Signal Performance
Transparent, honest track record from Sentinel's prediction ledger. All numbers are computed from pre-registered signal predictions and realised forward returns.
Hit Rate by Signal
Minimum 10 resolved predictions per signal. A "hit" means the signal's direction matched the sign of the 30-day forward return.
| Signal | Tier | Hit Rate | Correct / Total | Avg 30d Return |
|---|---|---|---|---|
| Conflict Intensity | 2 | 44.1% | 259 / 587 | +6.08% |
| Revenue Yoy | 1 | 53.1% | 285 / 537 | +6.04% |
| Options Put Call Skew | 2 | 52.9% | 209 / 395 | +6.21% |
| Institutional Flow | 1 | 61.8% | 239 / 387 | +6.47% |
| Social Options Interaction | 2 | 51.3% | 192 / 374 | +4.81% |
| Sentiment Combined | 2 | 51.0% | 172 / 337 | +4.47% |
| Stocktwits Net Sentiment | 3 | 57.8% | 160 / 277 | +3.43% |
| Polymarket Divergence | 2 | 52.1% | 134 / 257 | +2.57% |
| Factor Grade Quality | 2 | 53.4% | 135 / 253 | +2.32% |
| Borrow Cost Acceleration | 1 | 58.4% | 129 / 221 | +5.73% |
| Price Momentum 5d | 1 | 55.3% | 115 / 208 | +6.63% |
| Factor Grade Growth | 2 | 50.0% | 102 / 204 | +1.88% |
| Earnings Confidence | 2 | 60.1% | 113 / 188 | +11.95% |
| Hiring Momentum | 1 | 48.9% | 88 / 180 | +4.84% |
| Factor Grade Momentum | 2 | 67.4% | 64 / 95 | +3.30% |
| Factor Grade Value | 2 | 39.0% | 30 / 77 | +9.22% |
| Insider Cluster Sell | 1 | 23.0% | 17 / 74 | +21.61% |
| Earnings Tone Divergence | 2 | 49.2% | 31 / 63 | +20.57% |
| Si Daily Momentum 5d | 2 | 55.3% | 26 / 47 | +1.35% |
| Contract Momentum | 2 | 41.9% | 13 / 31 | +1.38% |
| Options Sentiment Pcr | 2 | 50.0% | 8 / 16 | +0.12% |
| Insider Cluster Buy | 1 | 43.8% | 7 / 16 | +1.06% |
Hit Rate by Market Regime
Monthly Hit Rate
Monthly accuracy of directional predictions. Green bars beat 55%, red below 45%, blue is the inconclusive zone in between. The dashed line marks 50% (coin-flip baseline).
| Month | Signals | Directional | Correct | Hit Rate |
|---|---|---|---|---|
| May 2026 | 1600 | 1154 | 614 | 53.2% |
| Apr 2026 | 5013 | 3685 | 1919 | 52.1% |
Calibration Curve
Signal magnitude (|value|) as a proxy for confidence, bucketed against observed hit rate. A well-calibrated model follows the diagonal — points above the line mean the model is under-confident, below means over-confident.
| Confidence Bucket | Predicted Prob | Observed Hit Rate |
|---|---|---|
| 5% | 5% | 57.3% |
| 15% | 15% | 48.4% |
| 25% | 25% | 47.1% |
| 35% | 35% | 55.2% |
| 45% | 45% | 60.9% |
| 55% | 55% | 48.1% |
| 65% | 65% | 47.1% |
| 75% | 75% | 52.4% |
| 85% | 85% | 50.0% |
| 105% | 105% | 52.1% |
Signal Information Coefficients
30-day Information Coefficient (Pearson correlation between signal value and forward return). Higher |IC| = more predictive. Green = positive correlation with returns, red = inverse.
| Signal | IC (30d) | |IC| | Sample Size | Strength |
|---|
Important Caveats
- Not investment advice. Sentinel is a research tool, not a recommendation engine. Past signal performance does not predict future results.
- Look-ahead bias risk. Although signals are recorded at extraction time (pre-registered), the composite weighting is updated monthly from the same ledger. This creates a mild feedback loop that inflates apparent IC vs true out-of-sample.
- Small sample sizes. Many signals have fewer than 100 resolved predictions. Hit rates with n < 50 have wide confidence intervals and should not be treated as reliable.
- Survivorship bias. The company universe has changed over time. Companies removed from coverage are excluded from recent calculations but included in historical periods.
- Forward-only sources. Signals derived from Polymarket, Bluesky, Reddit, and StockTwits only accumulate from their deployment date. Historical depth is thin.
- Unaudited. These statistics are computed internally and have not been independently verified. We publish them in the spirit of transparency, not as performance guarantees.