VERO · Estimated Accuracy

VERO Accuracy

This is experimental oracle technology. The estimate below is not a guarantee and should be treated as an evolving operational metric.

Estimated Accuracy (Current)

87.5%

What This Number Means

The estimated accuracy is an informal, best-effort score intended to communicate that the oracle system is still being refined. It is based on internal testing and observed outcomes, and it may change frequently as the underlying models, providers, prompts, and retrieval settings evolve.

Important: The protocol resolves based on what VERO returns on-chain. If the oracle returns a result that seems “incorrect,” the smart contract still treats it as final.

Why Outcomes Can Appear Wrong

Missing or delayed information: sources may not have published the relevant data yet, or indexing may lag.
Ambiguous statements: unclear terms, missing timezones, or undefined thresholds can force a guess when VOID would be preferable.
Retrieval limitations: limited search breadth/time, source quality variance, and contradictory reporting can mislead evaluation.
Model behavior: even with a strict prompt, models can misinterpret evidence or overfit to noisy results.
Adversarial content: retrieved text can include prompt injection attempts; the system treats it as untrusted, but risk is never zero.

Why “VOID” Exists

The oracle is instructed to return VOID when the statement cannot be reliably determined. This is designed to minimize resolution failures and to discourage exploitation of ambiguous edge cases. Even so, some borderline statements may still resolve YES/NO.

Improvement Over Time

The DAO can iteratively improve the oracle configuration (providers, retrieval settings, prompts, and model choices). Long-term, the goal is to push reliability dramatically higher; you may see this referred to as an aspirational “seven nines” target (99.99999%)—not a promise, but a direction.