SPICE GFPOSC is returning incorrect sunrise (and sunset) times for topocentric frames

Question

I cannot get SPICE to produce accurate sunrise times for topocentric frames I specify.

I'm using spiceypy to interface with SPICE. As a first test, I set up topo coordinates for the MARE landing site, mare.tf:

KPL/FK

\begindata

  NAIF_BODY_CODE += ( 3010001 )
  NAIF_BODY_NAME += ( 'MARE-LS' )

  FRAME_MARE-LS_TOPO = 13010001
  FRAME_13010001_NAME = 'MARE-LS_TOPO'
  FRAME_13010001_CLASS = 4
  FRAME_13010001_CLASS_ID = 13010001
  FRAME_13010001_CENTER = 3010001

  OBJECT_301001_FRAME = 'MARE-LS_TOPO'

  TKFRAME_13010001_RELATIVE = 'MOON_ME'
  TKFRAME_13010001_SPEC = 'ANGLES'
  TKFRAME_13010001_UNITS = 'DEGREES'
  TKFRAME_13010001_AXES  = ( 3, 2, 3 )
  TKFRAME_13010001_ANGLES = ( 300, 66.6, 180.0 )

\begintext
   ...

and used pinpoint on the following definition, mare.def,

SPK/FK

\begindata

  SITES    =  ('MARE-LS')
  MARE-LS_FRAME  = 'MOON_ME'
  MARE-LS_CENTER = 301
  MARE-LS_IDCODE = 3010001
  MARE-LS_BOUNDS = ( @1950-JAN-01/00:00,  @2050-JAN-01/00:00 )
  MARE-LS_LATLON = (
                       23.4
                      -60.0
                        0.0
                     )
  MARE-LS_UP     = 'Z'
  MARE-LS_NORTH  = 'X'

\begintext
  ...

with the command

./pinpoint -def mare.def -spk mare.bsp -pck mykernels.bsp

and mykernels.bsp set to

KPL/MK

\begindata
  KERNELS_TO_LOAD = (
    'pck00010.tpc'
    'de432s.bsp'
    'earth_070425_370426_predict.bpc'
    'naif0012.tls'
    'de-403-masses.tpc'
    'moon_pa_de421_1900-2050.bpc'
    'moon_080317.tf'
    'moon_assoc_me.tf'
  )

\begintext
   ...

I've checked and re-checked this, but the sunrise times it gives me for 2021 are clearly wrong:

• 2021 JAN 02 03:45:06
• 2021 JAN 31 18:55:28
• 2021 MAR 02 09:18:43
• 2021 MAR 31 22:28:38
• 2021 APR 30 10:20:47
• 2021 MAY 29 21:11:41
• 2021 JUN 28 07:31:53
• 2021 JUL 27 17:57:18
• 2021 AUG 26 05:01:20
• 2021 SEP 24 17:08:23
• 2021 OCT 24 06:28:44
• 2021 NOV 22 20:55:05
• 2021 DEC 22 12:02:05

My code is here:

    from spiceypy import spiceypy as spice
    import spiceypy.utils.support_types as stypes
    import numpy as np

    spice.furnsh(['mykernels.bsp', 'mare.bsp', 'mare.tf'])

    max_intervals = 1500
    window = year_interval(2021)
    adjust = 0.0
    step_size = 300.0
    min_elevation = 10.0 * np.pi/180.0
    landing_site_frame = 'MARE-LS_TOPO'
    abcorr = 'LT'
    landing_site = 'MARE-LS'


    inc_constraint = np.pi - 2.0 * min_elevation
    daytime_windows = stypes.SPICEDOUBLE_CELL(max_intervals * 2)

    spice.gfposc('SUN', landing_site_frame, abcorr, landing_site,
                 'LATITUDINAL', 'LATITUDE', '>', min_elevation,
                 adjust, step_size, max_intervals, window,
                 daytime_windows)

    for ii in range(spice.wncard(daytime_windows)):
        b, e = spice.wnfetd(daytime_windows, ii)
        print(spice.et2utc(b, 'C', 0, 51))

I've also tried calculating sunrises for Cape Canaveral using the same strategy. For January 1, 2021, I get 13:34:40 UTC, which is 8:34:40 AM local time in Florida. But Google tells me the correct answer is 7:15 AM.

Does anyone know what I might be doing wrong? It must be something relating to the coordinates I'm providing in the home-rolled kernels, but I can't figure out what. Is this just the wrong GF* function to be using? If so, which one should I be using?