Try automated grading on one course
One term of suggested scores, checked by people, then a clear decision.
- Length
- One term, up to 16 weeks
- Price, indicative
- From $2,400
A vendor demo grades tidy sample answers. A pilot grades the answers your own students write in week nine, against your rubric. Every suggested score is reviewed by a person, and a blind sample is compared before anyone decides to go further.
How the term runs
- Weeks 1 and 2
Set up the rubric
Your rubric and two to four short-answer or essay questions loaded, with LTI 1.3 connected.
- Weeks 3 and 4
Score anchor answers
Your graders score 40 to 60 past answers that become the reference points for suggestions.
- Weeks 5 to 12
Grade with suggestions
Each answer arrives with a suggested score and reason; a grader accepts or changes it.
- Weeks 13 and 14
Compare a blind sample
A second grader scores a random sample without seeing suggestions, then the two are compared.
- Weeks 15 and 16
Write the memo
Agreement figures, grading time per answer and a go or no-go recommendation, in writing.
What you receive
- Who it is for
- Course leads and exam owners grading up to 600 candidates
- Format
- Online, inside your LMS or the Exampit platform
- Your rubric configured with anchor answers and score reasons
- A suggested score for every answer, each accepted or changed by a grader
- A blind sample comparing machine-suggested and human scores
- An audit trail of every suggestion, change and grader
- A go or no-go memo with agreement figures and grading time
Is this the right program?
Right for you if
- One course or exam with short-answer or essay questions
- Up to 600 candidates in the term
- A written rubric your graders already use
- A grader who can give an hour a week to the comparison
Look elsewhere if
- Your rubric is still being argued over; start with a Rubric Calibration Sprint
- You want scores released with no person checking them; we do not offer that
Example: a 480-student biology course
A community college in Las Cruces runs introductory biology for 480 students, with weekly short-answer lab questions and four teaching assistants. Through the spring term the assistants review suggested scores instead of starting from blank. A blind sample of 300 answers is scored twice, and the memo recommends going ahead on two question types but not the third.
Grading Pilot questions, answered
Does the machine decide any grades?
No. Every suggested score is accepted or changed by a grader before a student sees it, and each decision is logged with the grader’s name and the time. The pilot measures how often graders agree with the suggestion. Our responsible AI page sets out where a person always decides.
What happens after a go decision?
You move to standard automated grading, from $3 per candidate per exam, with the rubric, anchors and settings from the pilot carried over. A no-go costs nothing further. The pricing page lists both, and the $2,400 pilot fee is fixed for one course or exam of up to 600 candidates.
How do you compare machine and human scores?
A random sample, usually 200 to 300 answers, is scored by a second grader who cannot see the suggestion. We report exact agreement, agreement within one point, and where the two disagree most by question and score level, so you see where suggestions help and where they mislead.
Does the pilot work inside our LMS?
Yes, through LTI 1.3 with Canvas, Moodle, Blackboard and D2L Brightspace. Students submit where they already do, graders review suggestions in a panel beside each answer, and final scores return to the gradebook. Courses and exams without an LMS run the pilot on the Exampit platform instead.
Plan this program with us
A 30-minute call to see whether Grading Pilot fits your teams, your calendar and your budget.