Why far transfer is harder than it looks

Section 1

When a cognitive training programme reports that users improved by 20 percent, the first question is improved on what. In most studies, the measure is the same task users trained on. That is near transfer, performance on the trained task itself, and it is weak evidence for broad cognitive benefit.

Far transfer is different. It is improvement that carries to tasks that were never part of training, with different formats, demands, and contexts. Far transfer is what people mean when they ask whether cognitive training actually works.

Section 2

Repeated exposure produces procedural learning. Practice a task and score higher on that task. This is not automatically a capacity gain. Near-transfer trends and far-transfer trends can both look upward on charts, so they must be separated with explicit design.

That means testing with changed formats, changed rules, and delayed checks. Without those checks, familiarity effects can be mistaken for portable gains.

Section 3

Capacity Gym includes game swaps, boundary probes, and delayed re-checks. The goal is to test whether performance survives condition changes, not only whether in-task scores rise.

Re-checks are delayed rather than immediate. A gain that disappears once format or delay changes is not treated as far-transfer evidence.

A gain that does not survive condition changes is familiarity evidence, not far-transfer evidence.

Section 4

If your trend moves upward, the next question is whether it remains stable under game swaps and delayed checks. Probe outcomes are often a cleaner transfer signal than raw session score alone.

Protocol logic is publicly documented at /proof#protocols. Claims boundaries are explicit at /proof#claims.

Section 5

Transfer remains an active research area. The evidence base is mixed and should be interpreted with caution. IQMindware uses conservative language aligned to published boundaries and evolving summaries.