Response Fields
The main fields in the judgment that POST /detect returns.
| Field | Type | Description |
|---|---|---|
is_profane | bool | Whether the text is offensive (final call) |
label | string | 욕설 (profane) or 정상 (clean) |
decision | string | block · review · allow |
decision_label | string | Korean label for decision |
score | float | Final score 0–1 (compared against the threshold) |
severity | int | Strength, 0 (none) to 3 (strong) |
severity_label | string | 없음 · 약함 · 중간 · 강함 |
matched | string[] | Dictionary expressions that matched |
reason | string | Short summary of the reasoning |
tier | string | Tier of this request (free / paid) |
load | float | Recent load of that tier (%) |
decision vs. is_profane
These two values reflect different perspectives.
is_profane— a two-way call: is the score at or above the threshold?decision— a gate that splits the score into three bands:block/review/allow
review is the borderline band — "ambiguous enough that a human might want a second look." Rather than auto-blocking, you might route these into a review queue in your service.
The actual boundary thresholds are available in real time under
decision_boundsin theGET /healthresponse.