<![CDATA[Differentiated instruction is widely endorsed, but its implementation is limited by the preparation time required to map learner profiles and by the interval between the submission of a draft and the return of comments. This study examined the validity, practicality and effectiveness of WETEL–GAIS, a web ecosystem whose closed-domain agent, Eko-Bot, conducts Bayesian diagnostic profiling, routes students to multimodal content, provides Sentence-BERT formative scaffolding, and sets a transmediation task for summative assessment of observation report (LHO) writing on fishery ecology. An explanatory sequential mixed-methods design was employed, its quantitative strand a nonequivalent pretest–posttest control group quasi-experiment involving 66 Year 10 vocational students at SMKN 1 Sidayu Gresik, allocated by cluster random sampling to experimental (n = 33) and control (n = 33) classes. Data were analysed using Kolmogorov–Smirnov and Levene tests, independent-samples t-tests, Hake's normalised gain and Cohen's d, and triangulated with interview, observation and telemetry data. The ecosystem proved highly valid (92.43%) and highly practical (teachers 91.25%; students 88.71%), and its profiling agreed with a standard questionnaire in 87.88% of cases. Normalised gain in writing reached 0.61 against 0.34, t(64) = 9.216, p < 0.001, d = 2.27, and in ecological literacy 0.49 against 0.26.]]>
Copyrights © 2026
