It is hard to do any exact investigation, because you can't see Stamina as a number in-game. One approach used by @devsifmatch in early investigations was to take screenshots before and after a note and use a script to calculate lost stamina by examining the area of the gauge that is still coloured. The smaller the max stamina of the formation, the more accurate this could be. An alternative could be installing a cheat engine like GameGuardian to attempt to find and watch the exact memory location of the remaining stamina. — Suyooo 2020/10/14 13:26 UTC

I guess damage gimmick notes might work too, since they appear in the skill log. But do they follow the exact same formula? — Suyooo 2020/10/20 20:34 UTC

Here's a neat sheet that should help finding the correct note damage formula:

https://docs.google.com/spreadsheets/d/1_7xyZ5yeO_ANXYurxXe9EySpF8qr2GQBMeIDFM0GYB0/edit#gid=0

How to use - make a copy, fill in data (copy necklace strengths from the bottom, not from the game). The bordered calculation area below lets you define what should be tried - either FLOOR to round down, or the name of a column header in the Q to W range to multiply by that factor. You can also add new factors there if you think they are needed. Move parts of the calculation around until it hopefully matches all your data. — Suyooo 2020/10/20 20:36 UTC

Idea for the “stamina remaining affects accs” problem in the sheet: make several lines - one for the damage gimmick (where the skill log will tell you what the damage is), and then one for every note before that gimmick note (where you can't enter real damage). Make the “stamina remaining” field of each line calculate itself from the previous calculated damage. This way, a correct calculation for the damage gimmick line should imply all of the note damages are correct as well. — Suyooo 2020/10/22 16:31 UTC

I made another sheet to try the idea above, it is limited to checking early notes in a song though and needs a ton of manual input.

https://docs.google.com/spreadsheets/d/1VhTBUrY_OQt-7Uy0D7xuckFg1-ie4_HFQibA-L6ZIj8/edit#gid=0

I've started porting this into a Python script (so I can also just use the song data from the map DB so I need to input less info), and it seems the order AC > Accs > DR+Type is going into the right direction, but I'm still off 1 or 2 damage points occasionally. I believe this is something in how accessory strength is calculated, as it is correct when just handling Gd-type bonus and DR skills, and even function synthesis wasn't able to find a correct formula. I believe the formula for accessory strength is something like this:

\begin{align*} accBaseStr &= (strAtMinLevel + \frac{strAtMaxLevel - strAtMinLevel}{maxLevel - minLevel} \cdot (currentLevel - minLevel)) \\ accEffectiveStr &= accBaseStr \cdot \frac{currentStamina}{maxStamina} \cdot 3 \end{align*}

…but well, there the search for multiplication order and rounding just extends in there now. There's a few assumptions I've been making:

• Accessory Base Strength is rounded to four decimal places (two in percent format). All skills in the game only define effects to that accuracy, maybe here too…?
• The times 3 is the last step: it comes from the three units in the equipping subunit. The skill is probably independently calculated for each, so the actual skill effect should be just applied thrice. No rounding neccessary after that times 3 either, should be rounded before.
• The stamina factor is only applied after base strength, since I assume base strength is already known at Formation time (by only rarity and skill level), and stamina is the only factor only known at Live time.

None of them lead to a correct result yet though… so either there's some real weird roudning, or one of these assumptions must be off. You can download the script I have so far here. It's a goddamn mess. I'm sorry. If you want to try out something but have no idea how this garbage works, drop me a message and I'll help. — Suyooo 2020/10/25 20:42 UTC

Idea: only use accs that are at minimum level for testing, which knocks off the entire base strength calculation. The first three datasets in the script above only have an SL1 R necklace (0.5% effect), and there's still problems, so there might be something cheeky about that stamina factor. — Suyooo 2020/10/25 20:55 UTC