internal/stats/udist.go (9 lines):
	- line 124: // TODO: Actually, it's at most ⌈n*m/2⌉, but
	- line 174: // TODO: ~40% of this function's time is spent in mapassign on
	- line 180: // TODO: The worst case for this function is when there are
	- line 210: // TODO: The n1 and twoU values in the ukeys follow strict
	- line 223: // TODO: Is it possible to generate this table bottom-up? If
	- line 255: // TODO: Later computations depend on these, but these don't
	- line 341: // TODO: Use symmetry to minimize U
	- line 353: // TODO: Minimize U?
	- line 385: // TODO: More precise bounds when there are ties.


internal/stats/ttest.go (4 lines):
	- line 88: // TODO: Can we still do this with n == 1?
	- line 112: // TODO: Can we still do this with n == 1?
	- line 124: // TODO: Can we still do the test?
	- line 141: // TODO: Can we still do the test?


internal/stats/dist.go (3 lines):
	- line 79: // TODO: Add a Support method for finite support distributions? Or
	- line 83: // TODO: Plot method to return a pre-configured Plot object with
	- line 119: // TODO: For discrete distributions, use the step size to


analysis/app/kza.go (3 lines):
	- line 9: // TODO: This all assumes that data is sampled at a regular interval
	- line 16: // TODO: These can generate a lot of garbage. Perhaps the caller
	- line 55: // TODO: Generate less garbage.


benchstat/table.go (1 line):
	- line 88: // TODO: Control this better.


internal/stats/package.go (1 line):
	- line 23: // TODO: Put all errors in the same place and maybe unify them.


benchstat/text.go (1 line):
	- line 211: row.cols = append(row.cols, t.Configs...) // TODO Should this also trim common path prefix? (see toCSV)


analysis/app/compare.go (1 line):
	- line 275: // TODO: If the query is invalid, surface that to the user.