Model Performance
Backtest window Jan 2022 – Dec 2025 · 214 completed / terminated vehicles · simulated demo results.
Completion AUC
0.782
Higher is better
Brier score
0.163
Lower is better
Redemption MAE
8.4pp
Percentage points
Redemption R²
0.61
Fit vs actuals
Flagged before termination
83%
Deals scored ≥60 redemption
False-positive rate
14%
Model demo
Monthly accuracy — completion vs redemption
Returns by signal
| Signal | N | Hit rate | Avg 12m return |
|---|---|---|---|
| PIPE cut > $25M | 42 | 79% | -34.2% |
| Extension #3+ | 28 | 71% | -22.4% |
| Going-concern language | 11 | 91% | -48.6% |
| Cash mix reduced > 10pp | 19 | 63% | -12.9% |
| Sponsor conflict disclosed | 34 | 56% | -9.4% |
| PIPE upsize | 22 | 68% | +4.7% |
| Model completion > 75 | 61 | 74% | -6.8% |
High-risk vs low-risk outcomes
| Bucket | Completion | Avg 12m | Terminated |
|---|---|---|---|
| High-risk (redemption ≥ 70) | 31% | -32.4% | 44% |
| Mid-risk (40–69) | 62% | -16.8% | 19% |
| Low-risk (< 40) | 81% | -4.1% | 7% |
Confusion matrix (completion predictor)
Actual: Closed
Actual: Terminated
Pred: Closed
74
21
Pred: Terminated
23
96
n = 214 · precision 78% · recall 76%
Redemption calibration
Bars show actual mean redemption for each predicted bucket.
Backtest is a simulated demonstration. Results are not indicative of live model performance and are not investment advice.