How scoring works
AI multi-reviewer scoring uses six dimensions: issue alignment 20%, trust-frame fit 20%, objection handling 25%, openness activation 15%, clarity/register fit 10%, and respect/backlash safety 10%.
Evaluation panels score independently, compare supporting evidence, and surface disagreement or a no-clear-winner result for human review.
Scores are decision-support rubric results, not a predicted probability of audience behavior.
All scores use a 0–100 scale. Higher is better for receptivity, trust fit, and objection coverage; lower is better for objection risk.
Score bands: 75–100 Strong fit; 55–74 Usable with edits; 0–54 Needs refinement.
Legacy heuristic scoring starts from a 38-point baseline, then adds up to 25 points (25% trust fit), 20 points (20% objection coverage), and 17 points (17% topical relevance). It subtracts 4.5 points for each uncovered objection, up to three.
The legacy score is a deterministic text-overlap aid, not a predicted probability of audience behavior.
Saved tests
Compare prior runs and reopen a message track.