Science does not fix its own mistakes automatically; errors get caught only when particular people decide to check and speak up. In the case the book follows, fabricated results were exposed because individual scientists made the corrections, not because the system did it for them.
A fabricator can work backwards: decide the result first, then produce numbers that fit it. In the book's case, the supposed measurements were often calculated from equations rather than taken from any experiment. Data that matches theory too neatly can be a sign that the theory produced the data.
Colleagues who rely on a star's results without checking are not simply fooled; trusting so much is a conscious rationalisation. The excitement of the results also stops people asking basic questions about how the work was done, questions they would otherwise have asked.
Journals that pick papers for their power to make headlines let dramatic fakes through, even at the most prestigious titles. One author got fraudulent manuscripts published at a rate of roughly one every eight days. Labs under pressure to justify their research budgets have the same appetite for breakthroughs.
A fabricated breakthrough costs more than its own retraction, because other labs spend time and money trying to reproduce something that never happened. Failed attempts to replicate are an early warning, and keeping original data in a notebook is what makes a claim checkable at all.