AI Writing Score Review: Synthetic Case Demonstration by Suliman AbdelatyAI Writing Score Review: Synthetic Case Demonstration by Suliman Abdelaty

AI Writing Score Review: Synthetic Case Demonstration

Portfolio project
Suliman Abdelaty

A SYNTHETIC WRITING-SCORE COMPARISON

The app, responses and scores in this demonstration are fictional. The example shows the structure of a documented human review; it does not report an evaluation of a live client or AI vendor.
THE QUESTION
A fictional workplace-writing app needs to know where its scores diverge from a separate judgement using the same descriptors. The example uses 20 emails in an intended B1-C1 context.
THE REVIEW METHOD
First review each response without seeing its AI score. Record the relevant criterion and response evidence. Then compare the decisions and group disagreements by task requirement and rating boundary. A real engagement needs authorised, de-identified materials, a frozen rubric and a documented model configuration.
ILLUSTRATIVE FINDINGS
The constructed example has 11 exact matches out of 20 and 18 within one level. Seven of nine disagreements give the AI the higher rating. These invented counts demonstrate how an overall figure can conceal a directional pattern; they are not empirical reliability estimates.
PROPOSED RETESTS
Check whether a response completes the required communication before relying on fluent language. Review contrasting B1/B2 examples and add length-matched probes. Proposed rules need specialist review and testing before operational use.
COMMISSION A DIAGNOSTIC
The $550 AI English Scoring Second-Opinion Audit (https://contra.com/s/oeFY2ZFX-ai-english-scoring-second-opinion-audit?r=sulimanabdelaty) covers up to 20 written responses and a 20-minute readout. The $1,500 comparison audit (https://contra.com/s/LYvvAvWe-ai-english-scoring-comparison-audit?r=sulimanabdelaty) extends the sample to 60 responses. Neither scope establishes model reliability or provides certification.

0 likes · 3 views · Nov 2, 2026

A fictional 20-response comparison demonstrates blind review, disagreement analysis and proposed retests. No vendor result claimed.

Tags