AI Assessment QA: Evaluation Method Demonstration by Suliman AbdelatyAI Assessment QA: Evaluation Method Demonstration by Suliman Abdelaty

AI Assessment QA: Evaluation Method Demonstration

Portfolio project
Suliman Abdelaty
A traceable review of AI writing feedback
This self-directed demonstration shows an evaluation method within Abdelaty Assessment & Learning Design. The cases and system outputs are synthetic. No live client platform was tested and no achieved improvement is claimed.
The problem
Fluent feedback can still reward an irrelevant answer or invent a correction. This demonstration connects each finding to a response, a criterion and a retest.
The method and deliverables
The original framework covers assessment accuracy, rubric alignment, consistency, feedback quality, level sensitivity, edge cases, robustness and pedagogical risk. Its 50-case starter bank contains provisional writing anchors, controlled changes and targeted probes. Expected behaviours require specialist review before a commissioned evaluation.
The demonstration includes an operating protocol, an evidence workbook, a simulated report and five public examples. One constructed output says a writer supplied no examples even though the writing mentions double-sided printing and refillable soap dispensers. The acceptance check asks the feedback to recognise those examples before suggesting further development.
What this establishes
It demonstrates how to document an issue and design a retest. Vendor accuracy, independent reference quality, fairness, reliability and client outcomes remain unestablished.
Current service
The AI English Scoring Second-Opinion Audit (https://contra.com/s/oeFY2ZFX-ai-english-scoring-second-opinion-audit?r=sulimanabdelaty) is $550 for up to 20 written responses, one task family and one rubric. Delivery is within five UK working days of complete materials, with documented blind human review, findings and a 20-minute readout. A larger 60-response comparison is separately scoped.
Send your rubric, intended learners and main scoring concern through Contra.
Archived demonstration files
These original files describe an earlier offer. Their service names, prices and scopes are superseded by the current live listing.
Synthetic case F01: two examples in the response contradict the simulated feedback. The retest requires those examples to be recognised.
Synthetic case F01: two examples in the response contradict the simulated feedback. The retest requires those examples to be recognised.

1 like · 8 views · Sep 28, 2026

Self-directed AI writing QA demonstration with synthetic cases, simulated outputs and a traceable issue-to-retest method.

Tags