Auto-graded questions¶
Gradebird supports explicitly authored True/False and multiple-choice questions. These questions do not use rubrics, partial credit, assessment feedback, or deliberation.
Open the workspace¶
Open the exam's Grade workspace and select Auto-graded from the question selector/card. The page contains overall status, Answer key, answer-detection controls, a collapsible Recorded-answer summary, and Needs attention when exceptions exist.
Set or change the answer key¶
Select every correct printed option for each multiple-choice question, or one answer for True/False. Each change saves the draft selection. Clearing every selection leaves an incomplete draft; it cannot be applied. Draft choices do not become the scoring key until all questions have choices, draft saving has finished, and you select Set answer key.
Later selection changes show Unapplied change. Select Apply key changes to update the key. Gradebird immediately rescores every current recorded response.
Answer keys are set in Grade, not in the exam builder. Answer detection can run before a key exists. In that case a visible response can be Recorded but not Graded.
Detect visible answers¶
Accepted auto-graded responses can be detected as scan processing finishes. When responses remain and no detection is active, select Detect answers or Detect remaining answers. After selecting Cancel, wait for the status to become Cancelled.
Gradebird compares two independent readings of the complete visible selection. Matching sets of selected options (in any order), blank, or unclear are recorded automatically. Disagreement or an unsuccessful reading enters Needs attention.
Deterministic scoring¶
The recorded response is scored against the current key:
- selecting exactly the options in the key receives full points;
- missing a required option or selecting any extra option receives zero;
- Blank and Unclear receive zero;
- no current recognition or no current key means there is no current score.
Auto-graded questions therefore receive either full credit or zero. For example, a key of A and C gives full credit only for A and C; A alone or A, B, and C receives zero. A single-answer key still requires exactly one matching selection.
Status definitions¶
| Main status | Meaning |
|---|---|
| Graded | A definitive current response is scored against the current key. |
| Recorded | A definitive response exists but a current score does not, usually because the key is absent or changed but not applied. |
| Processing | Independent reads are active. |
| Needs attention | Readers disagreed, two successful reads were unavailable, or a multiple-answer key needs the selections from a legacy multiple-mark result. |
| Awaiting detection | An accepted crop has not been read, or changed after an earlier reading. |
The recorded-answer summary additionally separates Selected, Multiple selected, Blank, Legacy multiple marks, Unclear, Disagreement, Read failed, Processing, and Awaiting. Each count links to a bulk review filtered to that status.
Human review and correction¶
Open Review from Needs attention, a summary count, or an auto-graded result in Reports. The page shows the accepted crop, printed question and choices, current key, score, and how the answer was recorded. Select every visibly marked option and choose Save recorded answer, or use Blank or Unclear. Record the visible response independently of the key.
Saving the visible answer updates the score immediately when a key exists. It then advances to the next available exception; after the last exception it returns to Auto-graded. A reviewer can also correct an answer that was previously recorded automatically. The identity toggle appears only when the effective course-wide or per-question permissions allow student names.
If Scan replaces the accepted crop, the response returns to Awaiting detection. Selecting multiple options against a single-answer key earns zero.
Existing finalized exams¶
Change the grading-time key without republishing the exam, printing new packets, or uploading scans again. Published questions and scan geometry stay unchanged.
Older Legacy multiple marks results do not contain the individual option identities. Instructors can choose Recover selected options to reread affected AI results once from their existing crops. Recovery uses two independent readers and preserves earlier records. Failed or disagreeing recovery requires human review; it cannot be rerolled. Cancelling before a response is attempted leaves it eligible for recovery.
Human corrections are never automatically reread. Manually record their selections when necessary. A legacy multiple-mark result remains zero under a single-answer key. With a multiple-answer key it needs recovery or manual clarification before a current score or new report can be released.
Previously released reports stay unchanged. Review updated live results and use the normal release workflow to publish revised reports.
Reports¶
Reports show the printed choices, recorded answer, correct answer, and earned points. They do not show a rubric, partial-credit reasoning, or free-response feedback for an auto-graded row. Auto-graded questions appear in numeric class-performance statistics but do not offer rubric-based AI class analysis.
Auto-graded questions¶
This workspace sets answer keys, starts visible-answer detection, summarizes recorded outcomes, and links to responses needing human review. Deterministic scoring uses the current key; changing the key recomputes eligible scores without creating rubric-based partial credit.
Review auto-graded responses¶
This page presents a server-selected set of auto-graded responses sharing a review status. Record the visible answer for each item and save; the filter or selected status does not change the underlying exam page identity supplied to the chatbot.
Review an auto-graded response¶
This page shows one accepted crop, printed choices, current key, recorded answer, and resulting score. A human can correct the visible answer or exception state, after which Gradebird updates deterministic scoring and returns or advances according to the review workflow.