Ever hear of "Team B?"
Edward Jay Epstein explains what it refers to:
In January 1976, in response to pressure from the President's Foreign Intelligence Advisory Board (PFIAB) to examine the way the CIA arrived at its National Intelligence Estimates, George Bush, then the newly appointed director of Central Intelligence, agreed to a test in which both the CIA (called Team A) and a panel of non-CIA experts (called Team B) would independently analyze the same underlying material on three national security issues.
Team B members, all approved by the CIA, included Harvard political scientist Richard Pipes; Gen. Daniel Graham, who had headed the Defense Intelligence Agency; Paul Nitze, a former deputy secretary of defense; Gen. John Vogt, the former Air Force chief of staff; Thomas Wolfe, a top Rand Corp. executive; Gen. Jasper Welsh, the head of the Air Force's system analysis; and Paul Wolfowitz, who was at the Arms Control Agency.
The three topics selected by the National Security Council were:
A fourth proposed topic, the detectability of U.S. submarines, was rejected by the Navy. The exercise began in August 1976 and ended in December 1976, with both sides presenting their conclusion to PFIAB.
The same data produced two startlingly different results. On the issue of Soviet missile accuracy, for example, Team A concluded that Soviet missiles were relatively inaccurate (¼ of a nautical mile), and therefore did not pose a major threat to U.S. silos; whereas Team B concluded that Soviet missiles may have attained sufficient accuracy (1/15th of a nautical mile) to threaten these same silos. ( As Soviet missile testing later revealed. Team B turned out to be correct on this issue.)
The lesson of this extraordinary disputation was not that the Soviet Union had a greater or lesser capacity but that intelligence estimates, no matter how objective they may seem, are an inherently uncertain enterprise, based on questionable assumptions and selective exploitation of sources.
The facts of intelligence work are not like marbles that can be lined up, counted and weighed. They assume different meanings depending on who selects them and orders them into a mosaic. Intelligence estimates are at best, therefore, an incomplete product.