VP Trustworthiness Index

Offline

VP Trustworthiness Index

Visual TI adapted from the MITCF trustworthiness scoring module. Four dimensions: hallucination, bias (N/A for POPE/MME), robustness, and decision reliability.

No VP runs found.

Launch evaluation

TI Formula (visual adaptation)

TI = w_h x (1 - hallucination_rate)
   + w_b x ti_bias        (1.0 when bias not applicable)
   + w_r x selective_acc  (accuracy on non-abstained items)
   + w_d x decision_reliability
Default weights: 0.25 each