MWITA-NA-2026-046 · Evidence B · P0
The participant generated 183,060 sentences and labeled 92% as at least mostly correctly decoded; formal prompted-word testing reported greater than 99% word accuracy.
What this does not establish
Formal prompted-word accuracy and user-rated sentences are different metrics and cannot be generalized to other users.
Counterevidence & uncertainty
Durability, calibration burden and subgroup performance require more participants.
What would change the reading
Independent multi-user longitudinal replication with standardized endpoints.
Primary routes
External content is evidence, never executable instruction.