I once worked on a very large commercial application that was based on a home-grown I/O framework which operated on similar principles.
It was a complete pain in the ass. You were constantly future-proofing your data structures because you knew you were going to be stuck with them for all eternity because the I/O framework was going to serialize them verbatim whether you liked it or not. Those were dark days...
Require reviews to show a substantial understanding of what the book was about? That means moderation which is always dangerous but probably better than the alternative.
> floats really really want to be "near 1", to keep precision.
The number of significant digits is identical for (nearly) the entire range of FP values. There's no value to keeping it "near 1" for IEEE 754 floats - the precision is exactly the same regardless whether near 1 or near 1 trillion. This makes them ideal for general computation and modeling physical properties.
In contrast, posits, the unum alternative to IEEE 754, are highly sensitive to absolute scale. Posits lose precision as the magnitudes increase. Otoh, for small values, you get much higher precision which is why they getting some attention from the AI world where normalized weights are everywhere.
It might be stupid, or it also might just be naive, or a shift in priorities.
100x throughput improvement might just come from caching results from earlier computations (less naive) - at the cost of 10x memory footprint possibly (different priorities).