Scoring

Every check returns an alignment score from 0 to 100 and 2–3 short deviations.

Score

The model returns a model score based on a rubric: tone alignment (0–30), structure (0–25), lexical alignment (0–20), clarity and precision (0–15), and context fit (0–10). The server then applies deterministic penalties: emojis (-15), exclamation marks (-5), banned terms (-10 each, cap -30), sentence length far above norm (-10). The final score is clamped to 0–100.

Deviations

Deviations are 2–3 one-line explanations of what was off. They reference evidence (e.g. "Tone is more casual than client standard" or "Sentence length exceeds norm"). They are short and concrete, not long coaching paragraphs.

Signals

The optional signals drawer can show sentence length, hedging rate, and banned phrase hits. These are stored with each run for audit and tuning.