In their instantiation of the LTWM theory for experiments where positions are dictated, Ericsson and Kintsch state that:
The LTWM theory actually seems to be caught in a dilemma.
It is unclear what explanation the LTWM theory offers for the differential recall of random legal and random illegal games, becaus e in both cases, playing a move produces changes on the board that can be equally well encoded into the retrieval structure or into LTM elaborations.
While these results are compatible with the LTWM theory, it is unclear why they should 'clearly implicate LTWM in the maintained access to the updated chess positions' (Ericsson & Kintsch, 1995, p.
According to the LTWM framework, the presence of a powerful retrieval structure should allow experts to store units of information independently regardless of the type of grouping (by chunks, by columns, or in scrambled order) during presentation.
As predicted by the LTWM theory, interfering material presented after the presentation of a position affects recall only minimally (Charness, 1976; Frey and Adesman, 1976).
In summary, the evidence adduced in favour of the application of the LTWM theory to chess is much weaker than proposed by Ericsson and Kintsch (1995).
 However, it still seems fair to say that the available empirical evidence does not offer any special corroboration for the LTWM theory.
At about the same time as the LTWM paper appeared, two other theories of expertise were published, EPAM IV (Richman et al., 1995) and the template theory (Gobet & Simon, 1996b), which both take EPAM (Feigenbaum & Simon, 1984) as their theoretical framework.
By contrast, in the LTWM account: (a) a mechanism b ased on retrieval cues is used to explain how the retrieval structure permits a rapid encoding; (b) only groups of digits can be stored in the retrieval structure; (c) the retrieval structure offers encoding mechanisms different from those offered by schemas; (d) no time or decay parameters are given; and (e) a speed-up in basic cognitive processes, such as encoding and retrieval, is postulated.
As discussed above, the empirical data on chess skilled memory only weakly support the hypothesis of a 64-square hierarchical retrieval structure and even contradict some of the predictions of the LTWM theory, mainly because the proposed mechanisms (retrieval structure and rapid generation of LTM structures) are too powerful.
Why does the LTWM theory, which convincingly explains mnemonists' behaviour, have so many difficulties with data from expert chess memory?