I'm always surprised how people describe evolution as from a source information independent from fossil evidence. When, in fact, we tell what evolution was, up until now, from the evidence we have and not something else. The evidences shapes the theory (the sum of theories). If complex organisms become simpler, than that's how evolution works. If simple organisms become more complex, than that's how evolution works.
Epistemically, what's going on, I think, is that evolution is pretty much assumed to be true and from the evidence it's only concluded what results it had.
There is certainly no forwards and also no backwards. It is just an on going optimization process. It is on going because the constraints change. Most of the constraints can be summed up under the general term "environment". The effects of the climatic environment can be grasped more easily. However it is harder to grasp how all(or most) of the constraints feeding into each other, across both time and space.
With that said, life as we know it has a penalty on needless complexity. And, what does complex mean anyway? Complex genotype or complex phenotype? What is a good metric for measuring each one? Whatever the metrics are, it cannot be said that they are related monotonically. In this sense, we have four permutations of complexity i.e.[simple genotype, simple phenotype], [simple genotype, complex phenotype], [complex genotype, simple phenotype] and [complex genotype, complex phenotype]. Life as we know it is defined in the whereabouts of the genotype level. The "penalty" is on the [complex genotype, complex phenotype] type of complexity primarily because of the inaccuracy in the cell division process/processes (life as we know it!!). With many points that can go wrong in the foundation(cells->chromosomes->genes -- genotype), you don't want many points that can go wrong in the outcome(phenotype).
Of course life can (and does) beef up the genotype level with more complexity to try and ensure a rock solid phenotype (complex or not) but all factors withstanding, a simpler phenotype is easier to maintain.
The evidence shapes the theory, but the theory also shapes what we can understand from the evidence. As well, the fossil record is not our only evidence. When people describe evolution as a thing independent from the fossil evidence, they mean "The things we've learned about how evolution works." It's not really independent of the fossil record, but neither are they one and the same.
Similarly, people talk about physics as a mathematical field, even though what actual entities do in the real world is the only real source of information. It isn't that the real world can be "wrong," but if a piece of evidence doesn't fit with our accumulated knowledge, that needs to be explained, not just accepted.
I call this "Science of the Gaps". Where there is insufficient evidence to reach a conclusion, or to provide a mechanism for a process we know occurs, one is just made up (by non-experts seeking to bolster the image of science).
My understanding is that no one knows how life originated, but that doesn't stop many people from prematurely invoking the Anthropic Principal.
On complexity, logically, we cannot say that evolution has no tendency towards complexity, and also that evolution explains the complexity of modern organisms. Not that the OP was making the latter claim, but it's not an unfair assumption.
Also, your post seems to suffer from "feigned surprise" :-)
Epistemically, what's going on, I think, is that evolution is pretty much assumed to be true and from the evidence it's only concluded what results it had.