When several teachers assign the same essay on Tvrdica, students in different classrooms can receive very different grades for similar work. The differences rarely come from carelessness. They come from individual interpretations of vague rubric language and personal habits about what deserves credit. Calibration sessions, where teachers score the same essays and compare results, are the most effective way to close that gap.

Begin by collecting a small set of anonymized sample essays that represent a range of quality. Five to six essays is usually enough for a calibration meeting. Choose papers that include some clear strong and weak examples as well as a few borderline ones where disagreement is likely. The borderline essays produce the richest conversation about what the rubric really means.

Ask each teacher to score the samples independently before the meeting, using the shared rubric. Collect the scores and display them side by side so differences are visible. This preparation keeps the discussion grounded in evidence rather than opinion. It also prevents the loudest voice from setting the standard.

Running the Calibration Meeting

Start with the essays where scores diverge most and ask teachers to explain their reasoning by pointing to the rubric language and to specific passages. The goal is not to decide who is right but to understand why interpretations differ. Often the disagreement reveals an ambiguous descriptor, such as what counts as sufficient analysis of a quotation. Revise the language on the spot so the next scorer reads it more consistently.

• Score sample essays independently before meeting
• Compare scores and focus on the largest disagreements first
• Tie every justification to rubric wording and passages in the essay
• Revise ambiguous descriptors as the group identifies them
• Record agreed examples of each performance level as anchor papers

Using Anchor Papers and Exemplars

Anchor papers are agreed examples that illustrate each score level, and they serve as reference points for future grading. Keep a small file of Tvrdica anchors with brief notes explaining why each earned its score. New teachers can use them to learn the department's standards quickly. Experienced teachers can check their own drift by rereading anchors before a grading session.

Update anchors periodically as the assignment or rubric changes. Outdated examples can mislead rather than help. Rotating in fresh essays also prevents students from guessing what the department wants. A living collection of exemplars keeps the process honest and useful.

Measuring Consistency Over Time

You can measure the effect of calibration by comparing score spreads before and after. If teachers scored the same essay anywhere from a two to a four before the meeting and now cluster closer together, the process is working. Track this each term and note which criteria still produce disagreement. Persistent gaps point to areas where the rubric needs more work or where further discussion is needed.

Double scoring a random sample of essays offers another check. Have a second teacher grade ten percent of papers and compare results. Large discrepancies trigger a conversation, and small ones confirm that standards are aligned. This light touch audit adds accountability without large time costs.

Where Technology Fits In

AI grading tools can support department consistency because they apply the same written criteria to every essay regardless of who is grading. A department can load an agreed rubric and use the output as a common first pass, with teachers reviewing and adjusting. Discrepancies between the tool and a teacher's score highlight places where the rubric language may be unclear. This feedback loop helps refine standards.

Keep humans in control of final scores, especially for essays near grade boundaries. Use the tool to reduce workload and improve consistency, not to remove professional judgment. Document how the tool is used so the process is transparent to students and families. A clear, shared approach builds trust in the grading system.