If you've ever paid for an "AI horoscope" and ended up with 200 words of generic affirmation, you should know what makes Kismets different before you spend $29. This page is the long-form answer to that.
The five steps, end to end
Bazi computation — deterministic, not LLM-guessed
Your birth date, time, and gender go into a classical Bazi engine. We use lunar-javascript, an open-source astronomical calendar library that has been independently maintained for years. It outputs your Eight Characters (年柱 月柱 日柱 时柱), your Day Master, your Five Element distribution, your Ten Gods configuration, and your Eight Luck Pillars with start/end ages computed correctly for your gender.
This step is pure math. No LLM, no probability, no interpretation. If you input the same data, you get the same chart, every time. The same chart we compute matches what any reputable classical Bazi practitioner would produce.
A specific prompt — written, iterated, and tested
The Bazi chart data is then handed to Claude Sonnet 4.6 along with a prompt we wrote and have iterated through multiple versions. The current version is about 16 KB of instructions covering: report structure (five chapters, in a specific order), voice rules (conclusion-first, no fortune-telling language, terminology must be translated to felt-sense scenes), what to never say (no event predictions with dates, no medical claims, no name predictions, no specific dollar amounts), and what counts as "passing" each paragraph.
The prompt enforces what we call the three-part rule: every time a Bazi term is used (印, 官, 食伤, Day Master, Luck Pillar), it must immediately be followed by (a) a modern psychological translation and (b) a concrete behavioral scene the reader could verify against their own life. This is what makes a Kismets reading feel specific instead of generic.
A regex quality gate — catches the boring failures
Before any report reaches you, it goes through an automated quality gate. The gate is deterministic regex checks against the markdown, looking for:
- Structural completeness — all five chapters present, the one-page summary has the required anchor sentence + watch/avoid sections, the disclaimer is present
- Length — the report is in the expected 5000-7500 word range (not truncated by an API timeout)
- Hard legal rules — no medical condition names, no specific year + event predictions, no surname predictions, no specific monetary predictions
- Chart specificity — your Day Master must actually appear in the text at least twice (catches the case where the model drifts off your chart)
If the gate blocks the report, an editor (a real person) is notified by email with the specific problems flagged. The report doesn't ship until those are fixed.
Editor review — for the things gates can't catch
Most reports pass the gate clean. But edge cases — reports where the AI took a creative liberty that's technically legal but tonally off, or where a Luck Pillar interpretation is plausible but worth a second read — get a human editor's eyes before delivery.
The editor doesn't rewrite for style. The editor catches: factual drift from the chart, voice misalignment, anywhere the reading would feel like generic horoscope instead of structural diagnosis. If something needs revision, the report is regenerated with corrections before it goes to you.
PDF delivery — 48 hours, hand-checked
The final markdown is rendered into a branded PDF (Fraunces serif body, A4, ~8–14 pages depending on your chart's complexity) and emailed to you. From the moment you submit your intake form to the moment the PDF arrives in your inbox is typically under 48 hours, often under one. The 48 hours exists because of step 4 — we'd rather hold a report for editor review than ship something that misses the bar.
Common questions
So you're just using ChatGPT?
No. We use Claude Sonnet 4.6 (Anthropic's model), and what makes the output meaningfully different from raw ChatGPT is the prompt + the chart data + the quality gate + the editor. If you ask ChatGPT "tell me about my Bazi" with just your birth date, you'll get something generic. We give the model a full computed chart (8 characters + 5 element counts + 10 Ten Gods + 8 Luck Pillars + current annual flow), plus 16 KB of instructions on how to write specifically and what to never say. That's a different output.
Why $29? Most readings are $100-500.
Because our marginal cost per reading is around $3 (chart engine + LLM tokens + PDF rendering + email delivery + editor time). We could charge $200 and most of you wouldn't blink, but we'd rather build a reading you can give as a gift to a friend who's curious but not invested, instead of a reading you only buy when you're in crisis.
What if I think the reading is wrong?
Reply to the delivery email. We'll either revise it, clarify it, or refund — case by case, no script. See our refund policy for the formal version.
Will the same person get the same reading?
The chart computation is deterministic, so the structural facts (which Day Master, which Luck Pillar, etc.) will always match. The long-form writing is generated fresh per order, so the prose isn't word-for-word identical, but the diagnosis it points to is the same. If you ordered two reports for yourself, the substance would match; the sentences would differ.
Is this fortune-telling?
No. We don't predict specific events with dates. We don't tell you who you'll marry or what year you'll get rich. What we do is describe the structural patterns of how you're built and what kind of decade you're in — patterns you can verify against your actual life, and use to make better decisions. Some readers find this more useful than predictions. Others want predictions. If you want predictions, Kismets isn't the right product.
Why publish all this
The astrology / divination market is full of products that won't tell you how they're made. Some use templates filled in with your sign. Some use full LLM hallucination. Some are real human practitioners working from real charts. You usually can't tell which from the marketing.
We think transparency is part of trust. If you read this page and decide the method isn't for you, that's a clean outcome. If you read this page and the method increases your confidence in the product, also a clean outcome. Either way, you're choosing with information you couldn't get from most competitors.
Try the free version first.
Find your Day Master + four pillars + current Luck Pillar in 10 seconds. Same engine. No signup wall.