In every non-life example I can envision, a copy of a copy is always a degraded or less pure version of the original unless some outside influence acts to correct the copy back toward the ideal represented by the original. Photocopies get blurrier with each generation. Casts from a mold are distorted from the original from which the mold was made. In fact, each cast degrades the mold itself. When data is copied on computers or across networks, parity checks verify that no mistakes were made, but even then, every long once in a while, combinations of errors can cause a false positive in a parity check. So given enough time, the copies would degrade.
In eukaryotes, new individual organisms always begin as a zygote, so in all kingdoms, reproduction boils down to the genesis of a single cell. This involves correctly building the DNA as well as all of the other complex architecture of the cell. Why doesn't this cell degrade like every other example I can think of? In fact, cells are capable of such perfect reproduction that the system generally supports the introduction of additional randomness in order to promote the possibility of productive change. I can think of some probable contributing factors that make this work, but I must be missing something. I can't imagine that this model would actually work the way it does - so well in fact that the design actually improves over time. What am I missing or underestimating?
Contributing Factors (I guess):
Perfect Building Blocks: Cellular development follows a pattern at every level, and ultimately operates all the way down to the molecular level. At that level, nearly all building blocks are identical. Life is built of stable atoms, not something like plutonium, and in the rare event that an atom does change, the result is simply a different kind of atom, which still tends toward a stable form in the long term. Because the structures of life are ultimately made of stable components that are plentiful everywhere in the environment, the essential structures being copied are precise and can be copied precisely. Photocopies and casts are not precise to the molecular level, so copying them is more approximate by nature. Digital data propagation, however, is a very similar process. Bits are also theoretically perfect building blocks.
Fitness Correction: When mistakes do degrade the reproduction process, rather than maintaining or randomly improving upon it, there is a correction mechanism that removes the defects from the process. Those defects do not survive to reproduce. This evolutionary process acts to keep the reproductive pattern focused back upon a theoretical ideal which is independent of a specific physical form to be copied.
This seems like the most essential element of the explanation, because it is ultimately only through progression that digression can be avoided, but it is also the part that seems the most dubious. Astronomical quantities of defects would have to be produced before developing just one advantageous feature. I would expect living creatures to be 99% defective with only 1% surviving to breed. I would expect 99.9% of zygotes to expire without being born or sprouting from seed. I would expect all sexual organs (ovaries, testes, stamen, etc.), if not the majority of the whole body, to be mostly dead cells, with just a few successes surviving to fertilization. I would expect 99.9% of the genome to be experimental, almost completely unusable liability to the species. Essentially, I would expect premature death to far outweigh successful life everywhere and at all times. And even so, I would still expect evolution to be even slower than it has been.
Mutation Management Mechanisms: I understand that there are mechanisms in reproduction that decrease the likelihood of mutation in more established and stable parts of the genome compared with sections that are more open for discussion - epigenetic structures, HOX genes, etc. Portions of all genomes have been established and functional for hundreds of millions of years, so I gather that there are mechanisms for protecting them (I suspect probably far more than we have yet discovered).
Note: The numbers I present are fuzzy and are based not on calculations but on general impressions I get of the magnitude of the numbers involved and the relative rareness of useful mutations. Is there any place where this kind of calculation has been performed with more realistic approximations of probabilities?