RevisionDojo IA Grader Review: How Accurate Is It?

The short verdict from this RevisionDojo IA grader review is that the grader is useful for identifying rubric-related weaknesses, prioritizing revisions, and obtaining a provisional criterion breakdown. However, students should not treat its estimated mark as a reliable forecast of their final moderated IA score.

As of 17 September 2026, RevisionDojo has not published a sufficiently detailed matched dataset comparing grader predictions with final moderated IB IA marks. There is therefore no defensible public statistic showing that a particular percentage of predictions fall within $\pm 1$ or $\pm 2$ marks. The responsible conclusion is that its feedback accuracy may be useful, while its predictive accuracy remains publicly unvalidated.

What the RevisionDojo IA Grader does

The RevisionDojo IB Coursework Grader evaluates uploaded IA drafts against the rubric selected for the relevant subject and assessment. Its report includes an estimated total, criterion-level marks, annotations, identified strengths, weaknesses, and suggested improvements.

RevisionDojo states that the same rubric is applied each time and that the score uses three-run averaging. This can reduce variation between individual AI evaluations, but it is important to understand what that proves. Averaging several outputs may improve repeatability, yet repeatability is not the same as agreement with an IB moderator.

The grader is best understood as a rubric-based diagnostic system. It attempts to answer questions such as:

• Which criteria appear weakest?
• What evidence is missing or insufficiently explicit?
• Where is evaluation superficial?
• Which claims need justification?
• What should the student inspect before submitting the next draft?

That is more useful than treating the headline number as a promise.

How accurate is the RevisionDojo IA Grader?

There is currently no robust public basis for claiming a specific level of RevisionDojo IA grading accuracy. No published validation report provides the necessary predicted-versus-final pairs, subject breakdowns, error statistics, date range, or analysis of drafts changed after grading.

This means claims such as “90% accurate” or “usually within two marks” should not be made unless supporting data becomes available. Individual student reports on forums can raise useful questions, but they cannot establish platform-wide accuracy because they are self-selected, difficult to verify, and often compare different versions of a draft.

A score can still be informative without being precisely predictive. If the grader repeatedly identifies weak evaluation, missing methodological justification, or an unsupported conclusion, that pattern deserves attention even when the exact mark is uncertain.

Why predicted and final IA scores can differ

A simple predicted vs final IA score comparison is more complicated than it first appears. The uploaded version, teacher-marked version, submitted version, and moderated result may all represent different stages.

Students often revise after receiving a prediction

Suppose the grader awards a Biology draft 15/24. The student then improves uncertainty treatment, adds stronger evaluation, and submits a substantially revised document that receives 18/24 from the school. Comparing 15 with 18 would not measure grading error because the underlying work changed.

A credible study must preserve the exact document graded by the system and match it to the identical document assessed by the school or IB.

Teacher marks are provisional

For internally assessed work, the teacher initially applies the relevant criteria. The IB then uses moderation to check whether the school has applied the global standard accurately and consistently.

The IB’s guide to assessment for teachers and coordinators explains that moderation can cause a school’s marks to be raised, lowered, or left unchanged. The official IB assessment overview also describes dynamic sampling, through which a sample is checked and an adjustment may be applied to the school’s marks.

Consequently, an AI estimate might disagree with the teacher but be closer to the moderated result, or agree with the teacher before the cohort receives an adjustment. A study must specify which comparison it is making.

Rubric judgment contains legitimate uncertainty

IB assessment criteria often require professional judgment. Two informed readers may agree that an investigation is strong while differing by one mark about the depth, consistency, or significance of its evidence.

The IB explicitly recognizes acceptable variation between examiners through marking tolerances. This does not make marks arbitrary. It means that exact-point predictions are a stricter test than identifying the correct performance region and revision priorities.

Subjects do not behave identically

A Mathematics IA, History IA, Economics commentary, and science investigation reward different forms of evidence. Mathematical sophistication, source evaluation, economic application, methodological control, and reflective evaluation cannot be assessed through one generic checklist.

Accuracy should therefore be reported separately by subject, level, syllabus version, criterion, and total mark scale. A pooled percentage could conceal strong performance in one subject and weak performance in another.

What existing RevisionDojo data does and does not prove

RevisionDojo has published an analysis of IB coursework moderation. It reports approximately 4,000 coursework samples collected over two months in mid-2025, of which 2,459 contained both a teacher-awarded mark and a final moderated mark. The dataset covered IAs, EEs, and TOK work across 33 subjects, 75 countries, and 240 schools.

This is relevant because it demonstrates that teacher and final moderated marks can differ. It is not, however, a validation dataset for the IA Grader. The published pairs compare teacher-awarded and moderated marks, not Jojo AI predictions and final marks, so they cannot establish the grader’s error rate.

Similarly, RevisionDojo’s 2026 State of Learning Survey contains self-reported information about grades and feature use, but it does not publish a matched grader-prediction-versus-final-IA analysis. It should not be repurposed to answer a different statistical question.

What a credible accuracy study should report

A trustworthy benchmark would begin with near-final IAs graded before official results become available. Each prediction would be frozen, timestamped, and linked to the exact unchanged document eventually submitted.

The dataset should disclose:

• The number of complete matched pairs
• Collection dates and examination sessions
• Subjects, levels, syllabus versions, and schools represented
• Whether the comparison uses teacher marks or moderated final marks
• Whether each IA was sampled directly or received a cohort-level adjustment
• Any exclusions and missing results
• Whether researchers or independent reviewers verified the matches

For each subject, the study should report the signed error

$$e_i = P_i - F_i$$

where $P_i$ is the grader prediction and $F_i$ is the final moderated mark. A positive average error would suggest over-marking, while a negative average would suggest under-marking.

It should also publish mean absolute error, median absolute error, and the proportions within $\pm 1$, $\pm 2$, and $\pm 3$ raw marks. Criterion-level agreement matters too, because a plausible total can hide incorrect reasoning when over-awarded marks in one criterion cancel under-awarded marks in another.

A stronger study would include confidence intervals and compare the grader against qualified teachers using the same anonymized scripts. It would test whether performance changes across score bands, subjects, and languages. Until such results are published, a precise accuracy percentage would be premature.

How to use the grader without overtrusting it

Start by confirming that you selected the correct subject, level, assessment type, and current syllabus. RevisionDojo’s subject-specific IA guides can help you locate the relevant structure, but your teacher’s current course documentation remains the authority for formal requirements.

Upload a coherent draft rather than notes, placeholders, or disconnected sections. The grader can only assess evidence visible in the file. If your appendices, citations, tables, diagrams, or calculations are missing, its judgment will reflect an incomplete submission.

Then use this process:

1. Record the total and every criterion estimate.
2. Read the explanation before reacting to the score.
3. Highlight comments tied to visible evidence in the draft.
4. Separate factual issues from debatable judgments.
5. Revise the underlying reasoning rather than copying suggested wording.
6. Ask your teacher about disagreements that could materially affect a criterion.
7. Preserve the original and revised reports so you can see whether feedback remains consistent.

The RevisionDojo IA help guide offers a similar criterion-focused workflow. You can also compare structural choices with assessed work in the coursework library, but examples should be analyzed rather than imitated.

Feedback to trust more and feedback to verify

Some forms of automated feedback are easier to verify than others.

Treat objective omissions as immediate checks. Treat holistic judgments as prompts for comparison with the rubric, teacher feedback, and marked examples.

If an annotation is vague, ask Jojo AI to identify the exact sentence, criterion language, and missing evidence. Do not ask it to rewrite a submission-ready paragraph. The IB’s academic integrity policy states that AI-generated material included in assessed work must be acknowledged appropriately, and students remain responsible for submitting authentic work.

Common mistakes when interpreting the result

The first mistake is converting a provisional raw mark into a guaranteed subject grade. IA marks form only one component of a subject result, and grade boundaries and component weightings must be handled separately.

The second is assuming that a higher score after revision proves the work improved. The output might partly reflect normal model variation, especially when changes are minor. Examine whether the criterion explanation changed for a defensible reason.

The third is following every comment mechanically. AI feedback can overlook subject context, misread an argument, or recommend unnecessary additions that weaken focus or exceed a word limit.

Finally, students sometimes use the grader only at the end. It is more effective as a diagnostic checkpoint while there is still time to rethink analysis, methodology, evidence, and evaluation.

Final verdict

RevisionDojo’s IA Grader is most credible as a rubric-based second opinion, not as a substitute for a teacher or a forecast of the moderated mark. Its criterion breakdown and annotations can make weaknesses easier to locate, but no robust public matched dataset currently supports a specific claim about predictions falling within $\pm 1$ or $\pm 2$ marks.

Use the score as a range indicator and the comments as hypotheses to verify. For a balanced workflow, combine RevisionDojo’s Coursework Grader with the current rubric, teacher guidance, authentic exemplars, and Jojo AI follow-up questions focused on understanding rather than generating assessed prose.

Sources and referenced URLs

• RevisionDojo IB Coursework Grader
• RevisionDojo IA guides
• RevisionDojo IA help guide
• RevisionDojo analysis of IB coursework moderation
• RevisionDojo 2026 State of Learning Survey
• RevisionDojo guide to IA moderation
• IB guide to assessment for teachers and coordinators
• Official IB assessment overview
• IB academic integrity policy