Each brand claim is classified into one of four discrete verdicts based on its alignment between memory and research, tracking underlying provenance labels while preserving manual human rulings from automated overwrites.
flowchart TD
F[Research Finding] --> M{Matches Stored Claim?}
M -- Yes --> P{Parametric Recall?}
M -- No --> V4[Verdict: gap / new]
P -- Yes --> C{Confirms or Contradicts?}
P -- No: Grounded-Only --> V3[Verdict: gap + reground]
C -- Confirms --> V1[Verdict: correct]
C -- Contradicts --> V2[Verdict: incorrect]
H[Human Review] -. Overrides .-> V5[State: is_human True / Locked]
V1 -.-> V5
V2 -.-> V5
V3 -.-> V5
V4 -.-> V5Referenced by