Context-Placement Study (2026-08)
Year: 2026
Type: preprint
Abstract: This component holds the 2026-08 context-placement study, a preregistered test of whether the context packet that produced a null before the decode pays off after it instead. A downstream corrector, OpenAI gpt-4o, makes span-limited edits to the 2026-07 context-effect study's own no-context transcripts, with the audio, human-produced references, ASR systems and metric code all inherited unchanged, so the within-side paired contrast isolates where the context was applied; the primary outcome is recall of curated vocabulary terms rather than aggregate word error, and damage to already-correct text is registered as a co-primary harm. This component contains the preregistration, the hash-verified frozen corrector and scorer, and the derived results; personal data is held out of the public record.
Author: Theodore Cochran
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204031365