first of all you have accept that its not binary. if you read the linked report, they spend some time talking about discrepancies between models and some of the problem areas where they don't agree.

in general predictive accuracy is the bread and butter of this kind of simulation work. since we aren't actually simulating the earth, everything in an approximation. you get work published by analyzing the failure modes of these approximations, investigating new simulation techniques, and examining the impact of integrating more and more effects (like chemical reactions, or air-water heat exchange, or more detailed salinity models, or ...). Each these papers does their own analysis, often times by 'replaying history', that is taking a time period with sampled data, evolving the state of the simulated system and comparing it to the measured evolution of the actual system.

so 'is it accurate' is not really a meaningful question, 'it is sufficiently predictive to be useful with an acceptable confidence' is maybe a better question to ask.