Resource
Voice-First Written Assessment
VFWA is not a product you install; it is an assessment structure with a site, a working paper and an implementation guide behind it. The structure is two stages. In the first, students respond to a prompt they have not seen before, under supervision, in whatever language or modality suits them (speech, handwriting, an argument map, another language), and the response is checked for sufficiency rather than graded. In the second, they take that artifact home and develop it into the graded submission with AI tools permitted and optional. What the assessor judges is whether the final piece can be read as the student's accountable development of the first.
What you can do with it
Read the model, then the implementation guide, which carries the workflow options, the Stage 1 capture methods, templates and scaling advice a program team would need to run a pilot. The site also publishes a rationale-to-feature matrix mapping each design choice to its justification, three worked discipline examples (a 1,200-word human-rights memorandum in law, a results-and-discussion section from an unseen dataset in STEM, and a critical essay from an unseen passage in English), and a recorded webinar from March 2026.
The paper behind it is a preprint on OSF, linked from the same page, and it is the place to go for the validity argument rather than the practical one.
Who it is for
Program and module leads in higher education who still want extended writing to count but no longer trust a final essay as evidence of reasoning: law, health sciences, humanities and any discipline where the task is divergent rather than a single correct answer. Assessment designers looking for an alternative to detection will find the case made explicitly. It is not a fit where the learning outcome is mastery of written English, rhetorical performance or genre control, since those have to be assessed directly and securely.
Notes and caveats
The model has not been evaluated at scale. The paper is explicit that no sample, instrument or measured outcome exists yet, and that pilots are needed to test whether Stage 1 artifacts raise assessor confidence or interpretive consistency, whether students experience the design as fairer, and how a discipline sets a sufficiency threshold without turning Stage 1 into a second exam.
Two practical costs are real. Stage 1 requires observed sessions and secure capture at scale, and multilingual recording and translation are prerequisites rather than extras. And the interpretive judgment is deliberately left to markers: the model assumes disciplinary expertise that cannot be automated, which is a staffing assumption as much as a design one.