When more than one person grades the same exam, differences in scoring can undermine fairness. One grader may value creative interpretation while another focuses on accurate detail, and students pay the price for the mismatch. Anchor papers are real or sample responses that illustrate each performance level, and they give graders a shared point of reference. For an essay exam on Holberg's Jean de France, a small set of anchors can dramatically improve consistency.

To build anchors, select responses from a previous administration or a pilot that represent clear examples of each score point. Remove names and any identifying details, then annotate each one with a short explanation of why it earned its score. A top-level response might offer a debatable thesis about Hans Frandsen's role in the satire and support it with multiple scenes, while a lower one might only summarize. The annotations carry the teaching value.

Choose anchors that illustrate typical performance rather than extremes. Graders spend most of their time on responses in the middle of the scale, so examples at those levels are especially useful. Include papers that show common borderline cases, such as strong ideas expressed in messy prose, and explain how the rubric handles them. This prepares graders for the difficult decisions.

Running a Calibration Session

Begin by having graders read the rubric together, then score a new sample response independently. Compare scores and discuss any differences, focusing on the specific language in the rubric rather than on personal preference. The conversation often reveals ambiguous phrases that need revision. Repeat with a second sample until scores consistently fall within a narrow range.

• Select annotated anchors for each performance level.
• Include borderline examples that illustrate tricky decisions.
• Have graders score new samples independently before discussing.
• Revise unclear rubric language revealed by disagreements.
• Repeat calibration until scores converge reliably.

Monitoring Consistency During Grading

Calibration should not end when grading begins. Graders drift over time, becoming harsher or more lenient as fatigue sets in, so periodic checks are valuable. Insert a previously scored anchor paper into each grader's stack without telling them and compare the result to the established score. Large deviations signal a need for a conversation.

Double-scoring a sample of exams provides another check, with a third reader resolving significant disagreements. The proportion double-scored depends on the stakes and resources, but even ten percent offers useful information. Track agreement rates and share them with graders in a supportive way. The goal is to improve the process, not to criticize individuals.

Improving the System Over Time

After each exam, review which anchors proved most useful and which led to confusion. Replace or revise as needed, and add new examples that capture emerging patterns. A living set of anchors becomes more valuable each year. Share the collection with colleagues so that the benefits spread across the department.

AI-supported grading tools can complement human calibration by applying the same rubric to every response and highlighting scores that differ from the tool's analysis. These discrepancies are not necessarily errors, but they point to responses worth a second look. Human graders remain responsible for decisions, and the tool simply adds a consistent comparison point. Together, anchors, discussion, and technology build a grading process that students can trust.