Q1. How did you arrive at 22.9% precision?
The 22.9% is an out-of-sample figure — the only number we present as a performance metric. The model was trained on NHTSA complaint and TSB data from 2020–2023, then tested on a completely separate period: 2024–2026. Of the vehicles flagged HIGH or MEDIUM (CRI ≥ 25), 22.9% proceeded to an official NHTSA recall within the test window. The naive baseline is approximately 3–5%. RecallVantage is roughly 5x better than random.
Q2. What will the precision number be tomorrow?
The 22.9% is a stable, validated figure — not a daily fluctuation. The model runs daily, but the precision metric is recalculated on a rolling 12-month basis as new recalls are confirmed. A Bayesian learning layer (v1.0, in development) will update posterior probabilities with each confirmed recall, expected to improve precision incrementally over time.
Q3. How early do I see the signal before a recall is announced?
Median lead time: 85 days — roughly three months before the official NHTSA announcement. Flagship validated case: Nissan Rogue (Electrical System) — signal appeared 252 days before recall. Component-specific windows: ADAS ~30 days · Electrical ~75 days · Brakes/Engine ~90 days · Fire ~120 days.
Q4. What is the CRI score and how is it calculated?
The Catastrophic Recall Index (CRI) is a 0–100 score per vehicle (Make/Model/Year/Component). It is a weighted combination of: Safety TSB count (40%) · NHTSA complaint volume (35%) · Reactive signal — complaints preceded TSBs (15%) · Generic penalty — penalizes bulletins applied across 5+ models (-10%). Only TSBs with critical keywords are counted: FIRE, STALL, BRAKE, LOSS OF CONTROL, THERMAL RUNAWAY, AIRBAG, CRASH, INJURY, DEATH.
Q5. What are the confidence tiers?
HIGH (CRI ≥ 40): Strong statistical anomaly, multiple component convergence.
MEDIUM (CRI ≥ 25): Elevated trajectory, partial convergence.
SPECULATIVE (CRI < 25): Early-stage signal, limited confirmation.
Only HIGH and MEDIUM signals are delivered to clients.
Q6. What data sources feed the system?
Three primary NHTSA sources: (1) Complaints — 1.8M+ records, 1995–2026, updated daily. (2) Technical Service Bulletins — 2020–2026, safety-filtered. (3) Recall Database — 2010–2026, used for validation only. SEC EDGAR warranty reserve data is documented as a competitive moat — standard XBRL tags return zero for all major OEMs, requiring proprietary extraction (in development).
Q7. How do you avoid false positives?
Three filters in combination: (1) Minimum thresholds — at least 3 safety TSBs and 2 complaints required. (2) Generic TSB filter — TSBs applied to 5+ models are flagged as unreliable. (3) Component specificity — CRI is computed at the Make/Model/Year/Component level, not manufacturer level. At CRI ≥ 40, false positive rate is less than 20%.
Q8. How is this different from Bloomberg or existing risk tools?
Bloomberg and credit risk tools use public recall announcements — reactive by definition. By the time a recall appears on Bloomberg, the warranty accrual has already hit the balance sheet. RecallVantage reads the NHTSA complaint database and TSB registry — the pre-announcement signal layer. The information is public, but unstructured, high-volume, and not systematically read by the market.
The recall is not the event. The recall is the confirmation.
Q9. What is the expected financial exposure per signal?
RecallVantage runs 10,000-iteration Monte Carlo simulations per vehicle using CRI score, component type, and historical recall cost distributions. Output includes Expected Loss (median), VaR at 95%, and Expected Shortfall (tail risk). Example: a HIGH-tier electrical signal on a major OEM platform carries expected loss of $180M–$2.4B depending on affected vehicle count and remedy type.
Q10. Is the 22.9% the in-sample accuracy?
No — this is a critical distinction.
In-sample (structural validation): 94.6% — confirms the model correctly identifies component patterns within the training period. This is NEVER cited as a performance metric.
Out-of-sample (predictive precision): 22.9% — measured on data the model never saw during training. Citing 94.6% as predictive accuracy would be misleading and professionally indefensible.
Q11. What is your current coverage universe?
4 major OEMs actively monitored: Ford, GM, Stellantis, Toyota. Approximately 2,771 active vehicle-component signals in the current cycle. International expansion in pipeline: Transport Canada (9,211 records ready) and SAMR China (1,336 records ready) — not yet integrated.
Q12. How do I access the signals?
Tier 1 (Standard): CRI score, confidence tier, expected loss — all clients.
Tier 2 (Premium, NDA required): Component breakdown, lag analysis, alpha signals, Monte Carlo output.
Access by application only. Current cohort: selective Design Partners across High Yield, OEM, and Insurance.
RecallVantage is a Decision Support System (DSS). Not investment advice. All signals are analytical outputs intended to supplement, not replace, independent professional judgment.
© 2026 RecallVantage Intelligence · Petah Tikva, Israel · For authorized distribution only
