The General Overview Prompt: Why Synthesizing 12 Houses Is Harder Than It Looks
When the 12 energy texts are ready, it seems like all that's left is to tell the model "read these and write a portrait." In practice — a week of work and a few dozen euros on testing. Not because the task is technically hard, but because LLMs do three things by default that kill the quality of an astrological text.
Problem 1: LLMs Flatter
This is the main problem, and it's systemic.
I'm from the generation that has Saturn in Capricorn — formally a strong position, domicile. When I ran my chart and the charts of friends my age through early prompt versions, the result was always the same. Saturn in the 10th — "a born career achiever, sees things through to the end." Saturn in the 6th — "a tireless workhorse, carries projects single-handedly." Everyone across the board — "disciplined, unbending, goal-oriented."
In reality, half of these people are lazy asses (me), the other half are beaten-down neurotics crushed by life and circumstances (also me). Not because Saturn in Capricorn is a bad placement. But because the LLM only describes the "internal" mode — when the person has already integrated the energy and uses it as a tool. The "external" mode — when that same energy is felt as pressure, obligation, constraints — is simply ignored.
Saturn domicile ≠ "you are disciplined." It's a quality of pressure (from outside) or a quality of instrument (from inside). For most of their lives, most people live in the first mode.
The solution — an anti-compliment rule in the prompt: malefics (Saturn, Mars, Pluto) always start with the external mode. Pressure first, then the transition to instrument. Not the other way around. If a malefic in a strong position reads like a compliment — the external mode is lost, rewrite it.
Problem 2: The Model Tries to Say Everything
12 energy texts are 6,000–18,000 words of input data. Each contains rulers, aspects, tensions, supports, connections, contradictions. The model dutifully tries to mention everything, the text bloats, and if you raise the token limit — it bloats with filler and still cuts off mid-sentence.
The astrological style suffered especially. It's meant to be educational — with explanations of mechanics, examples and counterexamples. The model dives deep into one topic, elaborates at length on why Mars in exaltation works differently from Mars in domicile, draws a comparison — and there's no room left for the other topics.
The solution — a hard limit: 3 topics for the astro style, 3–4 for narrative. No more. The rule: a complete text on 3 topics is better than a truncated one on 4. If at 80% of the budget the last topic hasn't started — skip it and close the text.
And also — deduplication. The same configuration (for example, that same Saturn in Capricorn) is mentioned in the analyses of several houses. The prompt requires: describe the configuration once, then show its manifestations across different areas. Not "in career Saturn gives discipline, in relationships Saturn gives discipline, in health Saturn gives discipline."
Problem 3: The Model Starts With the Wrong Thing
Early versions began the portrait with the "loudest" element of the chart — usually a stellium or a tense configuration. Stellium in the 10th house mentioned in 8 out of 12 analyses? Then that's where we start.
But a stellium in the 10th isn't personality — it's circumstances. The person is their Ascendant, Sun, Moon, first house. A stellium is the conditions this person lives in. And the difference is enormous: the same career stellium is a completely different story for someone with Aries Ascendant (fire amplifies fire) versus someone with Pisces Ascendant (a dreamer shoved into a rigid structure by life).
The solution — an "identity first" rule: the first topic is always who you are (ASC, Sun, Moon, 1st house). Everything else is viewed through the lens of this personality.
Two Response Styles
Roughly midway through development, I started showing results to friends. And I realized that most of them had no idea what they were reading. "Mars in exaltation in the 10th house" means nothing to someone who hasn't studied astrology.
That's how the narrative style was born: the same analysis, the same conclusions, but without a single astrological term. Zero. No planets, houses, aspects, signs. The text reads like a chapter from a book about a person — behavior, habits, inner conflicts, recognizable situations.
ChatGPT also had a habit of spawning headers and subheaders with two-sentence paragraphs inside. The "book chapter" format — continuous text, no headers, no bullets — turned out to be both more pleasant to read and more honest in content.
The astrological style stayed for those interested in the mechanics: why the system reached a particular conclusion, what rules stand behind it, how it works in other charts. Essentially — an astrology textbook using your own chart as the example.
Both styles are generated from the same energy descriptions, through the same pipeline. The difference is only in the final synthesizer prompt.