Castle
Signal trajectory
realSignal 6.1 → 9.6 across 26 readings (+3.6). Composite of press, GitHub velocity, community heat, liveness and funding.
Cross-sectional profile
percentile rank vs 79,105 modeled productsPercentiles computed over the modeled cohort (embedding + signal + tags). Sparse dimensions — network and press — reflect that most products have zero, so a single edge or mention lifts the rank; read those with the coverage flag.
Mortality outlook
verified labelThis product carries a verified outcome label (“surviving”) — it is one of 600 hand-labeled cases the model is trained and tested on.
- LightGBM · binary mortality (dead vs surviving)
- Held-out AUC 0.71 · CV AUC 0.79 ± 0.05 · Brier 0.21
- Trained on 600 verified labels (150 dead / 450 surviving)
- Leakage-guarded: excludes status, discontinued_year, profile_maturity
- Top drivers: tag breadth, signal, vintage, liveness, business model
n=600 is thin → 95% CI on AUC ≈ ±0.10. Treat as a directional signal, not a calibrated probability.
Analyst read
generated from the metrics above · no external narrativeCastle sits in the 1th percentile of Launch Sentinel's 79,105-product universe by signal, ranked #598 of 931 in Security. The mortality model puts its probability of going inactive at 0% — a verified outcome label. It sits on the competitive-alternatives graph with 1 linked rival, so the comparable set below leans on curated edges. Signal momentum is positive (WoW +58.9).
Comparable set
curated alternatives → functional tag-similarity| Company | Basis | Signal | P(dead) |
|---|---|---|---|
| Oktadesign | curated | 30.0 | 1% |
| LaunchKeysecurity | similar | 15.7 | 95% |
| ThinAirsecurity | similar | 15.7 | 84% |
| Digital Signing 2.0 by Attoresfinance | tag-overlap | 15.7 | 34% |
| Fairwinds Insightsdev-tools | tag-overlap | 30.0 | 2% |
| Microsoft Azure| Blockchain as a Service (BaaS)dev-tools | tag-overlap | 15.7 | 43% |
| Release Deliverydev-tools | tag-overlap | 32.9 | 11% |
| Driedev-tools | tag-overlap | 15.7 | 44% |
Confidence reflects source and category coherence: curated = an editorially linked alternative; similar = strong functional tag-overlap in the same category; tag-overlap = shared tags only and may be noisy. Signal & P(dead) are real / model.
Structure
observed attributesFunding & team
not in sourceNo disclosed funding in our sources. Funding data covers only ~1% of the corpus (462 rounds across 79K products), so absence here is unknown, not zero— we don't estimate a number we can't source.