release: 6.0.0a63 — regenerate golden baselines on DE441 extended kernel

Regenerate every test golden baseline on the extended (DE441) precision tier
with libephemeris 3.0.0rc1, restoring the 9 pre-1550 historical subjects a
prior medium-tier regeneration had silently dropped (41 subjects again, not 32).
No kerykeion computation bug was involved; the diff decomposes into:

- Stale-baseline fixes now captured: the previous baselines predated two
  house-ring fixes (composite MC/IC no longer swapped by re-sorting non-monotone
  cusps; get_planet_house resolves non-monotone rings via the shortest arc, e.g.
  Horizon houses at the equator). Regenerated fixtures satisfy the invariants
  (composite 10th cusp == MC; equatorial-Horizon planets spread across houses).
- Documented libephemeris updates: houses_ex2 now reports true time-derivative
  cusp speeds, so sign-locked systems (Whole Sign/Aries/Krusinski) carry the
  guiding-point (Asc/MC) rate instead of 0; Interpolated-Lilith/osculating-apogee
  drifted ~0.02-0.06deg after the upstream apogee fixes. Report snapshots (which
  track the shipped base/medium kernel) regenerated accordingly.
- Heliocentric charts no longer carry geocentric lunar nodes.

Fix TestDavisonBCE to parse the extended-year ISO string instead of the absent
CompositeSubjectModel.year (latent failure; these BCE tests only run on the
extended kernel, so default-tier CI never surfaced it).

Verified green: full offline suite at default tier (9548 passed) and the CI
--tier=base command (8864 passed); data baselines pass at forced extended tier.

Libephemeris rc updated

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DEPVRxJVsgWGtkBGg6zAbT

Author: g-battaglia

CHANGELOG.md

@@ -2,6 +2,31 @@
 
 ## [Unreleased]
 
+### Changed (6.0.0a63 — golden-baseline regeneration on the DE441 extended kernel)
+
+Regenerated every test golden baseline on the `extended` (DE441) precision tier
+with libephemeris `3.0.0rc1`, restoring the nine pre-1550 historical subjects a
+prior `medium`-tier regeneration had silently dropped (41 subjects again, not 32).
+No kerykeion computation bug was involved; the diff decomposes into:
+
+- **Stale-baseline fixes now captured.** The previous baselines predated two
+  house-ring fixes (composite MC/IC no longer swapped by re-sorting non-monotone
+  cusps; `get_planet_house` resolves non-monotone rings via the shortest arc — e.g.
+  Horizon houses at the equator). The regenerated fixtures encode the corrected
+  (invariant-satisfying) values: composite 10th cusp == MC, and equatorial-Horizon
+  planets distributed across houses instead of collapsed into the 1st.
+- **libephemeris behavioural updates (documented upstream).** `houses_ex2` now
+  reports true time-derivative cusp speeds, so sign-locked systems (Whole Sign,
+  Aries, Krusinski) carry the guiding-point (Asc/MC) rate instead of `0`; the
+  Interpolated-Lilith/osculating-apogee position drifted ~0.02–0.06° after the
+  upstream apogee fixes. Report snapshots (which track the shipped base/medium
+  kernel) were regenerated accordingly.
+- **Heliocentric charts** no longer carry geocentric lunar nodes.
+- **Test fix.** `TestDavisonBCE` derived the era from `davison.year`, but
+  `CompositeSubjectModel` exposes only ISO datetimes; it now parses the
+  extended-year ISO string. These BCE tests only run on the extended kernel, so the
+  latent failure was invisible to the default-tier CI.
+
 ### Fixed (pre-6.0.0 zero-bug review campaign, rounds 26–35)
 
 Ten further review rounds, each rotating a fresh runtime-reproduced lens. Every

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "kerykeion"
-version = "6.0.0a62"
+version = "6.0.0a63"
 authors = [
     { name = "Giacomo Battaglia", email = "kerykeion.astrology@gmail.com" }
 ]
@@ -46,7 +46,7 @@ classifiers = [
 requires-python = ">=3.12"
 dependencies = [
     # TODO before tagging 6.0.0: bump the floor to the stable 3.0.0 once released.
-    "libephemeris>=3.0.0rc1,<4",
+    "libephemeris==3.0.0rc3",
     "pydantic>=2.5",
     "svg-polish>=1.0.0",
     "requests-cache>=1.2.1",

tests/core/test_davison_composite.py

@@ -5,6 +5,20 @@
 from kerykeion import AstrologicalSubjectFactory, CompositeSubjectFactory
 
 
+def _astro_year_from_iso(iso_datetime: str) -> int:
+    """Astronomical year from a kerykeion ISO 8601 extended-year string.
+
+    ``CompositeSubjectModel`` exposes the Davison event date only through its
+    ISO datetime fields (it has no scalar ``year`` attribute like a natal
+    subject). BCE years carry a leading ``-`` (e.g. ``-0072-04-30T...``), so
+    the sign must be stripped before splitting on the date separators.
+    """
+    negative = iso_datetime.startswith("-")
+    year_token = iso_datetime.lstrip("-").split("-", 1)[0]
+    year = int(year_token)
+    return -year if negative else year
+
+
 @pytest.fixture(scope="module")
 def subjects():
     s1 = AstrologicalSubjectFactory.from_birth_data(
@@ -247,7 +261,7 @@ def test_bce_pair_round_trips_to_midpoint_jd(self):
 
         davison = CompositeSubjectFactory(s1, s2).get_davison_composite_subject_model()
 
-        assert davison.year < 1
+        assert _astro_year_from_iso(davison.iso_formatted_utc_datetime) < 1
         err_seconds = abs(davison.julian_day - mid_jd) * 86400.0
         assert err_seconds < 1.0, (
             f"Davison JD {davison.julian_day} is {err_seconds:.1f}s away from "
@@ -261,7 +275,7 @@ def test_mixed_pair_with_bce_midpoint_round_trips(self):
 
         davison = CompositeSubjectFactory(s1, s2).get_davison_composite_subject_model()
 
-        assert davison.year < 1
+        assert _astro_year_from_iso(davison.iso_formatted_utc_datetime) < 1
         assert abs(davison.julian_day - mid_jd) * 86400.0 < 1.0
 
     def test_bce_davison_sun_matches_ephe_at_midpoint_jd(self):