> For the complete documentation index, see [llms.txt](https://help.timelygrader.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.timelygrader.ai/user-guides-get-started-here/grading-settings.md).
# Calibrating grading suggestions
What happens if you think AI is wrong? The controls in the Grading Settings page can help you influence how TimelyGrader interprets your rubric.
## Before you start
* Make sure you've already looked a few processed submissions. You will need to know where the gaps are before trying to address them with these controls.
* Make sure Grading Suggestions are turned on.
* If you are trying to change the structure of the feedback, go see our article on feedback structure. If you are trying to change the content of the feedback, you are in the right place.
## Steps
{% stepper %}
{% step %}
### Make sure Grading Suggestions are turned on.
Your institution may have turned off Grading Suggestions to follow institutional policies or guidance on AI use. If this setting is disabled at the institutional level, you cannot turn it back on yourself. Contact your administrator to discuss whether it can be enabled.
{% endstep %}
{% step %}
### Add general instructions
General grading instructions apply across the whole rubric. Use them to explain expectations that should guide the entire assessment, such as what to always address, what to avoid commenting on, or how to approach borderline performance.
For example:
“When identifying a weakness, explain what the student could do to improve.”
“Do not comment on grammar or spelling unless it affects the meaning or is explicitly assessed by the rubric.”
“Give credit for a valid alternative approach when the student explains and supports their reasoning.”
Be specific about what “generous” or “strict” means. Instead of “Be generous,” explain where flexibility is appropriate: “Accept equivalent terminology when the student demonstrates an accurate understanding of the concept.”
Keep these instructions consistent with your rubric. If an instruction applies to only one criterion, place it under that criterion rather than applying it to the whole assessment.
{% endstep %}
{% step %}
### Criterion specific instructions: look-fors
Look-fors describe the evidence that would demonstrate successful performance within a criterion. Ask yourself: What would I read—or, for video, see or hear—if the student had done this well?
Add one observable statement per look-for. For example:
“Names the alternative they rejected and explains why.”
“Connects the recommendation to evidence from the case.”
“Explains the purpose of each step while demonstrating the procedure.”
Avoid broad statements such as “Shows critical thinking.” Describe the action that demonstrates it, such as comparing alternatives, questioning an assumption, or explaining a trade-off.
Use look-fors to clarify the rubric, keeping them aligned with the assignment instructions students received. They should make existing expectations easier to recognize rather than introduce additional requirements.
{% endstep %}
{% step %}
### Criterion specific instructions: violations and omissions
Violations and omissions describe concrete problems the AI should look for within each criterion.
A violation is something present in the work that conflicts with an expectation. An omission is something required that is missing. For example:
Violation: “Interprets a correlation as proof that one variable caused the other.”
Omission: “Provides a recommendation without explaining the reasoning behind it.”
Omission: “Does not address the required counterargument.”
Explicit descriptions help direct attention to missing requirements as well as problems in what the student submitted.
Be careful to distinguish an omission from an acceptable variation. A student may demonstrate the required understanding using different wording, a different example, or another valid approach.
For video, also consider whether the evidence can reasonably be seen or heard. Something that is not visible in the recording is not necessarily something the student failed to do.
{% hint style="success" %} **Hugo’s tip:** “Suggested from your rubric wording” offers look-fors drawn from your rating descriptions. Accept only those that match your intended expectations, and edit any that are too broad or introduce a requirement you did not mean to assess.
{% endhint %}
{% endstep %}
{% step %}
### Adding custom instructions for criterions
Beyond look-fors and violations, use criterion-specific instructions to explain how the AI should interpret evidence for that particular criterion.
These instructions are useful when a criterion requires context, an exception, or a distinction that does not fit neatly into a positive or negative marker.
For example:
“For this criterion, focus on the reasoning behind the recommendation. More than one recommendation may be defensible.”
“Assess the accuracy of the explanation separately from the student’s use of technical vocabulary.”
“When discussing calculation errors, distinguish between an incorrect method and an arithmetic mistake within an otherwise appropriate method.”
Avoid repeating the same instruction in several places. Keep assessment-wide guidance in general instructions and local guidance under the relevant criterion. If two instructions conflict, revise them so the intended expectation is clear.
{% endstep %}
{% step %}
### Common issues that can be addressed
* **The AI keeps missing the same requirement:** Check that it is stated explicitly under the relevant criterion. Replace broad language with a clear description of the evidence or omission.
* **The feedback comments on irrelevant issues:** Clarify the assessment’s scope in general instructions, including what should be excluded.
* **The suggestions are too strict or too generous:** Identify which judgment is inappropriate and explain how it should be handled. A specific example is more useful than simply asking for greater strictness or generosity.
* **The suggestions remain inconsistent:** Check for conflicts between the rubric, general instructions, and criterion-specific instructions. Simplify overlapping guidance and test one change at a time.
After making changes, review new suggestions across a few submissions with different levels of performance. A change that helps with one submission may need adjustment to work consistently across the class.
{% endstep %}
{% endstepper %}
## Troubleshooting
The AI keeps raising something I don’t care about.
Mute it, or add it to the general instructions as something to never comment on. Settings are per assignment.
My look-fors seem to be ignored.
Check the master switch. With suggestions off, the AI reads nothing and you grade from the rubric alone.
Flags are vague.
Add violations to that criterion. A concrete negative marker gives the AI something specific to find.
Grading feels too harsh or too lenient.
That’s the rubric, not these settings. Nothing here decides a score.
## FAQ
Do these settings change the grade?
No. They decide what evidence reaches you. You decide the band.
How specific should a look-for be?
Specific enough to check. “Names a company and a date” is checkable; “shows good analysis” isn’t.
Are these the same settings oral activities use?
Yes. Oral Quiz and Oral Defense have the same Grading Settings tab, with a few extra options for Oral Defense.
## Related articles
* [Building rubrics that grade consistently](/user-guides-get-started-here/rubrics.md)
* [Grading a submission](/canvas-support-articles/assignments/grade-submissions.md)
* [Grading and feedback settings for oral activities](/canvas-support-articles/tutorials-oral-assessments-canvas/oral-grading-feedback.md)
**Hugo’s tip:** “Suggested from your rubric wording” offers look-fors drawn from your rating descriptions. Accept only those that match your intended expectations, and edit any that are too broad or introduce a requirement you did not mean to assess.
{% endhint %}
{% endstep %}
{% step %}
### Adding custom instructions for criterions
Beyond look-fors and violations, use criterion-specific instructions to explain how the AI should interpret evidence for that particular criterion.
These instructions are useful when a criterion requires context, an exception, or a distinction that does not fit neatly into a positive or negative marker.
For example:
“For this criterion, focus on the reasoning behind the recommendation. More than one recommendation may be defensible.”
“Assess the accuracy of the explanation separately from the student’s use of technical vocabulary.”
“When discussing calculation errors, distinguish between an incorrect method and an arithmetic mistake within an otherwise appropriate method.”
Avoid repeating the same instruction in several places. Keep assessment-wide guidance in general instructions and local guidance under the relevant criterion. If two instructions conflict, revise them so the intended expectation is clear.
