Understand recommendations and scores

Relevance and editorial value answer different questions

Relevance describes how well a passage fits the request or topic. Editorial-value signals help rank candidates for the intended cut. Neither replaces listening to the passage or checking the full context.

A strong match can still begin mid-thought, omit a qualification or need a different speaker boundary. Review the actual source before approving it for translation.

Use scores as review aids

• Compare candidates for the same task rather than interpreting different score types as one common scale.
• Read the recommendation and play the surrounding context.
• Check complete statements, topic fit, repetition and natural boundaries.
• Change the brief or selection when the suggested moment does not serve your audience.

What a score does not establish

A score does not verify factual claims, forecast views or conversion, or guarantee a good translation. Ranking behavior can change with the model and app version. This guide deliberately avoids presenting internal weighting formulas as a stable public API.

Related guides

• Extract topics and choose highlights
• Make and refine an AI rough cut