How this practice actually works

Who’s behind the readings

House Call is one editor, three anonymous tradition consultants who built the reference corpus the drafter draws from, and a custom LLM trained on that corpus that produces the first draft of every reading. The consultants did one-time corpus work — sourcing the Vronsky / BaZi / Vedic reference passages, voice samples, and structural conventions the model synthesises against. They’re not per-reading authors. They stay anonymous because they have their own 1:1 practices and were paid for the corpus work, not to be public- facing for House Call.

The practitioner edits and ships every reading — the document you receive. That means: read the model’s draft against the chart math, cross-check every named transit, cut passages that drift into generic Sun-sign prose, rewrite the synthesis where the three traditions strain against each other, write the “short answer” callouts where the chart deserves one. The model is fast; the editor is the reason a House Call reading is specific to your chart instead of generically plausible.

The detailed AI disclosure — what the model can and can’t see, where it fails, how the editor catches it — lives at /how-we-use-ai.

The synthesis method

Three independent traditions compute their views of the same birth chart and the editor (with the LLM drafter as first-pass typesetter) weaves them into a single document. Three lenses, one chart, with the editor doing the interpretive work of saying where they agree and where they honestly disagree.

• Western (Vronsky-school)— classical natal chart math: planets in signs, planets in houses, aspects between them, the chart ruler.
• Chinese BaZi (Four Pillars)— day-pillar day master + the five elements as expressed across year, month, day, hour pillars. Older system; reads structure, not story.
• Vedic Jyotish— sidereal positions with Lahiri ayanamsa, plus Vimshottari Dasha timing. Reads trajectory in 16-20 year arcs.

Where all three agree, the reading speaks at high confidence. Where they diverge, it stays honest about the uncertainty rather than picking a winner.

What the model does (and what it doesn’t)

A custom LLM, trained on the Vronsky / BaZi / Vedic reference corpus plus the editor’s voice samples, drafts the synthesis. It reads the chart math and the curated reference passages keyed to your configuration, then writes a structured first draft of every section. The editor reads, rewrites, and ships it.

What the model specifically does notdo: run the chart math (Swiss-Ephemeris-equivalent positions do that deterministically); decide what’s true (the editor does that, cross-checked against the chart math); serve a reading without human review (a hard rule, the DB schema enforces it). No reading has ever gone out un-edited.

If you’re here because Reddit pointed you at us with the question “isn’t this just AI slop?” — the honest answer is: a generic ChatGPT spitting out Sun-sign prose is slop. A custom LLM trained on a curated multi-tradition corpus, with a human editor reading every word against the chart math, is what makes three-tradition synthesis economically possible at this depth. The speed comes from the model; the specificity comes from the corpus and the editor.

How a reading is actually built

For each commissioned reading the workflow is:

1. Intake. The practitioner collects birth data + context questions from the buyer.
2. Chart math. Tropical positions, Lahiri- ayanamsa sidereal positions, and BaZi four-pillar resolution are computed deterministically from the birth data.
3. Drafting pass (model).The chart math plus the curated reference passages keyed to this chart’s configuration are fed into the custom LLM. It produces a first draft of each tradition section + a synthesis pass + the short-answer callouts.
4. Editorial pass.The editor reads the draft line by line against the chart math. Catches false transit claims, cuts generic Sun-sign prose, rewrites passages where the synthesis strains the source tradition, checks that the three traditions actually converge where the draft says they do, validates that named predictions are falsifiable-by-date. Where the chart genuinely doesn’t support a claim, the claim gets cut — not softened.
5. Voice + format pass. Final pass for voice consistency, structural rhythm, and short-answer callout placement. Then it ships.
6. Customer follow-up. The practitioner runs the customer-facing question + feedback loop; the model drafts the answer against the same corpus; the editor reviews and ships.

Each reading takes 5–14 days end-to-end depending on depth tier — and the time is mostly the editorial pass.The model’s draft lands in minutes; the days are the editor reading every word against the chart math, cross-checking three traditions, and rewriting whatever didn’t hold up. A fast machine draft is cheap; the editorial cost is what the price covers.

What we got wrong in the first five readings

Beta means we’ve been wrong about things. The honest list:

1. Saturn-return timing was off by ~14 months in one reading because the model anchored to the tropical return without weighting the Vedic sidereal lag enough. The editor has since formalised a cross-system timing rule in the drafting prompt that catches this before it ships.
2. BaZi was under-servedin the first two readings — the day master got named but its element interactions didn’t. We added a requirement that the synthesis section anchor on a BaZi paragraph before it’s allowed to ship.
3. The “short answer” calloutat the top of each section was added after a reader said the long-form prose was beautiful but they couldn’t tell what the chart was actually saying. They were right.
4. One reading was too long— 18,000 words where 12,000 would have hit harder. Now the tier word counts are ceilings, not targets.
5. Predictions weren’t falsifiable enough in the first cohort. We now stamp each prediction with a check-by date and an outcome chip (true / false / mixed) so the practice gets feedback on its own accuracy over time.

Where we’re still stuck

• Resolving Western Mercury-retrograde claims against the same period in Vedic dasha terms — the systems sometimes describe the same months in language that looks contradictory but is actually pointing at the same underlying texture. The editor is still finding the vocabulary for this and the drafting prompt still produces awkward seams.
• Drafting against a sparse natal (born outside major aspect windows, weak BaZi element interplay) — the same framework that sings on a busy chart can feel forced on a quieter one. Working on a “quiet-chart” voice that doesn’t over-claim.
• The follow-up question rate-limit (3/week) is honest about team capacity but feels stingy. Considering a per-cohort pool instead.
• Practitioner attribution. Specialists currently stay anonymous (protecting their other client work); we’d like to credit them per-section once they opt in. No firm date for the switch.

What this is not

• Not a daily horoscope generator.
• Not medical, legal, financial, or psychiatric advice.
• Not deterministic predictions about specific events.
• Not a subscription product. Path A is a one-off reading at three depths.
• Not open to everyone yet. Invitation-only while the founders’ cohort runs.
• Not hand-written from a blank page. A custom LLM trained on a curated reference corpus produces the first draft; the practitioner edits, cross-checks, and ships. Both halves of that sentence are load-bearing.

If you want to try the practice

Three soft entry points, easiest first:

1. Free Sun-snapshot — 10 seconds, no signup. Tells you the strongest aspect today’s sky is making to your Sun.
2. Sample reading — three real reading sections, anonymised. The depth + voice you’d get if you booked.
3. Book the full reading — AED 300/750/1,500 depending on tier. Refundable until delivered. Founders’ cohort pricing locks in for the first 25 readers.

Questions about how the practice works — or where you think we’ve got it wrong — are welcome. Reply to any email from us, or write to hello@housecallastro.com (The practitioner runs the customer-facing channel personally).