Before you release an English writing scorer or change its prompts, check whether its scores and feedback follow your rubric. I help EdTech product and assessment teams investigate inconsistent ratings, invented errors and fluent but off-topic answers, with documented evidence and acceptance tests for the next release.

EdTech teams building or improving an English writing assessment or feedback feature. A plausible score is only the beginning: feedback should follow the rubric, identify real evidence and help the learner improve.

• 30 unique synthetic test cases selected for one agreed writing workflow and reviewed by me.

• 50 planned system executions: one per case plus two further executions of ten cases to examine repeatability.

• A PDF findings report with priorities and practical recommendations.

• An editable evidence workbook linking observations to inputs, outputs and criteria.

• Acceptance tests your team can use when checking fixes.

• A 45-minute walkthrough of the findings.

• One consolidated clarification or report-correction round, requested within seven calendar days of delivery.

Task relevance, rubric alignment, scoring consistency, feedback accuracy, unsupported claims and controlled response variations. Each finding connects learner evidence to the AI judgement, explains the possible consequence and sets out a retest. Specialist reference judgements are documented and frozen before I inspect the AI outputs.

1. We agree the feature, supported task types, rubric, configuration and evaluation questions.

2. I select and review the cases and record specialist reference judgements.

3. I execute the planned tests through authorised access, or review complete client-provided outputs under an agreed export protocol.

4. You receive the report, workbook, acceptance tests and findings walkthrough.

Your product link, authorised test access or agreed output exports, sufficient test credits, the applicable rubric and score scale, the target learner profile and one project contact. Synthetic cases are used by default; do not send identifiable learner data or production credentials.

USD 950 for one English writing feature, one rubric and one frozen configuration. Delivery within 10 business days of the agreed start date, after scope, access and required information are confirmed. Platform charges and third-party testing costs are separate. Additional configurations, new features and retests after changes are quoted separately.

I am Suliman Abdelaty, FHEA, FRSA, an English Language Assessment and AI Quality Specialist with 12+ years in education. My background includes accepted assessment-development work for an international EdTech client and published research on language-assessment reliability. I deliver the review personally.

The audit supplies specialist diagnostic evidence within the agreed sample. It does not certify a product, establish population validity or inter-rater reliability, or deliver official exam scores. It is a single-specialist review. Audio, pronunciation, clinical content, software implementation and formal psychometric studies require a separate proposal. Send your product link, writing feature and rubric through Contra to confirm fit and an available start date.