Apart from the highly questionable and suspiciously timed dropping of follow up actigraphy (which according to the results of previous trials would have been a null result and further discredited the deconditioning hypothesis), I would guess that their intentions and expectations were relatively optimistic towards the start of the trial. They expected good results and the definition of clinical improvement and recovery were generally much stricter back then compared to what it became.
Since then they have made big changes to the individual-level definitions of clinical improvement and recovery, seemingly well after the trial was already over and after they had already become very familiar with the outcomes data. Even the changing of the primary end point (to continuous scores instead of categorical outcomes), which we were told happened after the collection of data was over but before the unblinding of data, coincided with the FINE Trial failure which used the same or very similar outcomes as PACE originally planned. The PACE Trial was open label so it's not like general impressions about how the trial was going while it was still running were adequately suppressed either.
But the imbalanced politics of ME/CFS are such that, while it is widely acknowledged outside our community that changing protocol after seeing data is an obvious source of potential bias, it is politically incorrect to remotely suggest that the PACE authors, after having largely built their reputations on the success of CBT/GET and with a clear stake in the controversy, could possibly have been influenced by the poor trial results (subconsciously or otherwise) when drastically lowering the thresholds for success. There are not just one or two minor infractions which could have happened by chance, but a long list of questionable changes, poor justifications, published errors, weak dismissals for the lack of objective improvements, spin doctoring, and media hype or misinterpretation: *all* of which just happened to coincide in their favour.
Despite all the spin and posturing, there should be no doubt that the PACE Trial has ironically dealt a massive blow to the cognitive behavioural model of CFS. Even after 80% of ME/CFS candidates were excluded from participation leaving an Oxford criteria cohort, there was only modest self-reported improvements for a small fraction of remaining participants but no relevant objective improvements on average. The results are consistent with the placebo response and response biases in an open label trial which aimed to change participants' perceptions about their illness. The "recovery" criteria was so poor that the results for this outcome can basically be ignored. The "mediators" paper actually admits that the direction of causation is uncertain! If there is anything at all to CBT/GET, they should be finding out what that is, and who it specifically applies to, without subjecting all patients by default to a non-representative model promoted beyond its worth.