Can a horary engine reproduce William Lilly's own judgments?

Why Lilly's charts can be used as ground truth

Most astrological claims cannot be checked, because the method is not written down precisely enough to reproduce. Horary is the exception. William Lilly (1602–1681) set out his procedure in Book 2 of Christian Astrology and then applied it to real questions in public — printing the figure, naming the significators, giving his reasoning, and stating a verdict. In many cases he also recorded the outcome.

That gives a rare fixed target. If an engine claims to judge horary by Lilly's rules, it should reach Lilly's verdict on Lilly's charts. Not because he was infallible, but because reproducing a documented judgment is the only way to show the rules were implemented rather than approximated.

What was computed

Seventeen questions from Christian Astrology were encoded as fixtures with Lilly's printed date, time and place, his stated verdict, and his reasoning. Dates were converted from Julian to Gregorian (+10 days for 1582–1700) and 1640s London local mean time was treated as UTC. Each chart was then cast fresh through the production horary service — the same code path a question in the app uses — in Regiomontanus houses, and judged automatically.

The engine sees the chart, not the answer. It assigns significators from the house governing the question, checks the considerations before judgment in Lilly's order, and tests whether the significators perfect by direct application, translation of light, collection, or mutual reception before either changes sign. The verdict is computed from those rules. It is not written by a language model.

The result

All 17 charts cast without error. The engine reached Lilly's recorded verdict on 14 of them — 8 of the 9 he judged yes, and 6 of the 6 he judged no. The three disagreements are listed in full in the table below, and none is a case of the engine failing to run or returning a hedge.

The three misses, and why they stay

**Purchase of Mr. B's houses.** The engine finds an initial promise and then a prohibition, and judges no. Lilly judged yes. The prohibiting body sits at a longitude his 1630s tables placed differently from DE442; on his own figure the prohibition does not occur. This is a disagreement between his ephemeris and modern astronomy, not between his rules and ours.

**The missing dog.** Lilly judged that the dog would return, reasoning from the Sun's applying trine to the lord of the Ascendant. The engine routes a lost-animal question to the 6th house and finds no perfection between the 6th's ruler and the querent's significator. Lilly's own text is inconsistent on whether small beasts take the 6th, and here he reasons from the Ascendant lord instead.

**Attaining the preferment desired.** Lilly judged yes 'with some difficulty', resting on the Moon's placement in the 10th plus a mutual reception between Saturn and Venus. The engine requires a perfecting aspect between the significators and finds none, so it returns no. Whether placement in the house of the matter can substitute for perfection is a rule Lilly half-rejects elsewhere in the same book.

We publish the misses because a validation figure nobody can inspect is not evidence. Each of the three is a documented disagreement about a specific rule or a specific 17th-century longitude, and each is reproducible by anyone who runs the script.

What this does and does not show

It shows that the rules are implemented as written and that the engine reaches classical judgments on classical charts. It does not show that horary predicts the future, and nothing here should be read as such a claim. Whether the sky answers questions is a matter of belief; whether a program follows Lilly's method is a matter of test, and this is the test.

The full fixture set and the runner live in the Shymea backend repository at `astro-calc-service/backend/tests/run_lilly_charts.py`. The table below is downloadable as CSV under CC BY 4.0. The byline is Shymea Editorial; no astrologer is named as author, because the figures are the output of a program, not a person's opinion.