Slow-mode dynamic throttling

So this simplistic feature of linear delays between posting messages is really inconvenient.
Several other websites with chat functions now feature a delay which is variable and responsive.
When users post too fast, the site throttles them. I think discord could really use a similar option.

How it would work according to me: 

Multiple checks keep count of how often users post on a long, medium, and short timeframe, and collectively achieve limits text flow. Line counts and character counts within timeframes count towards ALL checks.

The definition of as used here: "cooldown" = the user stops long enough for the step 3 triggers to wear off.
Short:

1. If a user posts more than 3 lines or 200 characters consecutively in 1 second, the server flags them to be throttle-checked but doesn't throttle them yet, and warns them they are typing too fast.
2. If they post another 2 lines or 100 characters within the next 2 seconds, the site warns them that they are talking too fast and blocks them from sending a message for 2 seconds.
3. If within 3 seconds of the initial block ending, they send another 3 lines or 200 characters, they are warned again and the block time doubles. 
4. After step 3 completes for the first time, steps 2 and 3 are repeated, but each time step 3 is repeated, the number of seconds is doubled one more time on top of previous doublings.
5. each following incident where the initial condition is triggered, if the user has already been throttled once, they don't just get a warning- the initial step results in a throttle of 2 seconds.

Medium:

1. If a user posts more than 5 lines in 6 seconds or 300 characters in 5 seconds, the server warns them.
2. if within 6 seconds of the initial block ending, they send another 5 lines or 300 characters, it throttles them for 2 seconds. 
3. After step 2 completes for the first time, it is repeated, but each time step 2 is repeated, the number of seconds is increased by 10.
4. each sequential time after the initial condition is triggered, the process starts over with the throttle second count in step 2 set to an initial delay of 3 seconds and repeat seconds increase by 15.

Long:

1. If a user posts more than 20 lines in 30 seconds or 1500 characters in 30 seconds, the server warns them.
2. if they post another 20 lines or 1000 characters in the next 30 seconds, the server throttles them for 5 seconds.
3. each time step 2 repeats without going back to step one the server throttles them for one minute and one additional minute for each time it happens and tells them "you are doing that too much".
4. if the user successfully triggers the initial condition more than 4 times in an hour, or is continually throttled for one minute five times in a row without a cooldown, the step 3 throttle time increases to an hour and the warning changes to "you need to go outside".

Automatic Slow-Mode configuration

The exact settings for how much is too much in 1 second, 5 seconds, and 30 seconds is variable and often dependent on the activity in the channel. Moderators could be offered an option to enable some rate limitation globally and select an option to dynamically, automatically enable other rate limiters that influence conversation flow in very active channels during peak interaction periods. This option would be called "automatically enable as needed".

"Automatically enable as needed" observes flow in interaction on a moderately long term timescale and switches limitations on and off in 2 modes, standby and active.

Standby:

• steps 4 and 3 are initially off off on all levels
• no warning is issued for step 1.

If more than 20 people post within a 1 minute timeframe or more than 400 messages are sent to the channel within a 10 minute timeframe, active mode is triggered.

Active:

• steps 3 and 4 are enabled,
• mages count as 2 lines, not 1,
• all initial warnings come with a 2 second delay. 

Active mode returns to standby if less than 20 people post within a 30 minute timeframe or less than 40 messages are sent to the channel within that 30 minute timeframe.