1. Choose an indicator
2. Provide evidence
Calibrated on a small handful of historical findings (few-shot examples, not a trained model). Treat as an illustrative style nudge, not a statistically validated match to house style — use "Compare styles" to judge for yourself on a given document.
Assessment result
Close-out points
Style comparison
Same document, same indicator, run twice — once with house-style exemplars, once without. Compare the rationale and close-out phrasing; the determinations themselves may also differ, since calibration can shift borderline judgement calls.
Without house style
With house style
Traceability
Identical input yields identical output. This engine is deterministic (temperature=0) and every run is logged with exactly what produced it — use "Run again" above to verify live.