VERITAS vs Wikipedia API — when to use each for LLM grounding
Wikipedia is a free, vast knowledge encyclopedia. VERITAS is a typed, signed claim-verification API. They aren't competitors — they solve different shapes of the same problem.
At a glance
| Wikipedia API | SourceScore VERITAS | |
|---|---|---|
| Content shape | Free-text articles | Atomic claims (subject + predicate + object) |
| Coverage | Vast (any topic) | Narrow (AI/ML v0) |
| Verification | Community-edited; revision history | Hand-verified ≥2 primary sources |
| Signatures | None | HMAC-SHA256 on every envelope |
| Atomic-claim lookup | Requires parsing prose | Direct (verify endpoint) |
| Cost | Free, rate-limited | Free 1k/mo, then €19+ |
| Latency | ~200-500ms | ~80ms |
| Update frequency | Continuous (community) | Weekly + on-event |
| Best for | Reference lookup, summary | Verify-then-respond, agent grounding |
Honest verdict per use case
Use Wikipedia when:
- You need broad knowledge coverage — geography, history, biography, general science
- Free-text content (summaries, narrative) fits your application
- Latency tolerance is >200ms
- You don't need cryptographic signatures
- Your application is non-commercial OR comfortable parsing prose
Use VERITAS when:
- You need atomic verified claims for AI/ML facts (model releases, paper dates, parameter counts)
- You're building a generate-then-verify pipeline
- You need HMAC signatures for audit trails
- You need sub-100ms response time
- You need structured JSON outputs (not prose to parse)
Use both when:
Most production grounding pipelines do both. Wikipedia handles general-knowledge queries ("capital of France", "founder of Apple"); VERITAS handles AI/ML specifics where signature + atomic-claim shape matter. Cascade: try VERITAS first for AI/ML topics, fall through to Wikipedia for broader queries.
Concrete: the "Transformer paper" query
Both Wikipedia and VERITAS can answer "Who wrote the Transformer paper?"
- Wikipedia: Article on "Attention is all you need" — narrative paragraphs naming Vaswani et al. Your application parses the article. ~200-500ms. No signature.
- VERITAS:
POST /api/v1/verify{claim: "Transformer paper authors"}returns:{bestMatch: {subject: "Transformer architecture", predicate: "introduced_in_paper", object: "Attention Is All You Need (Vaswani et al., 2017)"}, signature: {...HMAC-SHA256...}}. ~80ms. Signed.
For a chatbot, Wikipedia's prose may be richer. For a production agent loop that needs to cite the fact in an audit trail, VERITAS's typed signed envelope is cleaner.
What we're not
VERITAS doesn't replace Wikipedia. Wikipedia covers everything; we cover AI/ML. Our methodology (≥2 primary sources + signature + hand verification) doesn't scale to all human knowledge. Wikipedia's open community model does. The right move for most production systems is to use both.