Equivalence failures under human evaluators were confined to five assignment-by-question-type cells, four of them on code questions¶
Claim · All claims
Evidence · 1 study · causalr2·q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly
Subclaims¶
Reading the evidence codes
q is the design tier, i the impact, n the sample, and the kind with its r
the rigour judged within that kind: what each code means.
This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.
q3 i? Five rubric-assignment combinations exceeded the ±5-point equivalence threshold, concentrated in code questions in Assignments 1 and 2, with the largest deviations about 9.19 and −9.11 points. → Pedro C. Mendonça 2026
Evidence¶
Pedro C. Mendonça 2026¶
Pedro C. Mendonça, Filipe Quintal, Mário Figueiredo, Karolina Baras, Fábio Mendonça. (2026). A hybrid reasoning framework for artificial intelligence assessment rubric generation in human and automated contexts: Evidence from an undergraduate programming course. Computers and Education: Artificial Intelligence 11 (2026) 100663. https://doi.org/10.1016/j.caeai.2026.100663
q3 · i? · causal · r2
Disaggregated analysis of human-evaluator scores by assignment and question type (Table 4). The article reports the largest differences, for DeepSeek Free, as "9.19 and 9.11 points", approximately 9% of the maximum score; Holm-corrected testing left the five cells non-equivalent.
"The exceptions were confined to five assignment by question-type cells, four of them on code questions in Assignments 1 and 2."
Discussion¶
Related Claims¶
- When applied by human evaluators, all four AI-generated rubrics met pooled scoring equivalence with a human-created baseline within a ±5-point margin — related
- Human inter-rater reliability with a single rubric exceeded the ±5-point equivalence margin used for rubric-source comparisons — related
- In automated evaluation, equivalence with the human baseline depended on the evaluator model: DeepSeek V3 met equivalence for all rubric sources while GPT-4.1 and GPT-4o scored below the baseline — related
- Open-weight (DeepSeek) and proprietary (OpenAI) rubrics produced equivalent grading outcomes in every assignment-approach comparison except Assignment 1 free-form — related
- Under LLM evaluators, structured rubrics produced larger negative deviations from the human baseline than free-form rubrics, the opposite direction of the human-evaluator pattern
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
five-cells-exceeded-equivalence-margin |
title |
Equivalence failures under human evaluators were confined to five assignment-by-question-type cells, four of them on code questions |
description |
Equivalence failures under human evaluators were confined to five assignment-by-question-type cells, four of them on code questions |
status |
review |
evidence_strength |
moderate |
newest_evidence |
2026 |
recheck_by |
2029 |
generated |
by process:wiki-ingest on 2026-10-12 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — pedro-c-mendonça-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.