diary.txt

## 2026 September 20 Sunday

I am nearly done improving `ts.utc()` so that any fraction attached to
its `day` argument is handled with the same caution as fractions
attached to its `second`, which should fix this issue:

https://github.com/skyfielders/python-skyfield/issues/1141

Having nearly worked out the kinks, I was surprised to find that I’d
broken the test `test_velocity_in_ITRF_to_GCRS2()`.  With the old logic,
it receives a time with these attributes, and succeeds:

    .whole          [2459047. 2459047.]
    .tai_fraction   [0.8691782407407408, 0.8691898148148148]
    .tt_fraction    [0.8695507407407408, 0.8695623148148148]

But the new code produces these values and makes it fail:

    .whole          [2459047. 2459047.]
    .tai_fraction   [0.8691782407407407, 0.8691898148148148]
    .tt_fraction    [0.8695507407407407, 0.8695623148148148]

But does this really mean that the new value is worse?  The test is a
finicky one, testing whether the velocity matrix returned from the
no-longer-used routine `ITRF_to_GCRS2()` really predicts how far a
position will spin around the frame’s axis in 1 second.  Does the test’s
failure really mean that the newly rewritten `Timescale._utc()` logic
needs to be tightened up?  Or was it mere coincidence that the old logic
produced a final decimal place that made the test slightly tighter than
otherwise?

Let’s instead judge the new decimal on its own merits.  The time 8 AM is
actually 20h because Julian dates start at noon; and 37 leap seconds had
accumulated by 2020.  So we can ask, what fraction of a day would the
time 20:51:37 AM ideally produce?  Let’s use `dc` to compute it without
the limited precision of machine floating point:

99k 20 24 / p 51 1440 / p 37 86400 / p + + p

The high-precision result is:

.86917824074074074074074074074074074074074074074074074074074074074074\
0740740740740740740740740740739

The `407` looks to be a true repeating decimal.  So the old logic that
produced an `8` at the end was flawed, because the final digit should
have rounded down, not up; and so it’s just a coincidence that the old
logic happened to make this test happier.  I will relax its bounds.