Why it is a question
THEORY-334 holds what Ballard, Hayhoe, Pook and Rao (LIT-1075) observed. Seven subjects copying a pattern of coloured blocks looked back at the model about 1.3 times per block. Their commonest cycle was the one that holds least in memory, and it is the slowest in the authors’ per-block tallies. The authors infer that holding information costs more than re-fetching it. That is inferred from choices, not measured, and with no statistics. THEORY-346 (Gray et al.) sets against it a model that prices memory and display access in time alone and fits a copying task. Bogacz’s commentary on the article reads the same timings as task unfamiliarity. Until 2026-10-20 the test that would decide between these was THEORY-334’s promote_when.
What would answer it
The condition as it stood, unchanged:
- A block-copying or similar study with enough participants for inferential statistics. The cost of looking and the cost of holding items are varied independently, and the time each strategy actually takes, encoding included, is measured per person.
- For the carrying-cost reading. People still prefer re-fetching where it costs more total time, so that memory carries a cost beyond time.
- Against it. Their choices follow total time, as THEORY-346’s model does.
What cannot settle it. More fixation counts in the original design, because its timings do not separate encoding spread over a trial from the cost of each look.