Skatesum uses final rate of graph matrix, which is sometimes totally wrong

[dreeves]

Desiderata

Preview Give feedback

1. Consolidate the notes below and get it all spelled out here in the top-level description
2. Solve the circular dependency
3. Fix the bug
4. Tweet the UVI

HT Grayson:

Aggday skatesum takes only the final rate into consideration when it decides what the cap value is for the plot. If your goal will do anything fancy at all -- anything other than stay at the rate you gave it when you started it -- then the graph will break when you change things and derails will boom into existence.

Skatesum is not officially supported. If you want your graph to have breaks, rate changes, or other not-the-same-rate-everywhere-forever, then skatesum won't work the way you're wanting it to.

Above is from a forum thread.

This is a nontrivial fix -- we can't just change rfin to rcur due to a circular dependency.

See line 592 of beebrain.js:

// HACK: aggday=skatesum needs to know rcur which we won't know until we do
// procParams. We do know rfin so we're making do with that for now...
br.rsk8 = gol.rfin * SID / gol.siru // convert rfin to daily rate

Cognata

• https://forum.beeminder.com/t/how-skatesum-works-was-when-does-a-skatesum-goal-update/7139/5?u=dreev
• https://github.com/beeminder/beeminder/issues/1295
• Skatesum uses rate in rate units instead of daily rate #169
• Possible floating point error with skatesum #250

Verbata: skatesum aggday, #baneful bug,

[dreeves]

I'm not sure if this will be an easy fix -- changing rfin to rcur -- but let's find out!

Oof, this is easier said than done!

Aggdaying happens in procData.
But aggday=skatesum needs to know the current daily rate.
But that's based on rcur which is computed in procParams.
And we do procData before procParams; we don't know rcur yet in procData.
I'm thinking we can get rcur in procRoad and that can happen before procData.

May require thoughtful refactoring.

I did fix a different bug with skatesum along the way: #169

PS: This was broken in Pybrain too so does not count as a zombie.