Astrology API Accuracy Benchmark

The RoxyAPI Astrology API Accuracy Benchmark is an open, reproducible benchmark that checks the RoxyAPI astrology and insight API domain by domain against named authorities: NASA JPL Horizons for planets, angles and sidereal positions, and published calendar tables and definitions for calendars, time zones and rule based domains. Every reference value in the RoxyAPI benchmark is cited to its source, every pass band states its reason, and the MIT licensed runner points at any astrology API. The cases are built to break formulas, not to resemble everyday birth data: births on a leap day and inside a daylight saving change, high latitudes such as Reykjavik and Tromso, instants seconds from a sign, nakshatra, tithi or gate boundary, a planet at its station, and calendar year edges such as Li Chun, so an ordinary chart is an easier test than any of them.

For the Sun, Moon, planets and Chiron, RoxyAPI returned 231 of 231 positions within 0.32 arcsec (0.000089°) of NASA JPL Horizons. In the open accuracy benchmark run of 2026-10-09, RoxyAPI returned 2,701 of 2,701 values within tolerance across 17 domains, with a median angular deviation of 0.048 arcsec (0.000013°).

By unit: positions median 0.048 arcsec (0.000013°) over 661; instants median 1.6 seconds over 133; calendar counts median 0 days over 179; discrete values 1,728 of 1,728 exact. Target roxyapi.com, generated from results/latest.json.

Domain	Authority	Values	Within band	Measured deviation
Western planets	NASA JPL Horizons	231	231 of 231	median 0.048, max 0.32 arcsec
Western angles and houses	NASA JPL Horizons sidereal time and obliquity, standard spherical astronomy	200	200 of 200	median 0.12, max 1.5 arcsec
Vedic sidereal chart	NASA JPL Horizons with the Lahiri ayanamsa by its published definition	894	894 of 894	positions median 0.039, max 1.5 arcsec; days max 0.49; 638 exact
Panchang	NASA JPL Horizons Sun and Moon longitudes by the classical definitions, sunrise from the US Naval Observatory	73	73 of 73	13 exact instants; 60 exact
Forecast events	NASA JPL Horizons	29	29 of 29	median 2.3, max 44.9 seconds
Moon phases	U.S. Naval Observatory primary moon phase tables, Universal Time dates	11	11 of 11	11 exact days
Human Design bodygraph	NASA JPL Horizons with the 88 degree solar arc and the Rave Mandala	100	100 of 100	instants median 0.014, max 0.28 seconds; 90 exact
Chinese calendar	Hong Kong Observatory Gregorian-Lunar conversion tables, sexagenary cycle from a published anchor day	135	135 of 135	103 exact days; 32 exact
Feng shui	Printed Qing almanac and Xuan Kong rules recomputed, Hong Kong Observatory Li Chun dates	256	256 of 256	13 exact days; 243 exact
Mesoamerican calendar	GMT correlation 584283, cross-checked against the FAMSI converter	56	56 of 56	16 exact days; 40 exact
Vastu	Brihat Samhita chapter 53 in the Iyer and Kern translations	118	118 of 118	all exact
Numerology	Pythagorean rules recomputed, cross-checked against published worked examples	21	21 of 21	all exact
Kabbalah	Published gematria letter table and the arithmetic Hebrew calendar	72	72 of 72	all exact
Biorhythm	Sine cycles of 23, 28 and 33 days, recomputed from the published definition	70	70 of 70	10 exact days; 60 exact
Ayurveda	NASA JPL Horizons Sun ingress by the classical season rule, sunrise from the US Naval Observatory	105	105 of 105	instants median 1.7, max 29.6 seconds; 24 exact
I Ching	King Wen sequence, cross-checked against Legge and the Unicode Standard	306	306 of 306	all exact
Location	IANA Time Zone Database	24	24 of 24	all exact

The RoxyAPI coverage map: each reference authority, the domain it checks and the quantities compared, generated from the references of every domain in the run.

flowchart LR
  d0("Western planets")
  a0["NASA JPL Horizons"]
  a1["NASA JPL Horizons small-body integration"]
  q0_0["Sun, Moon, planets and Chiron"]
  d1("Western angles and houses")
  a2["Spherical astronomy formulas"]
  q1_0["Ascendant, Midheaven and Placidus cusps"]
  d2("Vedic sidereal chart")
  a3["Lahiri ayanamsa, Calendar Reform Committee definition"]
  a4["IERS Conventions 2010, Table 5.3a"]
  a5["Positional Astronomy Centre, Kolkata"]
  a6["Brihat Parasara Hora Sastra"]
  q2_0["Lahiri ayanamsa and sidereal longitudes"]
  q2_1["Nakshatra and pada"]
  q2_2["Navamsa and Vimshottari"]
  d3("Panchang")
  a7["Surya Siddhanta, Burgess translation"]
  a8["US Naval Observatory Astronomical Applications API"]
  q3_0["Tithi, yoga, karana and vara"]
  q3_1["Sunrise"]
  d4("Forecast events")
  a9["US Naval Observatory seasons"]
  q4_0["Sign ingress instants"]
  q4_1["Station instants"]
  q4_2["Exact transit instant"]
  d5("Moon phases")
  a10["U.S. Naval Observatory Astronomical Applications moon phase service"]
  q5_0["Phase dates"]
  d6("Human Design bodygraph")
  a11["Introduction to the Human Design System"]
  a12["Jovian Archive"]
  a13["Human Design Gates by Zodiac Degrees"]
  q6_0["Design instant"]
  q6_1["Personality and design Sun and Earth gates and lines"]
  q6_2["Profile"]
  d7("Chinese calendar")
  a14["Hong Kong Observatory Gregorian-Lunar calendar conversion table"]
  a15["Sexagenary cycle"]
  a16["Ganzhi"]
  q7_0["Solar terms and Lunar New Year"]
  q7_1["Four pillars"]
  d8("Feng shui")
  a17["Xieji Bianfang Shu"]
  a18["Gujin Tushu Jicheng, Yangzhai Shishu"]
  a19["Shen Shi Xuan Kong Xue"]
  q8_0["Kua number and the Li Chun year boundary"]
  q8_1["Annual flying star plate"]
  q8_2["Natal period, mountain and water stars"]
  d9("Mesoamerican calendar")
  a20["GMT correlation definition"]
  a21["FAMSI date converter"]
  q9_0["Long Count and day count"]
  q9_1["Tzolkin sign and number"]
  q9_2["Haab month and day"]
  d10("Vastu")
  a22["Varahamihira, Brihat Samhita, translated by N. Chidambaram Iyer"]
  a23["Kern, Verspreide Geschriften vol. 2"]
  q10_0["Entrance side and pada from facing bearing and door position"]
  q10_1["Square of the nine by nine plan"]
  q10_2["Classical effect of the entrance pada"]
  d11("Numerology")
  a24["Pythagorean numerology rules"]
  a25["Hans Decoz worked examples"]
  q11_0["Life Path, Expression and Soul Urge"]
  d12("Kabbalah")
  a26["Mispar Hechrachi letter table"]
  a27["Dershowitz and Reingold, Calendrical Calculations"]
  a28["Hebcal date converter"]
  q12_0["Mispar Hechrachi and Mispar Gadol word values"]
  q12_1["Hebrew year, month and day, with the leap year flag"]
  d13("Biorhythm")
  a29["Cleve Moler, Biorhythms, MathWorks blog"]
  a30["US Patent 4,240,153, Biorhythm display device, USPTO"]
  q13_0["Physical, emotional and intellectual cycles"]
  d14("Ayurveda")
  a31["Sushruta Samhita, Bhishagratna translation"]
  a32["Charaka Samhita, Sutrasthana 6"]
  a33["Ashtanga Hridaya, Sutrasthana 1.8"]
  q14_0["Season at a date, both zodiacs and both hemispheres"]
  q14_1["Season opening and closing instants"]
  q14_2["Sunrise and sunset"]
  q14_3["Brahma muhurta window"]
  q14_4["Dosha periods of the day and night"]
  d15("I Ching")
  a34["List of hexagrams of the I Ching"]
  a35["The Unicode Standard, trigram and hexagram blocks"]
  a36["Legge, The Yi King, Sacred Books of the East vol. 16"]
  q15_0["King Wen hexagrams and trigrams"]
  d16("Location")
  a37["IANA Time Zone Database"]
  q16_0["Time zone and UTC offset"]
  a0 --> d0
  a1 --> d0
  d0 --> q0_0
  a0 --> d1
  a2 --> d1
  d1 --> q1_0
  a0 --> d2
  a3 --> d2
  a4 --> d2
  a5 --> d2
  a6 --> d2
  d2 --> q2_0
  d2 --> q2_1
  d2 --> q2_2
  a0 --> d3
  a7 --> d3
  a3 --> d3
  a5 --> d3
  a8 --> d3
  d3 --> q3_0
  d3 --> q3_1
  a0 --> d4
  a9 --> d4
  d4 --> q4_0
  d4 --> q4_1
  d4 --> q4_2
  a10 --> d5
  d5 --> q5_0
  a0 --> d6
  a11 --> d6
  a12 --> d6
  a13 --> d6
  d6 --> q6_0
  d6 --> q6_1
  d6 --> q6_2
  a14 --> d7
  a15 --> d7
  a16 --> d7
  d7 --> q7_0
  d7 --> q7_1
  a17 --> d8
  a18 --> d8
  a19 --> d8
  a14 --> d8
  d8 --> q8_0
  d8 --> q8_1
  d8 --> q8_2
  a20 --> d9
  a21 --> d9
  d9 --> q9_0
  d9 --> q9_1
  d9 --> q9_2
  a22 --> d10
  a23 --> d10
  d10 --> q10_0
  d10 --> q10_1
  d10 --> q10_2
  a24 --> d11
  a25 --> d11
  d11 --> q11_0
  a26 --> d12
  a27 --> d12
  a28 --> d12
  d12 --> q12_0
  d12 --> q12_1
  a29 --> d13
  a30 --> d13
  d13 --> q13_0
  a0 --> d14
  a31 --> d14
  a32 --> d14
  a7 --> d14
  a3 --> d14
  a5 --> d14
  a9 --> d14
  a33 --> d14
  a8 --> d14
  d14 --> q14_0
  d14 --> q14_1
  d14 --> q14_2
  d14 --> q14_3
  d14 --> q14_4
  a34 --> d15
  a35 --> d15
  a36 --> d15
  d15 --> q15_0
  a37 --> d16
  d16 --> q16_0

Latest RoxyAPI accuracy report: roxyapi.github.io/astrology-api-benchmark (PDF, JSON, CSV)

What does not count as an accuracy check

A wrapper checked against the library it wraps is a self-test. It shows that arguments pass through, not that the numbers are right.
A wrapper checked against NASA JPL Horizons measures the library, not the vendor. The agreement belongs to the library authors; the vendor added a network hop.
The widely used calculation libraries are licensed copyleft or sold under a commercial licence. A buyer of an API built on one has to ask which licence applies to the product the buyer ships.
RoxyAPI reads the NASA JPL DE440 ephemeris directly, verified against NASA JPL Horizons. No third party calculation library sits between the ephemeris and the response, so the comparison measures RoxyAPI work end to end.
A check that omits the bodies, dates or charts where its numbers look worse, or publishes only a median, hides its worst case. This benchmark publishes every value it measures, Chiron and the 1879 chart included, with the maximum beside the median.
Planet positions are one slice of accuracy. Angles and houses, sidereal frames, calendars, time zones and discrete rules each fail in their own way and need their own authority, so every domain here has its own folder, source and pass band.

First published here

RoxyAPI published the open Astrology API Accuracy Benchmark first, in April 2026, as the commit history of the repository records. Other astrology API providers have since published accuracy checks of their own, and some have run the RoxyAPI benchmark against their own APIs. An open benchmark is built for exactly that: the method, the references and the runner are public, so anyone can check anyone.

Getting started

You need a RoxyAPI key, from roxyapi.com/pricing, and uv, which also installs the right Python.

1. Install uv (official instructions).

macOS and Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows (PowerShell):

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

2. Clone and install.

git clone https://github.com/RoxyAPI/astrology-api-benchmark.git
cd astrology-api-benchmark
uv sync

3. Add your key. Copy the sample and set BENCHMARK_API_KEY; every other setting has its default. .env.local is gitignored.

cp .env.example .env.local

4. Run every domain. This writes results/latest.json and results/latest.csv, and exits nonzero when any value fails or is missing.

uv run python -m benchmark run

Add --domain western-planets (repeatable) to run one domain.

5. Build the report page, then open dist/index.html in a browser. It is one self contained file.

uv run python -m benchmark site

6. Refresh this README from the new results. The generated blocks are rewritten in place.

uv run python -m benchmark readme

Run against another API

One domain at a time

Many astrology APIs cover a single domain and list it as several products: a natal chart, synastry and daily horoscopes sold as three. The RoxyAPI benchmark counts the way the RoxyAPI API reference does, one domain per distinct body of calculation, so those three are one domain here (Western astrology, checked by the Western planets and Western angles folders), and every domain in the coverage map above is a separate one. To test a single-domain provider, run only the folder that matches what it sells:

uv run python -m benchmark run --domain western-planets
uv run python -m benchmark run --domain western-planets --domain western-angles

--domain takes a folder name under domains/ and repeats; without it every domain runs.

Against another base URL

Set that provider key as BENCHMARK_API_KEY in .env.local. The runner sends it in the X-API-Key header; a provider that expects another one sets BENCHMARK_API_KEY_HEADER too, with no code change.
Adapt the check() function in domains/{id}/check.py for each domain you run. It is the one place a response is read, mapping it to the quantity names in references.json.
Run that domain against the provider and keep its results apart:

uv run python -m benchmark run --target https://api.example.com/v1 --domain western-planets --out runs/example

The references, pass bands and statistics stay exactly as they are, so a single-domain provider is measured by the same numbers as the full RoxyAPI run.

Method

flowchart LR
  src["Reference source: NASA JPL Horizons, USNO, HKO, published rules"] --> pull["domains/id/pull.py"]
  pull --> refs["Cited references.json"]
  refs --> run["Runner calls the API under test"]
  run --> res["results/latest.json"]
  res --> out["README, report page, PDF and release"]
  cron["Monthly scheduled run"] -.-> run

The runner reads committed references only. The scheduled workflow reruns it every month, rewrites the results and republishes the README, the report page, the PDF and the release.

Time. The API receives the local birth time with its UTC offset or IANA zone name, exactly as a user would send it. The reference is computed at the UT instant from standard datetime math. A wrong time zone resolution moves every position, and the Moon, at about 13 degrees a day, shows it first.

Reference. Each domain names its authority and commits its reference values with the source, the retrieval date and the licence. Machine sources such as NASA JPL Horizons are pulled by domains/{id}/pull.py. Rule based domains are recomputed from a published definition, and a second source is transcribed as cross checks the recomputation must reproduce. The runner reads committed files only and never contacts a reference server.

A check. For every case the runner calls the API, reads each quantity and measures the deviation in the unit of that quantity: arcseconds for angles with 0 to 360 wraparound, seconds for instants, days for calendar counts, exact match for discrete values. A value inside its pass band is PASS, outside is FAIL, absent or unreadable is MISSING. Pass bands are vendor neutral floors, never sized to one implementation, so read the largest deviation per quantity for regressions.

Two choices on the planets. Jupiter to Pluto are compared with the Horizons system barycentres, which is what a planetary ephemeris integrates, because the Horizons planet centres come from separate satellite solutions that sit thousands of kilometres off it. The residual left on the oldest chart, from 1879, is the reference frame model: Horizons prints longitudes on the IAU 1976 and 1980 precession and nutation model, which drifts from the current IAU 2006 and 2000 model by about 0.3 arcseconds per century away from the year 2000.

Subjects. The shared subject corpus is data/charts.csv. Named charts carry an AA, A or B Rodden rating from the public Astro-Databank record, so anyone can check the birth data. Synthetic charts exercise time zone edge cases on purpose: daylight saving gaps and overlaps, local mean time before standard zones, half hour offsets, high latitudes and a skipped calendar day. Their value is coverage, not provenance.

Domains

Western planets

Authority: NASA JPL Horizons. Covers: Sun, Moon, planets and Chiron. Endpoints: POST /api/v2/astrology/natal-chart.

Pass bands: Moon 72 arcsec; every other quantity 36 arcsec.

Per quantity

Quantity	Reference	Median	Max	Worst case
Pluto	NASA JPL Horizons	0.056 arcsec	0.32 arcsec	`einstein`
Saturn	NASA JPL Horizons	0.049 arcsec	0.31 arcsec	`einstein`
Neptune	NASA JPL Horizons	0.048 arcsec	0.31 arcsec	`einstein`
Chiron	NASA JPL Horizons small-body integration	0.049 arcsec	0.31 arcsec	`einstein`
Uranus	NASA JPL Horizons	0.048 arcsec	0.31 arcsec	`einstein`
Jupiter	NASA JPL Horizons	0.048 arcsec	0.31 arcsec	`einstein`
Mars	NASA JPL Horizons	0.048 arcsec	0.31 arcsec	`einstein`
Sun	NASA JPL Horizons	0.048 arcsec	0.31 arcsec	`einstein`
Venus	NASA JPL Horizons	0.048 arcsec	0.31 arcsec	`einstein`
Mercury	NASA JPL Horizons	0.048 arcsec	0.3 arcsec	`einstein`
Moon	NASA JPL Horizons	0.048 arcsec	0.27 arcsec	`einstein`