science is repeatable verifiable observations

science is not blind faith nor is it an absolute certainty. It is repeatable observations verifiable to a discernible degree. And predictions thereof.

The overstretching of reason is precisely a departure from scientific method which prescribes degree of certainty to every knowledge; a measure of fitness for each pattern match. Human intelligence is largely about pattern matching (automatic mode) and pattern creation (learning mode). Mis-application of previously learned patterns often leads to overstretching, but that's the trade-off that was apparently advantageous to the previous 100,000 years of human evolution or so as a sociable animal. IOW it is much easier for us to believe than truly understand, maintaining the proof base for each and every one of our assumptions.

Obviously some trade-offs had to be made, for efficiency's sake. Which might yet prove fatal at some point in the future.

True intelligence/reason is like a self-aware Bayesian network, constantly re-checking assumptions at base of its knowledge. It's when we stop this re-checking when we overstretch our patterns, or as we call them, "understanding". Assumptions should not be confused with certainty.

So this mis-understanding is not necessarily erratic.

The world is very complex so over-reliance on assumptions might lead to wildly distorted visions of reality, that nevertheless maintain good degree of local match to the data selected (unconsciously) by the individual(s) interested in maintaining said vision (i.e. denial of fact). IOW it will be successful until a sudden break-off point where the force-fit mismatch finally ruptures, leading to a seemingly erratic break-down event.

Inadequate vision of reality will lead to wrong policies, which will fail, eventually.

As for the second law of thermodynamics, it only speaks of closed systems. Ours has an energy influx which pumps it up, so that natural selection provides for ever growing complexity.