How the STAAR ECR calibrations were built

What the calibrations were built from

TEA releases a scoring guide for every STAAR RLA test: the prompt, the reading selection, the rubric, and thirteen student responses with the score each one received and the state’s explanation of why. The guides from 2023 through 2026, for grades 3 through 8 and English I and II, are the only material the calibrations were built from and tested on. No student work from any school was used.

Every released paper was read into AutoMark exactly as printed, errors included, and its printed scores were checked against the guide before it was used.

How each grade’s setup was built

There is one setup per course: grades 3, 4, 5, 6, 7, 8, English I and English II. Each uses the rubric printed in that course’s 2025 scoring guide, grading guidance written from TEA’s explanations of the scored papers, and the released 2025 papers as examples of each score point. The guidance tells the grader what the state treats as a claim, what earns each Ideas score, and how Conventions is judged at that grade.

Scoring is structured. Before a score is written, AutoMark records what the paper contains: the claim in the writer’s words, each piece of evidence and whether it is explained, the introduction and conclusion, repetition, and the conventions errors. The score then follows from those findings and the rubric, so a long paper with no claim cannot score well on Ideas and a short paper with real evidence can.

How each setup was tested

Each setup was tested on the released papers from the other years, which answer different prompts about different passages. Those papers were held out: none of them was used to build the setup. AutoMark’s score for each paper was compared with the state’s printed score, trait by trait and in total, on one reader’s 0 to 5 scale.

Across the eight courses, AutoMark scores within one point of the state’s score about 9 times in 10 on state-released papers, and scores run slightly conservative by design. Each test was run once, and nothing was tuned after seeing a result.

What the figure means for your class

The state’s own readers do not agree with each other on every paper, and TEA sends papers whose two scores are far apart to an expert reader. AutoMark will also miss some papers, usually by one point and usually low. That is why every score is a draft until you approve it.

A calibrated score is one reader’s score. Double it to compare with a 0 to 10 state score.

What we do not claim

AutoMark is not the state’s scoring engine and does not predict what a student will score on the test. The calibrations are built so that classroom writing is scored on the same rubric, with the same standards, as the released papers. How a student’s next essay scores is up to the student.

Known limits

The standard for each grade is carried by the released sample papers, not by the rubric wording, so a prompt unlike any TEA has released may be scored a little differently than the state would. The grade 3 to 5 setups are the newest and have been tested once. Handwritten papers depend on the scan being readable. And the setups will be rebuilt as TEA releases new guides.