Compare two birth charts in your AI chat, with the math already done

A language model is good at what synastry is mostly about: turning a pattern of tensions and easy meetings into words. It is poor at the step before. Where Venus stood at a given minute over a given city is an ephemeris lookup, not a fact sitting in its training data, and a model asked for it produces a confident degree that is wrong. Two charts double that risk, and one wrong degree changes which contacts exist at all.

The comparison is mechanical too. Ten bodies on one side against ten on the other is a hundred pairings, each either inside the orb its aspect family allows or outside it and therefore absent. The server runs them and prints only the pairs that hold, each with its orb in degrees and arc-minutes.

Both charts in full, first: positions to the arc-minute with sign and house, retrograde markers, the four angles, the twelve house cusps, each person's own aspects and their element and modality balance. Then the inter-aspect list, which exists only once there are two charts: every contact running from one chart to the other, sorted tightest first, each row labeled to say which body belongs to whom. Without it a model mistakes a contact between you for an aspect inside one person's own chart, and the reading drifts into two horoscopes.

After that, the house overlays: where each of your bodies falls among your partner's houses, and where theirs fall among yours. Both directions are printed, each row spelling its direction out in words, because inverting an overlay turns a reading upside down. The text closes with the instructions that keep the answer honest: stay inside the data, work down from the tightest contacts, quote the orb of anything you interpret, never claim an unlisted contact, and write about the couple rather than two people in turn.

Date, place, time, and their agreement. The time is the awkward one: it has to come from a document or a family memory rather than an estimate: a quarter of an hour moves the house cusps enough to change every overlay.

Someone else's birth data is theirs, and a reading built on it will say things about them they never asked to hear. The prompt is text on your clipboard, not a profile on a site, but the conversation goes better when both of you know it is happening.

The tool does not quietly fall back to noon. With that person's toggle on, everything the Earth's rotation decides is removed from their chart rather than hedged: no houses, no Ascendant, no Midheaven, no Part of Fortune, and the Moon given as the span of degrees it covered across the birth day instead of one exact position. The prompt tells the model in as many words not to infer any of it, so you are not handed a rising sign nobody computed.

One consequence is easy to miss: an overlay needs both sets of houses, so a single missing time takes that block down entirely, not by half. What remains is still most of a synastry reading, because planet-to-planet contacts and their orbs do not depend on the clock: the Sun and Moon contacts, the Venus and Mars story and the Saturn pressure points all survive.

It will not hand you a verdict. Synastry describes a texture between two people, not a pass or a fail, and a single figure rating a couple is only a summary laid over the contacts. The prompt asks for the difficult contacts as honestly as the easy ones; an answer that is all good news has skipped half the list, so say so and ask again.

Exact data is also not an exact reading: the positions are computed, but the model still writes from patterns and will sometimes attach a meaning no astrologer would recognize. Make it cite the row and the orb behind every claim, and treat the answer as a dynamic you can test against your experience, never as advice on whether to stay or go.