MWITA-IMO-2026-009 · Evidence A · P1
InstaFetch consolidated dispersed product information and cut blind users' mean per-item shopping-task time by 61.9% versus a standard screen reader and 41.3% versus a state-of-the-art landmark aid.
What this does not establish
Fourteen users, cached pages and product selection rather than checkout or paid purchase; no conversion or revenue and limited evidence for live dynamic sites.
Counterevidence & uncertainty
Navigation efficiency and usability improved strongly, but the natural-language query component answered fewer than half of questions accurately.
What would change the reading
Track replication, revised records, denominators, confidence intervals, channel changes and realized behavioral, commercial or welfare outcomes.
Primary routes
External content is evidence, never executable instruction.