> For the complete documentation index, see [llms.txt](https://examind.gitbook.io/v1/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://examind.gitbook.io/v1/feedback-machines/review-adjust-and-approve.md).

# Review, Adjust & Approve Results

Review a machine's evaluations on the Results page, adjust the criteria from real submissions, and approve the grades your students see.

The **Results** page is where you check the machine's work before it becomes a grade. You can see the whole class at once, drill into any single submission, correct anything you'd have graded differently, and then approve — which is what releases the final result to the student.

Open a machine's **⋮** menu and choose **Results**.

{% hint style="info" %}
Results is available to anyone who can manage the machine — its owner, and instructors in its class. It's no longer enabled per institution.
{% endhint %}

## See the whole class

The Results page opens on a **Heatmap**: one row per student, one column per rubric part, so you can read the class by part rather than one submission at a time. A part where most of the class scored low is usually a problem with the criterion or the assignment — not with thirty students.

Results grades a set of submissions: your students' submissions, one per student, or a [bulk import](/v1/feedback-machines/bulk-grading-assist.md). Submissions you run through the machine yourself from **Use** are for checking the machine and don't appear here — find them under **Submissions**.

* **Heatmap / Overview** — switch between the grid and an **Overview** of the class as a whole: how many are graded, the average, median, standard deviation and range, the score distribution, and a **By part** list ordered weakest first, with each part's questions and their class averages beneath it. Selecting a part there takes you back to the Heatmap sorted by that part.
* **Scope** — choose which submissions to read: the class, or a particular [bulk import](/v1/feedback-machines/bulk-grading-assist.md) set.
* **Display** — show each cell as **Score**, **Percent**, or **Score / Total**.
* Select a part's header to sort by that part, lowest scorers first.

Selecting a part's cell expands a drill-down in place, so you can read that part's detail — and edit its summary — without leaving the class view.

If the page is empty, the machine has nothing to show results for yet. A machine that isn't part of a class has nothing to grade: [put it in a class](/v1/feedback-machines/use.md#put-the-machine-in-a-class), then have students submit to it or [import submissions](/v1/feedback-machines/bulk-grading-assist.md). Submissions made before the machine joined the class stay outside it.

### Final submissions awaiting review

When the machine holds anything until you approve it — results set to **After review & approval**, or grades sent to EXAMIND **When approved** (see [What Students See](/v1/feedback-machines/use/what-students-see.md)) — the Heatmap shows a **N awaiting review** count of the graded submissions you haven't approved yet. Those students are waiting on you, so this count is a queue to work through, not a statistic.

## Review one submission

Select a student's name to open their submission in a three-pane workspace:

* **Submission** — their original work.
* **Evaluation** — the rubric: each part with its score and written summary, and (when you expand a part) the criterion-by-criterion evaluation with the machine's reasoning.
* **Adjust** — a chat for correcting this evaluation.

Selecting a criterion's locate control scrolls the Submission pane to the exact spot in the student's work that the criterion refers to, so you can check the machine's reasoning against the evidence instead of taking it on faith.

{% hint style="info" %}
A chip in the Evaluation pane header tells you whether the criterion-level detail you're reading — **Detail visible to students** or **Detail hidden from students** — actually reaches the student, so you know where your review effort lands. See [What Students See](/v1/feedback-machines/use/what-students-see.md).
{% endhint %}

## Adjust the machine from real submissions

**Adjust** and the [Modify panel](/v1/feedback-machines/build-a-feedback-machine/modifying-a-feedback-machine.md) both change the underlying machine and how it grades from then on. The main difference is that the Adjust interface has access to student submissions and evaluations to help inform the changes and test their effect, whereas the Modify interface works from your written direction only.

That access is what makes a nuanced correction possible. You can say:

> Alice should be scoring higher for concept mastery and Joel should be scoring lower — they were told to do X in our lecture.

and the machine examines what Alice and Joel actually wrote, works out what distinguishes them, and updates the criteria to reflect your intent.

### How a change lands

{% stepper %}
{% step %}

### Point the AI at the evidence

Pull the specific criteria you're talking about into the chat, using the wand controls (🪄) in the Evaluation pane:

* **A whole rubric part** — hover over the part's heading and a small wand button appears beside its name. Select it.
* **A single question** — each question row has an **Adjust** button, marked with a wand, at the right-hand end.

Each one you select is added to the chat as a reference chip. Add as many as you need: several criteria at once, or the same criterion across several students.
{% endstep %}

{% step %}

### Describe the change

Explain what should be different and why, in plain language. Context the machine can't infer — what you told the class in lecture, how you weigh a trade-off — is exactly what's worth saying.
{% endstep %}

{% step %}

### Review the preview

The machine proposes updated evaluations and shows them to you before anything is committed. Compare the proposed result against the current one.
{% endstep %}

{% step %}

### Accept, and it re-grades the class

Accepting commits the change to the machine and re-evaluates the affected criteria across every submission in the set — not just the ones you referenced. Criteria the change didn't touch are left alone. You can revert a committed change from its history.
{% endstep %}
{% endstepper %}

{% hint style="warning" %}
**Criteria apply to the whole class.** A Feedback Machine has one set of criteria, so an adjustment changes grading for everyone. You cannot currently hold one student to different criteria than another — and you wouldn't want to: the point of adjusting from real submissions is to make the *shared* criteria express your intent more precisely.
{% endhint %}

### Edit the words students will read

Two pieces of the evaluation are written prose you can rewrite directly:

* **Key takeaways** — the lessons the student should carry forward, shown with their final results. Select **Edit**, and your version is what students see. It's attributed to you in the student's view.
* **Part summaries** — the short paragraph under each rubric part that justifies its score. Edit any of them the same way.

Your edits are yours: they survive re-evaluation, an applied Adjust change, and a revert — the machine never overwrites them. The edited text is also what leaves the system, so it's what appears in the [detailed CSV and Canvas feedback exports](/v1/feedback-machines/bulk-grading-assist.md) and in the results sent back to a linked EXAMIND assessment.

{% hint style="success" %}
**Part summaries are the reviewable-sized justification.** Reading every criterion row for every student doesn't scale; reading one summary per part does. That's why summaries reach students even when the detailed breakdown is turned off — the grade stays defensible without publishing all of the reasoning.
{% endhint %}

## Approve

Approving is the act that makes a result final.

{% stepper %}
{% step %}

### Approve a submission

In the review workspace, select **Approve**. The control shows its progress while the approval saves, then reads **Approved**, and the workspace opens the next submission still awaiting approval — so you can work the queue without going back to the class view each time. On the last one awaiting approval you stay where you are.

To reverse an approval, open the **Approved** control and choose **Un-approve**. It takes a deliberate second choice rather than a second click, so moving quickly through a class can't undo a decision you just made.
{% endstep %}

{% step %}

### Or approve in bulk

From the Heatmap, select the students you want — then **Approve** (or **Unapprove**) the whole selection at once. If the batch doesn't go through, your selection is kept so you can retry without picking the students again.
{% endstep %}

{% step %}

### The student's view flips

An approved result carries **Reviewed and approved by** your name and the date, so students see a human decision rather than a machine output. If the machine's feedback access is **Until approved**, the formative feedback on that submission gives way to the final results.
{% endstep %}
{% endstepper %}

Un-approving reverts the student's view in Feedback Machines immediately, so a mistaken approval is not permanent.

{% hint style="warning" %}
Approving does **not** reveal results if the machine's Summative Evaluation View access is set to **Never** — in that case approval is for your records and your LMS. See [What Students See](/v1/feedback-machines/use/what-students-see.md).
{% endhint %}

### Approval and a linked EXAMIND assessment

When the machine is linked to an EXAMIND assessment, the grade travels back to that assessment — the score, plus the key takeaways and part summaries as you edited them — and on to your LMS gradebook if the assessment [posts grades](/v1/examind-platform/get-started/lms-integrated/post-grades.md). The machine's [**Send grades to EXAMIND**](/v1/feedback-machines/use/what-students-see.md#send-grades-to-examind) setting decides when: on your approval, or as soon as each submission is evaluated. Delivery runs in the background, so a grade lands in EXAMIND a moment after you approve — and a whole selection approved at once is delivered the same way.

A machine set to send on approval holds the grades of everything it has evaluated but you haven't approved. Switching that setting to **When evaluated** sends all of them at once, after asking you to confirm — see [Send grades to EXAMIND](/v1/feedback-machines/use/what-students-see.md#send-grades-to-examind).

{% hint style="warning" %}
**Un-approving doesn't withdraw a grade from EXAMIND.** Once a grade has reached the linked assessment it stays there; un-approving in Feedback Machines doesn't remove it, and your next approval replaces it.
{% endhint %}

## How much reviewing is enough?

That depends on how well the machine already matches your judgment, and it changes over time. There are two established workflows:

<table data-view="cards"><thead><tr><th></th><th></th><th data-hidden data-card-target data-type="content-ref"></th></tr></thead><tbody><tr><td><strong>Human-in-the-loop</strong></td><td>Review every submission and add your own comments. Use it to align a new machine, and where a personal touch matters.</td><td><a href="/v1/feedback-machines/review-adjust-and-approve/human-in-the-loop.md">Human-in-the-loop</a></td></tr><tr><td><strong>Human-on-the-loop</strong></td><td>Spot-check and look for class-wide patterns. Use it once a machine has proven itself aligned.</td><td><a href="/v1/feedback-machines/review-adjust-and-approve/human-on-the-loop.md">Human-on-the-loop</a></td></tr></tbody></table>

## Grading a whole class from a batch of files

If your students' work isn't in Feedback Machines yet — you have a folder of files, or a Canvas submissions download — start with [Bulk Grading Assist](/v1/feedback-machines/bulk-grading-assist.md) to create the submissions, then come back here to review and approve them.