 14,658 · Actual up 6,566 / non-up 8,092 · majority class Non-up
Direction Accuracy
Model direction
51.0%
Majority-class rule
55.2%
Price Proximity
Model reference price
90.5%
Assume no price change
92.3%
These naive baselines use the same verified sample as the model metrics. They only test whether the model adds information beyond simple rules; they are not trading strategies or returns.
Aggregated verified samples by market, horizon, and stability bucket; no portfolio or trading-strategy metrics are included.
| Segment | Samples | Direction Accuracy | 95% Range | Price Proximity | Avg Stability |
|---|---|---|---|---|---|
| A-Shares · 1D | 149 | 47.7% | 39.8-55.6% | 93.8% · N 149 | 73.2% |
| A-Shares · 3D | 101 | 2% | 0.5-6.9% | 86.6% · N 101 | 93.4% |
| A-Shares · 7D | 67 | 49.3% | 37.7-60.9% | 89.8% · N 67 | 75.8% |
| A-Shares · 5D | 8 | 50% | 21.5-78.5% | 91.4% · N 8 | 72.1% |
| HK Stocks · 7D | 3 | 0% | 0-56.2% | 72.6% · N 3 | 70.7% |
| Crypto · 7D | 1 | 0% | 0-79.3% | 97.6% · N 1 | 67% |
| Horizon | Samples | Direction Accuracy |
|---|---|---|
| 1D | 149 | 47.7% |
| 3D | 101 | 2% |
| 5D | 8 | 50% |
| 7D | 71 | 46.5% |
Stability has a weak negative relationship with direction accuracy. Do not present high stability as a higher probability of directional agreement.
These are cumulative samples (for example, ≥70% includes every sample with stability at or above 70%), showing sample size and direction consistency only.
Create model reference paths and review them against subsequent market outcomes