That's where it gets a little complicated. Each committee member is allowed to use whatever information they want to make their individual decision. This can be polls, power indexes, predictors, retrodictions, etc. as long as they disclose their sources. The one restriction is that the systems they use can not give weight to preseason rankings. Most computer ranking systems use preseason rankings, or performance from the previous year, for the first 6-7 weeks of the season until there is enough information from the current season to make accurate predictions. At that point the preseason information is removed from the algorithm. If they do not remove that information, then the committee members are not allowed to use that system.
After each round of individual voting, the members have discussion. This discussion has different rules than individual voting, it uses rules much like formal debate rules when it comes to sources. During the debate, you are not supposed to use rankings, indexes, predictors, etc. when debating teams because it is not expected that every member is aware of how each raking system works. Each system uses their own algorithm to determine rankings/SoS and sometimes that information is not disclosed. This leads to the discussion being based around more broad statistics like, "Team A's opponents had a winning percentage of 57%, but Team B's opponents only had a winning percentage of 48%." After the discussion, members are allowed to vote for whatever team they want based on their own information and the information presented during the discussion.
In short, people on the committee can use whatever the hell they want individually, but cannot use advanced metrics to argue their position with the other committee members.
