# Zodiac compatibility study data

Data and code behind [Most Compatible Zodiac Signs: All 78 Pairings Ranked](https://www.zodiscope.io/journal/most-compatible-zodiac-signs). Free to reuse with credit to ZodiScope (zodiscope.io).

## 1. Full chart scores of 39,000 made-up couples

Do the Sun-sign rankings hold up when the whole chart is read? For each of the 78 Sun-sign pairings we made up 500 couples (seed 20261119): each person has a random birth date from 1960 to 2005 whose Sun sign matches the pairing, and a random birth city from Austin, Denver, Chicago, Seattle, Miami, Boston, Atlanta, Phoenix, Portland (Oregon), Detroit, Toronto and London. No birth times.

Each couple is scored with the ZodiScope Full chart score as the free calculator at zodiscope.io/compatibility computes it when the birth times are left blank: the Sun, Moon and eight planets at 12:00 local time in each birth city, from NASA JPL's DE440s ephemeris (Skyfield, apparent geocentric, ecliptic of date), aspects between the two charts with the orb caps 10 / 10 / 8 / 7 / 6 degrees (conjunction, opposition, trine, square, sextile), and the same weights as the live scorer (`synastryHarmony`, angles off).

**Check against the live calculator.** On 100 couples already scored by the live service (docs/seo/research/2026-11-03-synastry-score-pairs.csv in the ZodiScope repository, the `noon_no_angles` column), this code gives the same score for 97 and is 1 point away on the other 3 (each within 0.02 of a half point before rounding).

| File | What it is |
|---|---|
| `fullchart_study.py` | The code. `validate` mode reruns the 100-couple check; `study` mode builds the 39,000 couples. |
| `study_results.json` | Summary: overall figures, correlations, groups by how far apart the Sun signs are, and every pairing. |
| `study_per_pairing.csv` | One row per pairing: Sign score, full chart median, mean, standard deviation, shares at 75 or more, under 60 and at 90 or more, and the median and mean with the Sun to Sun contact removed. |
| `study_couples.csv` | Every made-up couple with both birth dates, cities and scores. |
| `validation.json` | The 100-couple check, row by row. |

To rerun (needs `skyfield` and `de440s.bsp` from NAIF):

```
python3 fullchart_study.py validate de440s.bsp 2026-11-03-synastry-score-pairs.csv out
python3 fullchart_study.py study de440s.bsp signs.json out --n 500 --seed 20261119
```

`signs.json` holds every ordered pair's Sign score from the site model (`src/scripts/compatibilityScore.ts`).

**Limits.** The couples are made up and have no birth times, so the Ascendant, Midheaven and exact Moon are left out; with real times, scores move by a median of 6 points (earlier 100-couple test). At 500 couples a pairing, a share's 95% margin of error is at most about 4.4 points. The study compares two scoring methods; it does not test real relationships.

## 2. Sun signs of 85.7 million US births, 1994 to 2014

Daily US birth counts from FiveThirtyEight ([fivethirtyeight/data, births](https://github.com/fivethirtyeight/data/tree/master/births)): the CDC/NCHS file for 1994 to 1999 and the Social Security Administration file for 2000 to 2014. 7,670 days, 85,712,738 births. In the overlap years the SSA file counts about 2% more births than the CDC file.

Each day takes the Sun sign the Sun was in at 12:00 noon US Eastern time, from its exact position (PyEphem, tropical, ecliptic of date). 252 days are days the Sun changed sign, holding 3.32% of births; giving each of those days wholly to either sign moves no sign's share by more than 0.03 percentage points.

A random pairing's chance is pa × pb, doubled for two different signs, where p is a sign's share of births. That treats couples as random draws, which real couples aren't.

| File | What it is |
|---|---|
| `sun_sign_births.py` | The code: `python3 -I sun_sign_births.py CDC_CSV SSA_CSV SCORES_JSON OUT_JSON EPHEM_SITE_DIR` |
| `us_births_by_sun_sign.json` | Births and share by sign, births per day, all 78 pairing chances, band shares, and each sign's chance of a random partner being its best or worst match. |
