Same brief, two voices: what the measurement actually sees

A brief does not identify a voice

Give two writers the same topic, format and length: both can follow every instruction and still produce recognisably different texts. One opens with a scene, the other with a claim. One decides quickly; the other builds nuance. Those differences are voice habits, not briefing errors.

That is why Kaneme does not compare a text with an abstract idea of “good style”. It compares it with a reference corpus belonging to one specific person.

What becomes measurable

The measurement observes patterns that survive a change of subject. No single feature is enough; their combination creates a stronger fingerprint.

• sentence length and sequencing;
• the way a writer opens, pivots and concludes;
• degrees of certainty, nuance and confrontation;
• recurring linking words, images and concrete detail;
• the distance a writer keeps from the reader.

A verdict, not an intuition dressed as a score

A number is useful only when its question is clear. Kaneme separates comparison with generic AI from comparison with another human. When the signal is too weak or a text falls outside the learned register, the right result is abstention rather than forced certainty.

A useful verdict also points to the next action: sharpen the point of view, restore a concrete detail, recover the rhythm, or accept that this text deliberately explores another voice.