# Handling Results

> How to act on a moderation verdict: block, blur, queue for review, and tune thresholds for your content.

> For AI agents: the complete documentation index is at https://nosmai.com/llms/moderations.txt

Product: Nosmai Moderation
Group: core-concepts
Source: https://nosmai.com/docs/moderations/handling-results/

The SDK gives you a verdict; your app decides what to do with it. This guide covers the common patterns.

## Read the verdict

Every visual result carries a single `isUnsafe` flag plus the detail you need to decide:

- `isUnsafe` is true when an object is flagged or NSFW is `block`. Use it for a simple allow / reject.
- `nsfw` (`safe`, `warn`, `block`) lets you treat borderline content differently from explicit content.
- `detections` and `nsfwScores` give you the specifics for logging, messaging, or tuning.

Text results use `blocked`, plus `category` and `layer` for context.

## Choose an action

There is no single right response. Match the action to your product and the verdict.

| Verdict | Typical action |
| --- | --- |
| `block` / `isUnsafe` | Reject the upload or message, or replace it with a placeholder |
| `warn` (suggestive) | Blur, add a content warning, or age-gate. Advisory, so many apps allow it |
| Uncertain / borderline | Allow but queue for human review |
| `safe` | Allow |

## By surface

**Image upload.** Check before the image is stored or shown to others. Reject or blur on `isUnsafe`.

```text
const r = await analyzeImage(path)
if (r.isUnsafe) rejectUpload()
else if (r.nsfw === 'warn') blurUntilTapped()
else allow()
```

**Chat message.** Check before the message is sent or displayed. Block on `blocked`, and use `category` for the user-facing reason.

```text
const t = await moderateText(message)
if (t.blocked) showBlockedNotice(t.category)
else send(message)
```

**Live camera.** Handle the per-frame result. Because a single frame can flicker, act on the debounced verdict (the SDK already holds it) rather than every raw frame, and remember that a sustained `warn` escalates to `block`.

## Add a human review step

Automated moderation is the fast first pass; it is not a full trust-and-safety system. For the gray area, keep the sample and route it to your own review queue instead of making a hard decision. A good split:

- Clear `block`: act automatically.
- `warn` or borderline scores: allow with a soft action (blur, warn) and flag for review.
- Clear `safe`: allow.

The SDK does not include a review console or user-reporting flow, so pair it with your own reporting and review tooling.

## Tune, do not hardcode

If you see misses or false positives for your content, adjust the thresholds rather than adding ad-hoc rules around the result. Lower is stricter.

```text
setNsfwThreshold(explicit, 0.45)   // block bar
setNsfwThreshold(sexy, 0.55)       // warn bar
setThreshold(weapon, 0.70)         // object class bar
```

> [!TIP]
Start from the defaults and change one bar at a time. See [How moderation works](/docs/moderations/concepts/) for what each field means.