Every verification runs through three independent pillars:
Deterministic: Math, dates, and constants are computed — never guessed. Hallucination is impossible for these answers.
Retrieval: Multiple models research independently with enforced source diversity. They cannot recycle each other's sources.
Adversarial: A dedicated model's only job is to find reasons the consensus is wrong.
The result is a confidence-scored, source-verified answer — not another chatbot response.
How Veracore Works
Your question is classified by risk level (Tier 1–4)
Deterministic math and date queries are answered without any LLM
Higher-risk queries run the full 3-pillar pipeline
An adversarial model actively hunts for errors in the consensus
A cross-evaluation layer scores each model's accuracy
A final consensus engine produces a confidence-scored answer
Beta Notice
Veracore is in public beta. This means:
Capacity is limited — you may see a "server busy" message during peak times
Daily verification limits apply per IP address
Results depend on the underlying AI models — confidence scores reflect uncertainty, not guaranteed truth
No user accounts exist yet — tier access uses a header token system for beta testers
Limits & Usage
Questions are limited to 250 characters
20-second cooldown between verifications
Daily caps reset at midnight UTC
Save Result stores verifications on your device only — no server, no account, no uploads
Share Result copies a plain-text summary to your clipboard — nothing is sent anywhere
Subscriptions increase your daily limit. They do not change the engine or guarantee outcomes.
Veracore is not a substitute for professional medical, legal, or financial advice.
About The Good Neighbor Guard™
Veracore is built by The Good Neighbor Guard™, an independent AI safety company focused on protecting people from misinformation — especially in medical, legal, and financial contexts.
Veracore checks AI responses using multiple models, adversarial analysis, and deterministic computation — so you can see when an answer is actually reliable.
Pillar 01 — Deterministic
Compute, Don't Ask
Math, dates, and constants bypass AI entirely. A calculator cannot hallucinate.
Pillar 02 — Retrieval
Source Diversity Enforced
Multiple models research independently. Source overlap is prevented at the orchestration layer.
Pillar 03 — Adversarial
Structured Skepticism
A dedicated model's only job is to find reasons the consensus is wrong.