See where an AI English writing scorer disagrees with an independent human review before commissioning a larger evaluation. For EdTech product teams and assessment providers with a defined scoring rubric.
What you receive
Blind human review of 20 written learner responses, using your agreed rubric before model scores are revealed.
A response-level comparison table and descriptive agreement summary on the supplied sample.
A short findings report with evidence for the largest score or feedback disagreements and three improvement priorities.
A 20-minute readout call.
Scope and delivery
One task format, one model/version and one scoring rubric with up to five criteria. Maximum 600 words per response and 10,000 words overall. $550 fixed fee. Delivery within five UK working days of complete inputs. One human reviewer is included.
What I need from you
De-identified responses with stable IDs, original prompts, rubric and scoring scale. Supply model scores separately for the comparison stage. Confirm that you have permission to share the data. A secure transfer route and any NDA are agreed before data transfer.
Revisions
One consolidated clarification or factual-correction round within seven days. Rescoring changed responses, rerunning a model or a new rubric is a separate scope.
What this review covers
This is a diagnostic convenience sample, not a reliability, fairness or validity study. Human scores are reasoned review judgements, not ground truth. No model access, fine-tuning, code changes, official exam scores or certification are included.
The $550 diagnostic fee is credited against a larger AI English Scoring Comparison Audit booked within 30 days.
Start here
Message me with your task type, rubric and a de-identified example. Please agree the transfer method before sending learner data.
Inspect a sample
Download the free English Assessment & Course QA Samples. The four-page AI-writing report includes a fictional 137-word response, invented model scores, an illustrative reviewer rationale, comparison and retests. No model was run and no independent human rating was completed for that demonstration. A commissioned review uses the agreed client rubric and permission-cleared responses.
See where an AI English writing scorer disagrees with an independent human review before commissioning a larger evaluation. For EdTech product teams and assessment providers with a defined scoring rubric.
What you receive
Blind human review of 20 written learner responses, using your agreed rubric before model scores are revealed.
A response-level comparison table and descriptive agreement summary on the supplied sample.
A short findings report with evidence for the largest score or feedback disagreements and three improvement priorities.
A 20-minute readout call.
Scope and delivery
One task format, one model/version and one scoring rubric with up to five criteria. Maximum 600 words per response and 10,000 words overall. $550 fixed fee. Delivery within five UK working days of complete inputs. One human reviewer is included.
What I need from you
De-identified responses with stable IDs, original prompts, rubric and scoring scale. Supply model scores separately for the comparison stage. Confirm that you have permission to share the data. A secure transfer route and any NDA are agreed before data transfer.
Revisions
One consolidated clarification or factual-correction round within seven days. Rescoring changed responses, rerunning a model or a new rubric is a separate scope.
What this review covers
This is a diagnostic convenience sample, not a reliability, fairness or validity study. Human scores are reasoned review judgements, not ground truth. No model access, fine-tuning, code changes, official exam scores or certification are included.
The $550 diagnostic fee is credited against a larger AI English Scoring Comparison Audit booked within 30 days.
Start here
Message me with your task type, rubric and a de-identified example. Please agree the transfer method before sending learner data.
Inspect a sample
Download the free English Assessment & Course QA Samples. The four-page AI-writing report includes a fictional 137-word response, invented model scores, an illustrative reviewer rationale, comparison and retests. No model was run and no independent human rating was completed for that demonstration. A commissioned review uses the agreed client rubric and permission-cleared responses.