How do you grade a short constructed response?
Updated 4 September 2026
Score against the published state rubric, and set your anchors before you score a single paper. Read six responses without assigning anything, pick one you are sure sits at each score point, and keep them in front of you. Then work one dimension at a time across the whole set rather than finishing each paper before you move on, because scoring drift comes from re-deciding the standard sixty times, not from reading slowly. A short constructed response is a few sentences making a claim with text evidence, scored on one small item-specific scale. An extended constructed response is a full piece scored on two dimensions that move independently, development and organization plus conventions, so it cannot be graded holistically and stay defensible.
SCR and ECR, in one paragraph each
Short constructed response (SCR). One to five sentences. The student makes a defensible claim about a text and supports it with evidence from that text. Scored on a small item-specific scale, usually 0 to 2 or 0 to 3, where full credit needs both halves: the claim has to be defensible AND the evidence has to actually support it. Most lost marks are a correct-sounding claim with evidence bolted on that does not do any work.
Extended constructed response (ECR). A full multi-paragraph piece. On STAAR it is scored 0 to 5 for development and organization and 0 to 2 for conventions, and the two are reported together. They move independently, which is why a holistic gut score is the wrong instrument: a student who argues well and punctuates badly is an extremely common profile and a single number hides it from them.
The method, six steps
1. Read six papers and set your anchors before you score anything
Pull five or six responses out of the pile and read them without assigning a score. Choose one you are confident sits at each score point and keep those beside you. Almost all scoring drift is the standard moving between paper 5 and paper 55, and anchors are what stops it.
2. Score one dimension at a time, not one paper at a time
For an ECR, score development and organization across the whole set, then go back and score conventions. Holding one set of criteria in your head for sixty papers is far more consistent, and far faster, than switching between two rubrics on every paper.
3. Capture the set in one pass
Typed responses go into one upload. Handwritten responses are photographed page after page with a phone, without stopping to grade. Separating capture from scoring is what stops a class set from turning into sixty individual tasks.
4. Run a first pass against the published state rubric
Use the actual state scoring guide rather than a rubric written at 9pm, so the score means something against the test the student will sit. A first pass should give you a score per dimension with a written reason, because a bare number is not reviewable.
5. Review the disagreements first
Do not re-read everything. Go to the papers where the first pass disagrees with your anchors, and to the ones sitting on a band boundary. Verifying a stated judgement is much faster than forming one, and the boundary papers are the only ones anyone will ever argue about.
6. Feed the result back into instruction, once
Constructed responses fail in a small number of repeated ways: no claim, a claim with no evidence, evidence with no explanation of why it supports the claim. Pull the two or three that dominate your set into one mini-lesson instead of writing the same margin note sixty times.
Where our numbers come from, and what they do not mean
This is the one response type where we can be specific rather than vague, because it is what our published benchmark is measured on. Texas releases real student constructed responses together with the scores its trained examiners awarded. We graded all 78 of them, grades 3 through 8, from the Texas Education Agency’s 2025 released scoring guides. Nine in ten came back within one point of the examiner, and quadratic weighted kappa was 0.78 against 0.70 as the commonly accepted bar for an automated marking system.
What that is not: an official score. Nothing here is affiliated with the Texas Education Agency or with any state assessment programme, and the score that counts on the real test is awarded by the state. This is practice scoring, useful for telling a student in October what a May examiner is likely to say, and useless as an appeal.
The rubrics we hold for this, in full
Free to read, no account, reproduced in full where the publisher is a public agency. Score against the one your students will actually sit.
- STAAR Short Constructed Response, reading domain
- STAAR Short Constructed Response, writing domain
- STAAR grades 6-8 ECR, argumentative and opinion
- STAAR grades 6-8 ECR, informational
- STAAR grades 3-5 ECR, argument and opinion
- STAAR English I and II ECR, argumentative and opinion
- TNReady grades 6-8 argument rubric
- New York grades 3-8 ELA 2-credit constructed response
- Illinois IAR grades 4-5 prose constructed response
All 86 rubrics, including Florida BEST, MCAP, MCAS, CMAS, NJSLA and the AP and IB scoring guides.
Where this goes wrong
Item-specific SCR scoring is genuinely hard.A short constructed response is scored against the particular passage and the particular question, and a generic framework can only get you so far. Give it the passage and the item stem, not just the student’s answer, or you are asking it to judge evidence it has never read.
Two-sentence answers have very little signal. The shorter the response, the more one word swings the score, and the more a 0-to-2 scale turns into a coin flip on the boundary. Review the 1s.
It is not the state. No affiliation with any assessment programme, no access to the real item bank, and no standing to award a score that counts.
No plagiarism or AI-writing detection. We do not do it, and on constructed responses that is a live question we have no answer to.
No SOC 2 certification. Institution plans are FERPA and COPPA aligned with a custom data processing agreement. If your district requires SOC 2, that decides it.
Common questions
What is a short constructed response (SCR)?
A short constructed response is a written answer of roughly one to five sentences in which the student makes a claim about a text and supports it with evidence from that text. On STAAR it is scored on a small item-specific scale, typically 0 to 2 or 0 to 3, where the top score requires both a defensible answer and text evidence that genuinely supports it. It is not a fill-in-the-blank and it is not an essay: the whole point is that a student has to produce the reasoning rather than pick it.
What is the difference between an SCR and an ECR?
Length and scoring structure. A short constructed response is a few sentences on a single item-specific scale. An extended constructed response is a full multi-paragraph piece scored on two separate dimensions, development and organization plus conventions, which are combined into a total. On STAAR the ECR is scored 0 to 5 on development and organization and 0 to 2 on conventions. The practical consequence for a teacher is that ECRs cannot be graded holistically and stay defensible: the two dimensions move independently, and a strong writer with weak conventions is a real and common profile.
How do you grade constructed responses consistently across a whole class?
Anchor before you score. Read five or six papers without assigning anything, pick one you are confident sits at each score point, and keep those in front of you. Then score one dimension at a time across the whole set rather than finishing each paper before moving on. Scoring drift comes from re-deciding the standard sixty times, not from reading slowly. If you use an AI first pass, the same rule applies: check that it agrees with your anchors before you let it touch the rest of the set.
Can AI grade STAAR-style constructed responses?
It can produce a first pass you review, and this is the response type our published accuracy figure is actually measured on. We graded all 78 student responses in the Texas Education Agency’s 2025 released STAAR constructed-response scoring guides, grades 3 through 8, against the scores official examiners gave them. Nine in ten came back within one point of the examiner and quadratic weighted kappa was 0.78, against 0.70 as the commonly accepted bar for an automated marking system. That is agreement, not authority: the score that counts on the real test is awarded by the state, not by us or by any other tool.
Does it work on handwritten constructed responses?
Yes. Photograph the page with a phone and the response is read off the paper, marked in place, and anything genuinely unreadable is flagged for you rather than quietly scored wrong. There is no template to print and no scanner. Faint pencil, heavy crossing-out and pages shot at an angle all cost accuracy, so a flagged answer is still work you have to do.
Which state constructed-response rubrics can I score against?
The rubric library holds 86 state and national writing rubrics reproduced in full, including the STAAR SCR reading and writing frameworks, the STAAR ECR rubrics for grades 3-5, 6-8 and English I/II, the TNReady grade bands, the New York 2-credit and 4-credit constructed-response rubrics, and the Illinois IAR prose constructed-response rubrics. You can also load your own rubric instead. Using the published state rubric matters more than it sounds: a locally invented rubric produces scores that do not predict the test.
How long does a class set take?
We do not publish a stopwatch figure, because it depends on response length, whether the set is handwritten or typed, and how much of the review you do yourself, and a number that ignores all three would be marketing rather than a measurement. What we will say is where the time goes: capture is fast, the first pass is fast, and your review is the part that scales with class size. Budget your review time, not the machine time.
What does it cost?
A free account covers 20 graded items a month. Pro is $30 a month, or $25 a month billed yearly, for 400 items. Max is $50 a month, or $42 yearly, for 1,000. Short constructed responses bill as one item each. Extended responses bill by length: one item per 500 words or one per page, whichever is larger. Extra items are 10 cents and overage is capped, so a heavy testing week cannot become a surprise bill.
Check it against your own anchors first
A free account grades 20 items a month. Take the six papers you anchored on and see whether it lands where you did before you point it at the set.
Start grading freeRead next
- AI grading for handwritten work: photograph the page, get it back marked, and what happens when the handwriting cannot be read.
- How to grade 100 essays fast: the same problem at essay length, with the four routes weighed honestly.
- Can AI grade AP free-response questions?: FRQ, DBQ and LEQ, and why a College Board score is a different thing.
- AI grading for teachers: what the category does and what it still cannot do.