Risk score return value

[SamuelJanoska]

Hi,

I'd like to ask if the current logic for returning risk score values is going to stay that way, or if there are any plans to change. I noticed that if the prompt is evaluated as valid, score is set to zero, which I found misleading. I saw this in the BanTopics scanner, but it's true for multiple scanners.

[rlangenfeld-twin]

Going to bump this we are encountering the same issue with version 3.15.

def scan(self, prompt: str) -> tuple[str, bool, float]:
        if prompt.strip() == "":
            return prompt, True, 0.0

        output_model = self._classifier(prompt, self._topics, multi_label=False)
        label_score = dict(zip(output_model["labels"], output_model["scores"]))

        max_score = round(max(output_model["scores"]) if output_model["scores"] else 0, 2)
        if max_score > self._threshold:
            LOGGER.warning(
                "Topics detected for the prompt",
                scores=label_score,
            )

            return prompt, False, calculate_risk_score(max_score, self._threshold)

        LOGGER.debug(
            "No banned topics detected",
            scores=label_score,
        )

        return prompt, True, 0.0
        

This method will always returns 0.0 which makes it harder for us to know how close certain topics where to being above the threshold (in our usecase this would be extremely helpful).

I think it would be a fairly straightforward change to return the calculated_risk_score on the last line instead of just 0.0