Equivalent measured performance does not establish equivalent action.
I believe an audit must distinguish agreement on its reported measures from agreement on the actions a system takes. The same measured performance can leave the identity of selected people, retrieved evidence or final decisions unspecified. I want institutions to examine those differences before treating two systems as interchangeable, with a level of scrutiny proportionate to the consequences.
RIDI: the research question behind the position
Read the full argument →Capacity belongs inside the evaluation.
I regard the number of interviews, inspections or support places as part of the decision system. Changing capacity changes the boundary that a model acts on. I want institutions to compare plausible capacities and explain who bears the consequences of the chosen limit, rather than treating that limit as an invisible implementation detail.
Read my argument about capacity and accountability
Read the full argument →An institution must be able to explain what its categories mean.
I believe workforce intelligence starts with an explicit account of the work: its purpose, responsibilities, decision rights and requirements. A title or code cannot carry that meaning by itself. I want AI to make assumptions reviewable while the institution remains responsible for the definitions, revisions and decisions it adopts.
MIYAR: an implementation to examine
Read the full argument →Readiness requires evidence of judgment under changed conditions.
I believe a learner should explain a choice, test it, and revise it when a relevant constraint changes. Attendance, content coverage and a polished AI-assisted artifact do not establish that capability on their own. I want evidence appropriate to the discipline, including the learner’s reasoning and the limits of what we can infer from one task.
iSCARB and the educational design behind this position
Read the full argument →iSCARB · judgment, evidence & context
iSCARB: what makes a judgment defensible?
I want a learner to connect a claim to evidence, explain the limits of that evidence and revise a choice when a consequential constraint changes. I bring this position into iSCARB, with IMAM as a contextual design layer for language, local examples and educational purpose. I believe that context should help a student reason within the discipline while the university retains responsibility for curriculum, assessment and the meaning of readiness.
Read the education argument →
iSCARB: learning design & practice →I investigate why equivalence on declared performance measures need not imply equivalence in downstream action. I use RIDI to make that question inspectable, including how selection identity can change under a stated constraint. The public demonstration replays one SciFact case: the displayed retrieval metrics remain unchanged while the recorded answer changes. I use that case to examine the limits of the reported audit; I would need separate evidence to estimate the frequency or consequences of such changes in a deployment.
Implementation & project stages →