> ## Documentation Index
> Fetch the complete documentation index at: https://mirage.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Reviewing reports

> Approve or reject AI-extracted metrics so only verified data reaches donors

The **Approvals** page is your organization's review queue. This is where the "steward approved" half of Mirage happens: the AI proposes, you decide.

## What happens when a report comes in

When a reporter submits and finalizes a report, Mirage automatically:

1. **Analyzes the media**: audio is transcribed and language-detected (with an English translation when the original is in another language); photos and video get AI-written summaries.
2. **Extracts metric contributions**: for each of the project's metrics, the AI proposes values it found in the answers and evidence, each with a **confidence score** and a link to the exact question or attachment it came from.
3. **Queues everything for review**: contributions arrive as **pending** and count toward nothing until you act.

![Reporting workbench with pending metric contributions](https://mintcdn.com/mirage/eVR3n30GbOeZdkdj/images/steward-reporting-queue.png?fit=max\&auto=format\&n=eVR3n30GbOeZdkdj\&q=85\&s=a171f530f9a2004f59ec404cca799e7a)

The reporting workbench: a finalized field report with its AI-extracted contributions awaiting your approval.

## Review a report

**1. Open the submission**

From the **Approvals** page, open a finalized report. You'll see the reporter's answers, their attached photos, video, and audio (with transcriptions and AI summaries), and their free-form observation notes.

**2. Evaluate the contributions**

Each proposed metric contribution shows the metric, the extracted value, the AI's confidence, and its source. Click through to the source to check the AI's work against the actual evidence.

**3. Approve or reject each one**

* **Approve**: the value rolls up into the project's aggregated metrics and becomes available to donor reports.
* **Reject**: the value is excluded. Use this for misreads, duplicates, or evidence that doesn't hold up.
* Use **Approve all** / **Reject all** to act on every pending contribution for a report at once when they're all clearly good (or clearly not).

Every decision is recorded with who reviewed it and when: your verification audit trail.

## What the reporter sees

Your review is the feedback loop for the field. In the reporter app, each submission shows its status:

| Status | Meaning for the reporter |
| - | - |
| **Draft** | Still being written; not yet submitted. |
| **Pending** | Submitted and awaiting steward review. |
| **Accepted** | At least one of its contributions was approved; their work counted. |
| **Declined** | Its contributions were rejected. |

Consistently reviewing promptly keeps reporters engaged: they can see their work landing.

## Consent flags

Photos and video arrive with the consent metadata reporters captured at upload: whether **face permission** was obtained and whether the media is **internal only**. You don't need to police this manually at review time (donor-facing pages exclude non-consented and internal-only media automatically), but the flags are visible during review so you know what will and won't be publicly usable.

## Photo privacy and safety

Mirage applies several protections to field photos automatically. You can see and adjust them from the photo safety panel on each attachment during review.

* **Minors are blurred automatically.** When Mirage detects a minor's face, public and donor pages serve a blurred copy. The photo stays hidden publicly until that copy is ready.
* **Blur all faces** extends this to every face in the photo, not just minors. Use it for hostile environments or at-risk adults. While the fully blurred copy is being produced, the photo is withheld from public pages.
* **Location metadata is always removed.** Every photo Mirage serves has its GPS and EXIF metadata stripped, whether or not any face is blurred. A photo taken at a sensitive site never reveals its coordinates on a donor or public page. The original file stays in secure storage for verification.

The review surface intentionally shows you the original, unblurred photo: your job is to make the consent call, and you can't do that from a redacted copy. What you see during review is never what the public sees.

> [!TIP]
>
> High-confidence contributions from a reporter with a good track record can usually be approved quickly; spend your review time on low-confidence extractions and unusual values. Confidence scores exist exactly to help you triage.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.