Validation · v1 → v2
The algorithm was stress-tested against adversarial cases. v2 introduces the two model-level modifiers and re-weights AI Visibility upward — closing the gap to the estimated fair score on all three benchmark cases.
S1 · Chain with legacy halo
Large chain with massive review volume but mediocre AI signals. v1 over-rewarded sheer review count. v2 caps volume contribution and removes the implicit halo.
Fix: Volume bucketing + AI Visibility re-weighted to 25 pts.
S2 · Established local
Genuinely strong independent business under-scored by v1 because legacy authority wasn't captured. v2's Legacy Institution Boost closes the gap.
Fix: Legacy boost (+15) on 40+ year businesses with press/Wikipedia.
S3 · Fake review farm
v1 scored a clear gaming pattern as Elite. v2's Gaming Flag now catches cross-platform divergence and unnatural review velocity.
Fix: Gaming flag (−20) on divergence ≥1★ or rapid 4.8★+ inflation.