So t recap what this test was, I took an R50 ARC and made sure to hit the stat caps at 334PIE/303DEX. I then switched weapons around for different base damage / delays / levels by choosing different bows and arrows. Some bows had extra ATK bonuses so I used gorgets to balance ATK at static 511. For these tests, I will analyze the results as if the PIE/DEX stat caps have been reached in every case.
Here's my prelim data set:
http://imgbox.com/aawk6uhA
Some of the 'dev' ranges weren't near 8.0% but I felt this is a pretty good data set to eyeball general trends off of.
A lot of your initial thoughts turned out to be spot on. To go over a couple of key points I took out of this:
(1) Auto-Attack / Light Shot seems to be based on the base damage
Being able to play around with both the weapon and arrow combos made it possible to test what happened to light shot if the base damage was the same, but the level of the weapon and the DPS/delay was different. In separate 2 comparisons (one at base 72, other at base 106), the auto-attack damage was the same despite differences in DPS, delay, and level of the weapon. Note that the WS damage on Wide Volley did change. The left chart is a simple plot that shows the relationship between base damage and damage dealt by light shot when stats are capped. I did not force it to intercept 0 just to see how close it would be (it was pretty damn close).
(2) Weapon Skill damage appears to be based on DPS.
Couple things here. First off, when you plot DPS vs. damage dealt by wide volley (curve at the bottom right), you get an incredibly nice linear curve with an intercept very close to 0. This gives very strong evidence that the "DPS" calculated value is the major variable that determines the WS stat caps. We can also remove level of the weapon as a variable because when comparing trial 2 vs. trial 3, Wide Volley damage was higher with a lower level weapon. I think the prelim data set is pretty convincing - DPS is the main modifier for the WS stat caps.
I think the biggest question now is how do the stat caps change with increasing base damage for auto-attack and DPS for WSs. We could brute force our way to the answer by simply running a series of large scale tests similar to Seiken's; however, I think we can save ourselves some time. The curves were very linear. If we can find the "per point increase in damage per +1 stat", we can back calculate the stat caps. I'll try to think of a test methodology for this.
Another question is if all WSs have the same cap. This would take a bit more testing though. From eyeballing, it seems like it should. I remember talking to Valk a couple weeks ago and he noted that all WSs seem to follow similar patterns while auto-attack is separate.
Last question I would offer up is what happens in these scenarios if the stats are NOT capped? Does a higher base damage weapon become more favorable if stats cannot be capped? As your DPS value rises, it becomes more and more difficult to cap stats. Is it worth pushing "DPS" on the weapon if you cannot cap the relevant stats (or require sacrificing too much ATK to do so)?
****
Thinking ahead to how this information can be used to min-max in actual in-game scenarios...
I think the focus of weapon selection should shift towards DPS for melee jobs. The vast majority of your total melee damage comes from WS damage. The only possible exception to this is ARC/BRD, which deals a significant amount of damage output from Light Shot (which includes Barrage). You can potentially make the case for prioritizing Base Damage for ARC/BRD. If we exclude the ARC argument for now, this makes Garuda weapons followed by GC weapons the 2 best options for damage output.
Let me know if you have any critique or thoughts! Pretty interesting stuff that really changes the way we'll be thinking about melee min-maxing.