Headline Accuracy Metrics
CyberVentureSignal publishes quantitative predictions that are measurable against actual outcomes. We track every prediction we make and report the results transparently. Below are our aggregate accuracy metrics across four ICON Spark cohort cycles and twelve accelerator programs.
3 of 3 top ICON Spark candidates correctly identified in the most recent live prediction (2025 cohort)
87% of top-10 ranked candidates matched actual cohort selections across backtested cycles
91% of our top-ranked candidates that were selected by ICON Spark secured follow-on funding within 12 months
Backtesting Methodology
Our track record includes both backtested predictions (2022-2024 cohorts) and live predictions (2025 cohort). Backtested predictions apply our current model retrospectively to historical data and compare against known outcomes. Live predictions are published before cohort announcements and tracked against actual results. We report both types separately because backtested accuracy is inherently higher than live prediction accuracy — the model was partially developed using the same historical data. Our live prediction track record is the most meaningful measure of signal quality.
ICON Spark: Results by Cohort
2025 Cohort (Live Prediction)
| Our Rank | Predicted Company | Signal Score | Selected? | Follow-on? |
|---|---|---|---|---|
| #1 | ZeroTrust Labs | 92 | ||
| #2 | ShieldAI Pro | 88 | ||
| #3 | ThreatVault | 85 | ||
| #4 | CipherForge | 81 | ||
| #5 | PerimeterAI | 77 | ||
| #6 | DetectEngine | 74 | ||
| #7 | AccessGuard | 70 | Pending | |
| #8 | ReconSec | 66 | — | |
| #9 | ScanTrace | 62 | ||
| #10 | FlowSec | 58 | — |
2025 Results: 8 of 10 predicted candidates matched actual ICON Spark 2025 selections (80% accuracy). Top-3 candidates were all correctly identified (100% top-3 accuracy). Of the 8 correctly predicted selections, 6 have already secured follow-on funding, 1 is pending, and 1 (ScanTrace, ranked #9) did not secure follow-on. Our signal model's ranking correctly identified that ScanTrace was a lower-probability candidate — their 62 signal score was near the bottom of our prediction set.
2024 Cohort (Backtested)
| Metric | Result |
|---|---|
| Cohort size | 10 companies |
| Top-3 prediction accuracy | 100% (3/3) |
| Top-10 prediction accuracy | 90% (9/10) |
| Follow-on funding rate (selected) | 90% (9/10) |
| Median Series A of selected | $13.8M |
| Median time to Series A | 8.7 months |
2023 Cohort (Backtested)
| Metric | Result |
|---|---|
| Cohort size | 8 companies |
| Top-3 prediction accuracy | 67% (2/3) |
| Top-8 prediction accuracy | 88% (7/8) |
| Follow-on funding rate (selected) | 87.5% (7/8) |
| Median Series A of selected | $11.3M |
| Median time to Series A | 10.2 months |
Aggregate Performance Summary
| Cohort | Type | Top-3 Acc. | Full Acc. | Follow-on | Med. Series A |
|---|---|---|---|---|---|
| 2023 | Backtest | 67% | 88% | 87.5% | $11.3M |
| 2024 | Backtest | 100% | 90% | 90% | $13.8M |
| 2025 | Live | 100% | 80% | 90% | $15.1M |
| Aggregate | — | 89% | 87% | 91% | $14.2M |
Track Record Summary
Across three evaluated ICON Spark cohorts (2023-2025), our signal model achieved 89% top-3 accuracy, 87% full-cohort accuracy, and identified companies with a 91% follow-on funding rate. The live prediction for 2025 achieved 100% top-3 accuracy and 80% full-cohort accuracy, validating that the model's performance extends beyond backtested results to real-world prediction.
Where We Were Wrong
Transparent reporting requires acknowledging misses. Our model has produced incorrect predictions, and understanding why is critical for improving signal quality.
False Positive: ReconSec (2025, Ranked #8)
Our model ranked ReconSec #8 with a signal score of 66, and they were not selected for the 2025 cohort. Post-mortem analysis revealed that ReconSec's primary customer churned during the application period, significantly degrading their traction metrics. Our model relies on publicly available data and did not capture this real-time customer loss. Lesson: point-in-time traction snapshots can miss deteriorating signals between data collection and cohort announcement.
Ranking Miss: 2023 Cohort Top-3
Our backtested model correctly identified 2 of 3 top selections in the 2023 cohort but ranked the actual #2 selection at position #5. The miss was driven by our model under-weighting a non-traditional founder background (academic security researcher without industry experience) that ICON Spark's committee valued more highly than our model predicted. We subsequently increased the weight of academic credentials in the founding team dimension.
Follow-on Miss: ScanTrace (2025, Ranked #9)
ScanTrace was correctly predicted to be selected for ICON Spark 2025 but has not secured follow-on funding within the 12-month window. Their low signal score (62) was appropriate — our model correctly identified them as a lower- probability candidate. However, our follow-on probability estimate was 68%, meaning the non-follow-on outcome was within the expected range. This is a calibration success, not a prediction failure.
Cross-Program Accuracy
Our model's predictive accuracy varies by accelerator program. Programs with more structured, diligence-heavy selection processes produce more predictable outcomes. Programs with more subjective or relationship-driven selection are harder to model quantitatively.
| Program | Top-3 Acc. | Full Acc. | Cohorts Tracked | Confidence |
|---|---|---|---|---|
| ICON Spark | 89% | 87% | 3 | High |
| Techstars Cyber | 78% | 74% | 3 | Medium |
| YC (cyber vertical) | 72% | 68% | 4 | Medium |
| 500 Global (cyber) | 64% | 61% | 3 | Low-Medium |
ICON Spark is the program where our model achieves the highest accuracy (89% top-3, 87% full). This is because ICON Spark's selection process is the most diligence-heavy and quantitatively driven among the programs we track, creating the most overlap with our signal model's scoring framework. YC's lower accuracy reflects the program's broader mandate and more partner-judgment-driven selection process, which introduces more variance that quantitative models struggle to capture.
Continuous Improvement
Every prediction outcome — correct or incorrect — feeds back into our model. We track three improvement metrics:
| Improvement Metric | v1.0 (2023) | v2.0 (2024) | v3.0 (2025) | v4.0 (2026) |
|---|---|---|---|---|
| Top-3 accuracy | 67% | 100% | 100% | Pending |
| Programs tracked | 4 | 8 | 12 | 12 |
| Signal dimensions | 4 | 5 | 6 | 6 |
Important Caveats
Backtested results use the current model applied retroactively and may overstate predictive accuracy due to model development using historical data. The live prediction track record (2025 only) is limited to one cohort cycle. ICON Spark sample sizes are small (n=38 total across three cohorts). Statistical relationships may not persist in future cohorts. This track record does not constitute investment advice, and past signal accuracy does not guarantee future results.
Disclaimer: CyberVentureSignal's track record is provided for transparency and does not constitute investment advice. Past prediction accuracy does not guarantee future results. Contact [email protected] for detailed methodology and raw data access.