As I can see, you have talked about various numbers, like 66~75%, 20% etc. These appears to be magic number to me, like BIP 101's 8mb or BIP 103's 4.4% & 17.7%. How do you derive them ?
I've said stated all these figures need to be discussed. I believe these thresholds need to exist, but a decent value or a decent way to compute these values needs to be discussed. I can give you the reason these values need to exists but I have not done the research to determine which figures are the most appropriate, of it there is a way to set these dynamically as well.
The 66~75% figure is a proposal for the block space usage threshold at which a resizing should be tested against secondary conditions. The rationale is that organic market growth will always burst through any threshold (until it hits a hard cap eventually), whereas an attacker won't necessarily. Raising the space usage threshold increases the effort required by ill intentioned parties, and doesn't change a thing for natural market growth. As a reminder, the current threshold is 50%.
The 20% figure denotes the fee growth threshold, i.e. a resizing should only occur if fees have gone X% either way compared to the previous period. Currently there is no such threshold, making it trivial for any attacker to push up the block size and maintain it high.
As long as these thresholds are in place and tight enough, an effective decay function can be implemented. The goal is to distinguish between organic growth in demand and spam attacks, and use a safety net mechanism (the decay function) to correct all growth that is not supported by actual demand. It would actually mimic commodity prices in a speculative market: large speculators can pump the price for a while but eventually the market will always correct itself, with the valid demand as its baseline.
The first threshold is not very important. It will always be reached first when demand climbs, so its particular value is not all that important. It could be 90% for all I care, because fees won't start climbing until blocks are nearing max capacity. It needs to be > 50% to make room for the decay function.
The threshold that needs to be truly discussed is the second one. It can't be low enough that an attacker can throw a couple extra BTC at the network and trigger a size growth on the cheap under the right conditions. It can't be so big the network will get clogged with a massive backlog before it resizes. However, an increase in fee subsidy is an increase in revenue, which will translate into an increase mining power eventually. It can be expected that such tight thresholds will result in bursty cap growths, which is another reason for a decay function, but generally I believe we are better with high values than low ones.
