astronoby 0.9.0 → 0.11.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: becadc2ba01e7620f12b20a5192e8537ed9f717336bb43471734fd5be9561675
-  data.tar.gz: 47c5743867f09c58f49093645da7534def3836135883f44929d4da15f73a441a
+  metadata.gz: 1093c9a0d077cfa9c2a9fcf60e2af2a21ec4c1f74c268ef22dca8e93b5badadb
+  data.tar.gz: e12ccfad49e7a63821cde3b382e8584efdff3c0c06a42ffddf60119ad213b2f5
 SHA512:
-  metadata.gz: 2b04c971a79ac8201e21398dc0cc02b920cdadde538a0eb7d5d59dbc260661cbda51f3a72a434695b596634848f361a96b9533074f399d23d1e3d21fac89d772
-  data.tar.gz: f71973fd392c15a243a28ac09e521e813adf4900eefacfbc55712bd4f942c46b3351cfe44e183c1e809b0285b0ffb22ab320afd26b84c3d329be7fe79897378d
+  metadata.gz: cc73661c433669d3b9efcd816088513a031d7c8c609844988e0ecabed36959e9a361d15d8cbd87905f83bfcb3685640dbad1a7862d624a6895b0b78fe0aa54f7
+  data.tar.gz: 7c544cd27042ec8436d2c11a9f27c6269e039256dadb1ba80a62dac6b3140669420c336df950077aa6a54786158f5c2f3b91744b3cacf67de2003db686ffca37

data/.ruby-version

@@ -1 +1 @@
-3.4.4
+4.0.5

data/CHANGELOG.md

@@ -1,5 +1,240 @@
 # Changelog
 
+## 0.11.0 - 2026-10-31
+
+_If you are upgrading: please see [UPGRADING.md]._
+
+### Bug fixes
+
+* Apply nutation after precession for a `DeepSkyObject` without an ephemeris.
+  The two rotations were composed in the wrong order, so the apparent position
+  was off by up to about 0.1 arcsecond depending on the date, and the velocity
+  had the same error. The rotation now matches ERFA and Skyfield, and is the
+  one used when an ephemeris is given. Without an ephemeris there is still no
+  aberration correction ([#316])
+* Let the rise, transit and set refinement move an estimate back in time. The
+  correction was clamped to be positive, so an estimate that overshot its event
+  could never be brought back, and settings and transits came out up to a few
+  minutes late depending on the window asked for ([#301])
+* Refer equatorial coordinates to the epoch of the frame they come from
+  ([#302])
+* Return real Barycentric Dynamical Time from `Instant#tdb`, which returned
+  Terrestrial Time unchanged. The two scales differ by up to about 1.7
+  milliseconds ([#304])
+* Keep ΔT at its last measured value past the end of the IERS series. `iers`
+  0.3.0 stopped attaching an `available_range` to the `OutOfRangeError` it
+  raises for an uncovered date, and the fallback read that absence as "no data
+  at all" and answered 0. Every date after the bundled series, currently
+  2027-09-04, was off by about 69 seconds ([#304])
+* Read the constellation data files as UTF-8, so that `Constellations::Finder`
+  and `Constellations::Repository` work when the default external encoding is
+  not UTF-8, as under `LANG=C`
+
+### Features
+
+* Add `HorizontalTransformation`, which converts equatorial coordinates at
+  J2000 epoch into horizontal coordinates for one observer and one instant. The
+  rotation and the sidereal time are computed once, which makes converting many
+  points, like the stars of a catalogue, about 70 times faster than going
+  through `DeepSkyObject#at(instant).observed_by(observer).horizontal`. Points
+  are treated as fixed directions, as for a `DeepSkyObject` without an
+  ephemeris. `refraction:` is an option ([#317])
+* Add `AltitudeCrossingCalculator`, which finds when a body reaches any
+  altitude, with unrounded instants, the crossing direction, a grazing flag and
+  the culminations. `refraction:` and `reference:` (`:center`, `:upper_limb`,
+  `:lower_limb`) are options ([#311])
+* Expose lunar eclipse shadow geometry at greatest eclipse and at every contact
+  ([#300])
+* Add lunar eclipse local circumstances with `LunarEclipse#visibility_from`
+  ([#302])
+* Expose the Moon's apparent place at every lunar eclipse contact ([#302])
+
+### Improvements
+
+* Cache Greenwich mean and apparent sidereal times when the cache is enabled.
+  The cache is still disabled by default, which is now stated in the README
+  ([#320])
+* Compute the TAI and TDB time scales with the horologium gem, which owns the
+  scales that convert by definition or by model. UT1 stays in Astronoby
+  ([#304])
+* Read `Instant#tai` and `#tdb` as a `Rational`, which a `Float` Julian Date is
+  too coarse to hold ([#304])
+* Evaluate `Precession` in TT, as SOFA and ERFA do ([#304])
+* Raise the minimum `iers` version from 0.1 to 0.3.1, which `horologium`
+  requires ([#304])
+* Share the rise and set horizon convention through `Astronoby::Horizon`
+  ([#311])
+* Derive a lunar eclipse's magnitudes, axis distance and gamma from its
+  geometry ([#300], [#302])
+* Resolve the lunar eclipse shadow axis offset with vectors ([#302])
+* Reduce the Moon to the observer by vectors for lunar eclipse visibility
+  ([#302])
+* Answer lunar eclipse phase coverage in the words the whole eclipse uses
+  ([#302])
+* Pin CI runner to ubuntu-24.04 ([#312])
+* Fix CI for the standard and simplecov bumps ([#319])
+* Bump rubyzip from 3.4.0 to 3.7.0 by @dependabot ([#285], [#299], [#305],
+  [#308])
+* Bump standard from 1.55.0 to 1.56.0 by @dependabot ([#288])
+* Bump simplecov from 0.22 to 1.0.3 by @dependabot ([#287], [#289], [#291])
+* Bump iers from 0.1.0 to 0.2.0 by @dependabot ([#284], [#294])
+* Update json to 2.21.2 ([#296])
+
+### Backward-incompatible changes
+
+* `TwilightCalculator` is based on `AltitudeCrossingCalculator`. Times change
+  slightly: up to about 30 seconds at the standard angles, and up to about 100
+  seconds with `time_for_zenith_angle` and custom angles. The new values match
+  IMCCE, Horizons and Skyfield. Twilight that exists is returned when the Sun
+  doesn't rise or set, as in a polar night, instead of `nil`. Results are still
+  rounded to the second. `#event_on` is about 1.8 times slower than before,
+  because it solves for the real crossing instead of approximating the Sun's
+  position from a single evaluation ([#311])
+* Equatorial coordinates carry the epoch of the frame they come from ([#302])
+* `Instant#tdb` returns real TDB, and `#tai` and `#tdb` return a `Rational`
+  ([#304])
+* Remove `Constants::TAI_TT_OFFSET` ([#304])
+* `Util::Time.terrestrial_universal_time_delta` is replaced by `DeltaT.at`
+  ([#295])
+* `EclipsePhase` is now `LunarEclipsePhase`, `LunarEclipse#greatest_eclipse` is
+  now `#greatest_eclipse_instant`, `LunarEclipse.new` no longer accepts the
+  magnitudes, gamma or shadow axis distance, and `LunarEclipsePhase#duration` is
+  no longer rounded ([#300], [#302])
+* Remove `LunarEclipseVisibility#altitude_at` ([#302])
+* Lunar eclipse contacts are reduced with the IAU eclipse constant k. Contact
+  times may shift by a second or two, and magnitudes by up to about 0.001
+  ([#302])
+* Bisect to a millisecond rather than to a second when locating an event in
+  time. Conjunction and opposition times may move by a fraction of a second, and
+  searches are about 20% slower ([#302])
+
+**Full Changelog**: https://github.com/rhannequin/astronoby/compare/v0.10.0...v0.11.0
+
+[#284]: https://github.com/rhannequin/astronoby/pull/284
+[#285]: https://github.com/rhannequin/astronoby/pull/285
+[#287]: https://github.com/rhannequin/astronoby/pull/287
+[#288]: https://github.com/rhannequin/astronoby/pull/288
+[#289]: https://github.com/rhannequin/astronoby/pull/289
+[#291]: https://github.com/rhannequin/astronoby/pull/291
+[#294]: https://github.com/rhannequin/astronoby/pull/294
+[#295]: https://github.com/rhannequin/astronoby/pull/295
+[#296]: https://github.com/rhannequin/astronoby/pull/296
+[#299]: https://github.com/rhannequin/astronoby/pull/299
+[#300]: https://github.com/rhannequin/astronoby/pull/300
+[#301]: https://github.com/rhannequin/astronoby/pull/301
+[#302]: https://github.com/rhannequin/astronoby/pull/302
+[#304]: https://github.com/rhannequin/astronoby/pull/304
+[#305]: https://github.com/rhannequin/astronoby/pull/305
+[#308]: https://github.com/rhannequin/astronoby/pull/308
+[#311]: https://github.com/rhannequin/astronoby/pull/311
+[#312]: https://github.com/rhannequin/astronoby/pull/312
+[#316]: https://github.com/rhannequin/astronoby/pull/316
+[#317]: https://github.com/rhannequin/astronoby/pull/317
+[#319]: https://github.com/rhannequin/astronoby/pull/319
+[#320]: https://github.com/rhannequin/astronoby/pull/320
+[UPGRADING.md]: https://github.com/rhannequin/astronoby/blob/main/UPGRADING.md
+
+## 0.10.0 - 2026-06-26
+
+_If you are upgrading: please see [UPGRADING.md]._
+
+### Bug fixes
+
+* Fix constant name typo ([#245])
+* Fix matrix multiplication order in Apparent frame ([#250])
+* Remove spurious N·P from observer ECEF-to-TOD transformation ([#251])
+* Apply aberration in GCRS before precession/nutation rotation ([#253])
+* Fix Moon magnitude ([#254])
+* Fix astrometric position for Deep Sky Objects ([#272])
+
+### Features
+
+* Defer Solar System body computations lazily ([#244])
+* Integrate `iers` gem for Delta T and GMST ([#249])
+* Add IERS polar motion correction to observer ECEF-to-TOD transformation ([#252])
+* Fall back to polynomial GMST for dates outside IERS EOP range ([#255])
+* Add ICRS frame bias to precession matrix ([#257])
+* Extract `EarthRotation` class from `Observer` ([#260])
+* Add TEME reference frame for satellite tracking ([#261])
+* Add ECEF to geodetic conversion (WGS-84) ([#262])
+* Memoize per-instant conversions on `Instant` ([#268])
+* Add `ReferenceFrame#separation_from` for angular separation ([#269])
+* Introduce `Center`, `Body` and `Position` domain objects ([#270])
+* Add planetary phenomena ([#271])
+* Add lunar libration, position angle of axis and bright limb ([#273])
+* Add `Astronoby::Orientation` for arcsecond lunar libration ([#279])
+* Add Lunar eclipses ([#280])
+* Add `Astronoby::Duration` ([#281])
+
+### Improvements
+
+* Optimize nutation loop by eliminating array allocations ([#246])
+* Reuse single Nutation instance in observer matrix computation ([#247])
+* Memoize Observer geodetic computation methods ([#248])
+* Update documentation with recent values ([#256])
+* Update referential frames documentation ([#258])
+* YARD documentation update ([#259])
+* Update zlib, JSON and ERB for security reasons
+* Add references to README
+* Bump actions/checkout from 5 to 7 by @dependabot ([#239], [#276])
+* Bump rubyzip from 3.2.1 to 3.4.0 by @dependabot ([#237], [#267], [#275])
+* Bump irb from 1.15.2 to 1.17.0 by @dependabot ([#242])
+* Bump standard from 1.51.1 to 1.55.0 by @dependabot ([#243], [#274])
+* Bump rake from 13.3.1 to 13.4.2 by @dependabot ([#264])
+* Bump ephem from 0.4.1 to 0.5.0 by @dependabot ([#278])
+* Upgrade default Ruby to 4.0.5 ([#277])
+
+### Backward-incompatible changes
+
+* Minimum Ruby version bumped to 3.2.0 ([#249])
+* Delta T and GMST now use IERS data ([#249])
+* Defer Solar System body computations lazily ([#244])
+* `SolarSystemBody#ephem` is now publicly accessible ([#244])
+* `ReferenceFrame#center_identifier` replaced by `#center` ([#270])
+* `Sun#equation_of_time` now returns a `Duration` ([#281])
+
+**Full Changelog**: https://github.com/rhannequin/astronoby/compare/v0.9.0...v0.10.0
+
+[#237]: https://github.com/rhannequin/astronoby/pull/237
+[#239]: https://github.com/rhannequin/astronoby/pull/239
+[#242]: https://github.com/rhannequin/astronoby/pull/242
+[#243]: https://github.com/rhannequin/astronoby/pull/243
+[#244]: https://github.com/rhannequin/astronoby/pull/244
+[#245]: https://github.com/rhannequin/astronoby/pull/245
+[#246]: https://github.com/rhannequin/astronoby/pull/246
+[#247]: https://github.com/rhannequin/astronoby/pull/247
+[#248]: https://github.com/rhannequin/astronoby/pull/248
+[#249]: https://github.com/rhannequin/astronoby/pull/249
+[#250]: https://github.com/rhannequin/astronoby/pull/250
+[#251]: https://github.com/rhannequin/astronoby/pull/251
+[#252]: https://github.com/rhannequin/astronoby/pull/252
+[#253]: https://github.com/rhannequin/astronoby/pull/253
+[#254]: https://github.com/rhannequin/astronoby/pull/254
+[#255]: https://github.com/rhannequin/astronoby/pull/255
+[#256]: https://github.com/rhannequin/astronoby/pull/256
+[#257]: https://github.com/rhannequin/astronoby/pull/257
+[#258]: https://github.com/rhannequin/astronoby/pull/258
+[#259]: https://github.com/rhannequin/astronoby/pull/259
+[#260]: https://github.com/rhannequin/astronoby/pull/260
+[#261]: https://github.com/rhannequin/astronoby/pull/261
+[#262]: https://github.com/rhannequin/astronoby/pull/262
+[#264]: https://github.com/rhannequin/astronoby/pull/264
+[#267]: https://github.com/rhannequin/astronoby/pull/267
+[#268]: https://github.com/rhannequin/astronoby/pull/268
+[#269]: https://github.com/rhannequin/astronoby/pull/269
+[#270]: https://github.com/rhannequin/astronoby/pull/270
+[#271]: https://github.com/rhannequin/astronoby/pull/271
+[#272]: https://github.com/rhannequin/astronoby/pull/272
+[#273]: https://github.com/rhannequin/astronoby/pull/273
+[#274]: https://github.com/rhannequin/astronoby/pull/274
+[#275]: https://github.com/rhannequin/astronoby/pull/275
+[#276]: https://github.com/rhannequin/astronoby/pull/276
+[#277]: https://github.com/rhannequin/astronoby/pull/277
+[#278]: https://github.com/rhannequin/astronoby/pull/278
+[#279]: https://github.com/rhannequin/astronoby/pull/279
+[#280]: https://github.com/rhannequin/astronoby/pull/280
+[#281]: https://github.com/rhannequin/astronoby/pull/281
+
 ## 0.9.0 - 2025-10-31
 
 _If you are upgrading: please see [UPGRADING.md]._

data/README.md

@@ -11,7 +11,9 @@ Some algorithms are based on the following astrometry books:
   Duffet-Smith and Jonathan Zwart
 
 Solar System bodies' positions are computed based on ephemerides from the
-[IMCCE] or [NASA/JPL].
+[IMCCE] or [NASA/JPL]. The library also supports Earth-orbiting satellite
+tracking through the [TEME] reference frame, suitable for use with SGP4/SDP4
+orbit propagators.
 
 ## Content
 - [Installation](#installation)
@@ -45,6 +47,8 @@ previous versions, you can access the documentation in the README for each
   and objects
 - [Reference Frames](docs/reference_frames.md) - for coordinate systems
 - [Observer Setup](docs/observer.md) - for location configuration
+- [IERS Data](docs/iers.md) - for Delta T and Greenwich Mean Sidereal Time
+  (GMST) data management
 - [Glossary](docs/glossary.md) - for astronomical and technical terms
 
 ### Expected breaking changes notice
@@ -69,6 +73,19 @@ automated guiding of amateur telescopes and some professional use-cases.
 The sources used for comparison are: [IMCCE], [JPL Horizons], [Stellarium],
 and the [Skyfield] library.
 
+## Performance
+
+Astronoby has an internal cache for intermediate results like nutation,
+precession and sidereal time. It is disabled by default. It helps when the same
+computation is repeated at one instant, for example for many bodies or many
+points seen by one observer, and can make it about 2.5 times faster:
+
+```rb
+Astronoby.configuration.cache_enabled = true
+```
+
+See the [configuration documentation](docs/configuration.md) for the details.
+
 ## Development
 
 After checking out the repo, run `bin/setup` to install dependencies. Then, run
@@ -122,6 +139,7 @@ Everyone interacting in the Astronoby project's codebases, issue trackers, chat
 rooms and mailing lists is expected to follow the [code of conduct].
 
 [NASA/JPL]: https://ssd.jpl.nasa.gov/planets/eph_export.html
+[TEME]: https://github.com/rhannequin/astronoby/blob/main/docs/reference_frames.md#teme
 [Bundler]: https://bundler.io
 [`docs/`]: https://github.com/rhannequin/astronoby/blob/main/docs
 [release]: https://github.com/rhannequin/astronoby/releases

data/UPGRADING.md

@@ -7,6 +7,276 @@ changes to it as long as a major version has not been released.
 If you are already using Astronoby and wish to follow the changes to its
 public API, please read the upgrading notes for each release.
 
+## Upgrading from 0.10.0 to 0.11.0
+
+### Lunar eclipse contacts and magnitudes have changed slightly
+
+Lunar eclipse contacts are now reduced with the IAU eclipse constant
+k = 0.2725076 Earth radii, putting the Moon's radius at 1738.09 km, instead of
+its physical equatorial radius of 1737.4 km. This is what IMCCE and NASA reduce
+eclipses with. The atmospheric enlargement of Earth's shadow has been
+recalibrated alongside it, from 1/99 of Earth's radius to a 64 km shell.
+
+The two constants have to move together: either one alone makes agreement with
+IMCCE worse, because the old 1/99 enlargement was partly compensating for the
+smaller lunar radius. Together they cut the contact-time error against IMCCE by
+more than half, and the worst case over the 2026 to 2048 eclipses from 8.9
+seconds to 2.2 seconds.
+
+Contact times may shift by a second or two, and the magnitudes by up to about
+0.001. Greatest eclipse, `#gamma` and `#shadow_axis_distance` are unaffected,
+since they depend on neither constant.
+
+### `EclipsePhase` is now `LunarEclipsePhase`
+
+`Astronoby::EclipsePhase` has been renamed to `Astronoby::LunarEclipsePhase`.
+Its geometry is a `LunarEclipseGeometry`, so the class was never general to
+every kind of eclipse, and the old name reserved a place that solar eclipses
+will need.
+
+### A lunar eclipse and its phases are built from fewer values
+
+`Astronoby::LunarEclipsePhase.new` requires `starting_geometry:` and
+`ending_geometry:`, and `Astronoby::LunarEclipse.new` requires `geometry:`,
+each an `Astronoby::LunarEclipseGeometry`.
+
+`LunarEclipse.new` no longer accepts `umbral_magnitude:`,
+`penumbral_magnitude:`, `gamma:` or `shadow_axis_distance:`. All four are now
+read off the geometry at greatest eclipse, so an eclipse can no longer be
+built holding values that contradict its own geometry. The readers are
+unchanged.
+
+Both objects are built by `Astronoby::LunarEclipseCalculator`, so this only
+affects code constructing them directly.
+
+### `LunarEclipse#greatest_eclipse` is now `#greatest_eclipse_instant`
+
+The method returns an `Astronoby::Instant`, not an eclipse. `#instant` remains
+available and is the same value.
+
+### `LunarEclipsePhase#duration` is no longer rounded
+
+The duration was rounded to whole seconds. It is now measured on the
+terrestrial time scale, which is continuous, so it is unaffected by a leap
+second falling inside the phase, and it keeps the fraction of a second.
+`#duration.seconds` may now return a Float with a fractional part.
+
+### Equatorial coordinates carry the epoch of the frame they came from
+
+`#equatorial` reported the J2000 epoch on every reference frame, including
+frames referred to the equator and equinox of date. Precessing a mean of date,
+apparent, topocentric or TEME place was therefore a no operation rather than
+removing the precession accumulated since J2000, and `GeocentricParallax`
+carried the wrong epoch into anything derived from one.
+
+`MeanOfDate`, `Apparent`, `Topocentric` and `Teme` now report their own
+instant as the epoch, through a new `#epoch` reader on `ReferenceFrame`.
+`Geometric` and `Astrometric` are unchanged, since both are built on the ICRS
+and were already right.
+
+Code passing an of date coordinate to
+`Precession.for_equatorial_coordinates` will now get a precessed result where
+it previously got the coordinate back nearly unchanged.
+
+### Twilight times are computed from altitude crossings
+
+`TwilightCalculator` is now built on `AltitudeCrossingCalculator`, which finds
+the instant the Sun's centre reaches each altitude directly. The previous
+approach was an approximation that could be off by tens of seconds.
+
+Times change slightly: up to about 30 seconds at the standard angles, and up to
+about 100 seconds with `time_for_zenith_angle` and custom angles. The new
+values agree with IMCCE, JPL Horizons and Skyfield. Results are still rounded
+to the second.
+
+When the Sun doesn't rise or set, as in a polar night, the twilight that does
+happen is now returned instead of `nil`. A twilight that never happens, such as
+astronomical twilight at high latitude in midsummer, is still `nil`.
+
+`TwilightCalculator#event_on` is about 1.8 times slower than before, from around
+4 to around 7.5 milliseconds for one day. The previous version approximated the
+Sun's position from a single evaluation, which could be off by up to 30
+seconds. This one solves for the real crossing, evaluating the Sun's position
+about 20 times, and agrees with IMCCE, JPL Horizons and Skyfield to the second.
+
+### Rise, transit and set times are no longer late
+
+The refinement of a rise, transit or set time could only move its estimate
+forward in time. An estimate that overshot its event was never brought back, so
+settings and transits came out up to a few minutes late depending on the window
+asked for. Moon transits now agree with Skyfield to the second. Expect Moon
+times to move by up to a few minutes, and Sun and planet times by a few
+seconds.
+
+### Event times are located more precisely
+
+Bisection accepted a root once its bracket fell under 0.864 seconds, so a
+reported time could sit up to 0.43 seconds from the root of the model it came
+from. It now bisects to under a millisecond. Conjunction and opposition times
+may move by a fraction of a second, and a time that sat near a whole second
+may round to the next one.
+
+A sample landing exactly on a root is also no longer missed, and bisection
+stops if a bracket reaches the resolution of a Float rather than looping.
+
+### New `horologium` dependency, and `iers` raised to 0.3.1
+
+The TAI and TDB time scales are now computed by the
+[`horologium`](https://github.com/rhannequin/horologium) gem, which owns the
+scales that convert by definition or by model. UT1 stays in Astronoby: it
+follows the actual rotation of the Earth rather than a definition, and comes
+from `Astronoby::DeltaT`. Nothing else about time changes, and `Instant#tt`,
+`#to_time`, `#to_date`, `#to_datetime`, `#delta_t`, `#gmst`, `#gast`, `#lmst`
+and `#last` are untouched.
+
+`horologium` requires `iers` 0.3.1, so the minimum `iers` version is raised
+from 0.1 to 0.3.1. One `iers` API Astronoby used did change on the way: `iers`
+0.3.0 stopped attaching an `available_range` to the `OutOfRangeError` that
+`DeltaT.at` raises for a date it cannot answer. Astronoby reads the end of the
+published series directly now, so ΔT past that end still comes back as the last
+measured value rather than 0.
+
+Astronoby pins the precision it asks `horologium` for, so an application that
+calls `Horologium.configure` for its own purposes cannot change what Astronoby
+computes.
+
+### `Instant#tdb` now returns Barycentric Dynamical Time
+
+`tdb` previously returned Terrestrial Time unchanged, documented in the source
+as an approximation. It now returns real TDB, from the Fairhead and Bretagnon
+series that ERFA uses. The two scales differ by up to about 1.7 milliseconds.
+
+```rb
+instant = Astronoby::Instant.from_terrestrial_time(2460796)
+
+# Before
+instant.tdb          # => 2460796
+
+# After
+instant.tdb.to_f     # => 2460796.000000017
+```
+
+`Precession` is now evaluated in TT, as SOFA and ERFA do. It read `tdb` before,
+which was TT under another name, so precession results are unchanged.
+
+### `Instant#tai` and `#tdb` return a `Rational`
+
+Both are read as a `Rational` rather than a `Float`. A Julian Date is around
+2.46 million, so a single `Float` rounds to about 47 microseconds of the day,
+which is more than the whole TDB - TT correction. Call `to_f` where a `Float`
+is what you want.
+
+`tai` is now exactly 32.184 seconds behind TT. It was within a few femtoseconds
+of that for an instant built from an `Integer` or a `Rational`, the offset
+having been built from a `Float` literal, and up to 47 microseconds off for one
+built from a `Float`, where the result came back as a `Float` too. So the type
+changes in that last case, and `tai.to_f` can land one `Float` step from where
+it used to, closer to the true value.
+
+### Drop of `Constants::TAI_TT_OFFSET`
+
+`Astronoby::Constants::TAI_TT_OFFSET` has been removed. The offset between TAI
+and TT is `horologium`'s to hold now, and Astronoby no longer names it.
+
+### `Util::Time.terrestrial_universal_time_delta` replaced by `DeltaT.at`
+
+Delta T (TT - UT1) now has its own namespace instead of living in
+`Astronoby::Util::Time`. The returned value is unchanged.
+
+```rb
+# Before
+Astronoby::Util::Time.terrestrial_universal_time_delta(time)
+
+# After
+Astronoby::DeltaT.at(time)
+```
+
+`Astronoby::Util::Time` keeps `decimal_hour_to_time`.
+
+## Upgrading from 0.9.0 to 0.10.0
+
+### Minimum Ruby version bumped to 3.2.0
+
+The minimum required Ruby version is now 3.2.0 (previously 3.0.0), due to the
+new `iers` gem dependency.
+
+### Delta T and GMST now use IERS data
+
+Delta T (TT - UT1) is now computed from IERS data via the
+[`iers`](https://github.com/rhannequin/iers) gem instead of hardcoded lookup
+tables. Greenwich Mean Sidereal Time (GMST) is now computed using the IERS
+Conventions 2010 ERA-based expression instead of the Duffett-Smith polynomial.
+
+Both changes improve accuracy significantly (Delta T error reduced from ~0.5s
+to ~0.001s, GMST error reduced by 600-2200x vs Skyfield). As a result,
+computed values for sidereal times, horizontal coordinates, and
+rise/transit/set times may differ slightly from previous versions.
+
+The `iers` gem ships with bundled data files that work out of the box. See
+[IERS Data](docs/iers.md) for details on data coverage and updates.
+
+### `SolarSystemBody` initialization is now lazy
+
+`SolarSystemBody#initialize` no longer eagerly computes geometric positions,
+light-time corrections, or Sun data. All computations are now deferred until
+the corresponding method is first called. This means ephemeris-related errors
+will surface on first property access rather than during instantiation.
+
+### `SolarSystemBody#geometric` is now a lazy method
+
+`geometric` was previously exposed via `attr_reader` and eagerly computed in
+`initialize`. It is now a lazy memoized method. The public interface is
+unchanged — `body.geometric` still returns a `Geometric` object.
+
+### `SolarSystemBody#ephem` is now publicly accessible
+
+The `ephem` object passed to `initialize` is now exposed via `attr_reader`.
+
+### `ReferenceFrame#center_identifier` replaced by `#center`
+
+Every reference frame previously exposed `#center_identifier`, an `Integer`
+(e.g. `Astronoby::SolarSystemBody::EARTH`) for barycentric/geocentric frames or
+a `[longitude, latitude]` array for topocentric frames. It is replaced by
+`#center`, which returns an `Astronoby::Center` value object.
+
+```rb
+# Before
+frame.center_identifier # => 399 or [longitude, latitude]
+
+# After
+frame.center                 # => #<Astronoby::Center ...>
+frame.center.geocentric?     # => true
+frame.center == Astronoby::Center.geocentric # => true
+frame.center.observer        # => the Observer, for topocentric frames
+```
+
+Two topocentric centers are equal when their observers share the same
+latitude, longitude and elevation (atmospheric parameters are ignored).
+
+### `Sun#equation_of_time` now returns a `Duration`
+
+Lengths of time are now wrapped in a new `Astronoby::Duration` value object,
+consistent with `Angle`, `Distance` and `Velocity`. As a result,
+`Sun#equation_of_time` returns an `Astronoby::Duration` instead of an
+`Integer` of seconds. Call `#seconds` to get the previous value.
+
+```rb
+# Before
+sun.equation_of_time # => -185
+
+# After
+sun.equation_of_time         # => #<Astronoby::Duration ...>
+sun.equation_of_time.seconds # => -185
+sun.equation_of_time.minutes # => -3.08...
+```
+
+A `Duration` can be built with `from_seconds`, `from_minutes`, `from_hours`
+and `from_days`, and read back in any of those units. It is immutable,
+comparable, and supports `+` and `-`.
+
+Note that `EclipsePhase#duration`, also introduced in this release, returns
+an `Astronoby::Duration` as well.
+
 ## Upgrading from 0.8.0 to 0.9.0
 
 ### Constant `MINUTES_PER_DEGREE` renamed