Earlier I said that experiments involve careful observation of nature, trying to see it very accurately, with the overall hope of learning something new and surprising. I also emphasized the fact that nature is external to oneself and not totally under one's own power. Therefore it may seem incongruous that physicists seek experiments that can be repeated at will, under total control.
On one level there is no contradiction: we create experiments that can be repeated - however the first time we run the experiment we are in suspense about what will actually be observed. The certainty and repeatability is in our preparation of the experiment; the surprise is in the experiment's actual outcome, the details or whole of which may be different than expected.
The point of having totally repeatable control over an experiment is that the experiment's outcome should be exactly the same every time it is performed. Once an experiment is done once, the surprises should be over, and further repetitions of the experiment only serve to verify the scientific truth of the original result. In fact the only "truth" that physicists care to recognize while acting as physicists is those observations that can be repeated experimentally, as well as any mathematical techniques whose predictions have been repeatedly checked experimentally. If you repeat an experiment and obtain a different result than what was originally found, this casts doubt on whether the original result actually was "scientific truth."
For instance, the failure by many to reproduce cold fusion is said to have "discredited" the original experiment which observed cold fusion, leading most physicists to think that probably cold fusion did not occur even in the original experiment. Of course it is possible that fusion actually occurred the first time but not other times - but even if this were true it would not be a "scientific truth," simply because it can not be reproduced on demand by everyone. The only ways to establish the truth or falsity of this possibility would be similar to the procedures used when investigating and judging a crime - collection of historical evidence about the equipment that was used in the first cold fusion experiment and the data that was obtained, rethinking how to interpret all that was observed, judgments about the trustworthiness of the scientists, etc. There may also be reasoning based on currently accepted ideas of how the universe works, and efforts to establish whether or not these ideas indicate any possible way that fusion might have happened in the first cold fusion experiment. In any case, the result of this sort of process would not be "scientific truth," simply because it would be confining itself to a particular time, place, and set of scientists and observations. Physics does not deny the the possibility of truths of a particular, non-repeatable nature - the shape of a particular rock, the sound of a specific person's voice, etc. - but neither does it pay much attention to these things.