Numeric answer evals for finance bots should use:

Question

Accepted Answer

Tolerance-aware numeric comparison plus unit/currency normalization. — Finance numeric evals need tolerances and unit/currency normalization — not brittle string match.

Answer

Exact string match only including formatting noise.

Answer

No numeric checks.

Answer

Only BLEU scores.

Numeric answer evals for finance bots should use:

Short Explanation and Infographic

Full Explanation