We need a quantitative concept of the time value of increasing precision. Like, the idea that a 70% right answer today is more valuable than the 99% right answer a week from now. Can we turn this into a marginal value vs time graph (set of graphs actually, 1 per bit)
Conversation
Replying to
Depends. There's an intermediary company that makes money by offering to pay invoices earlier in return for a moving discount calculated and updated frequently for eg.
2
Replying to
Inflection point of d info d t as a rudimentary cutoff? Maybe with a satisfice threshold coefficient.
1
Replying to
reminds me of the IARPA forecasting challenge described in SUPERFORECASTING, where each day's prediction counts the same (right?).
There is probably a better allocation of these daily points related to what's being predicted... but are any marginal thresholds generalizable?
1
Replying to
You could let people bet on it. A precision futures market.
2
Replying to
This is very applicable to software estimation. People systematically underestimate the opportunity cost of estimation activities themselves. Also, bad estimation assumptions are arguably at least as dangerous as many of the hidden risks that estimates are meant to flesh out.
2
Replying to
Like a "net present value" for accuracy









