How we verify
SafeYard cards are extracted from OEM Emergency Response Guides and cross-checked against secondary safety publications. Below is exactly how we work and where the limits are. Read this once.
Sources we use
Every source we cite is tagged with a tier reflecting its authority. Higher tier = stronger evidence.
- Tier 1 — Authoritative. OEM official documents (Emergency Response Guides, Technical Service Bulletins, Service Information Systems) plus federal regulatory sources (NHTSA, FEMA U.S. Fire Administration, NTSB investigation reports, DOE / NREL national-lab technical reports).
- Tier 2 — Industry / safety bodies. NFPA Alternative Fuel Vehicle Safety Training, EV FireSafe research, IIHS crash test reports, OARA / ARA / ARC dismantling guides, SAE published battery safety standards (J2929 / J2950 / J3147), peer-reviewed academic papers.
- Tier 3 — Cross-reference. Trade publications, ERG aggregators (Energy Security Agency, Boron Extrication), spec databases (EV Database, EVSpecifications), Wikipedia. We only use these to cross-check non-safety facts like voltage class or year ranges.
When you see a chip on a field, hover or tap it — the tooltip breaks down how many Tier-1, Tier-2, and Tier-3 sources cite that specific value. A “Confirmed” field is strongest when at least two of its citing sources are Tier 1 or 2. Agreement among Tier-3 sources alone is weaker evidence because trade publications and aggregators often reprint the same upstream figure.
Per-field confidence labels
Every card carries a verification banner at the top and per-field confidence tags inline. The labels mean:
- Confirmed2 or more independent sources agree on this value. Most-trusted tier.
- 1 sourceOnly one source documents this field. Often correct, but no cross-check. Verify against the linked OEM ERG before relying on it.
- DisputedTwo or more sources disagree. The card shows the more conservative value. Read the field note for the conflict.
- VerifyNo source confirmed in our extraction pass. The card uses an industry default or a TODO placeholder. Treat as unverified.
Verified vs Draft cards
A card is Verified when every tracked field carries the “Confirmed” label. Otherwise it's a Draft and the banner names exactly how many fields still need review. We show Draft cards anyway — partial reference is more useful than no reference — but you should treat the verification banner and the inline chips as load-bearing, not decorative.
Honest limits
- Source agreement is not proof. Three sources quoting the same incorrect industry rumor will all agree. Cross-source agreement is evidence, not certainty.
- AI-assisted extraction has blind spots. Our pipeline uses LLMs to parse PDFs and compare values. They make mistakes, especially on diagram-heavy or model-year-specific content.
- Safety-critical info may live only in OEM bulletins that aren't publicly indexed. Where that's the case we cannot triangulate.
- No human dismantler has reviewed cards yet. A named expert review will be added once we recruit one. Until then, the verification status shown is the entire trust signal — there is no additional human checkpoint behind it.
If you spot a mistake
Every card has a “Something wrong?” link at the bottom that opens a prefilled correction form. We treat user corrections as a source: a second confirming user report can promote a field from “1 source” to “Confirmed” (and we'll note it in the changelog).