Most astrology apps generate interpretations by feeding a birth date to a language model and hoping for the best. The output reads well. It also reads the same way for everyone born within a two-week window. That is not forecasting. That is content generation wearing a costume.
StellaCarta takes a fundamentally different approach. Every forecast begins with astronomical calculations that have nothing to do with AI, passes through a deterministic verification layer, and only then reaches the interpretation stage. Here is how the pipeline works, and why the order matters.
The Calculation Layer: Swiss Ephemeris
The foundation of every StellaCarta report is the Swiss Ephemeris, the same astronomical library used by research institutions and planetariums since 1997. It computes planetary positions to sub-arcsecond precision using NASA's JPL Development Ephemeris data.
This is not a lookup table or an approximation. The Swiss Ephemeris solves the gravitational n-body problem for each query, accounting for perturbations, precession, and nutation. When StellaCarta reports that Saturn is at 0.43 degrees Aries on February 14, 2026, that position has been computed against the same dataset used by the Jet Propulsion Laboratory to navigate spacecraft.
Core architectural principle: AI interprets. AI never calculates. All planetary positions, house cusps, aspect geometry, and transit timing come from the Swiss Ephemeris. The AI layer receives pre-calculated data and produces interpretations from it.
This separation matters because language models hallucinate. They are excellent at pattern recognition and synthesis, but unreliable for precise numerical computation. By keeping calculations deterministic and feeding verified data to the AI, we eliminate the most common source of error in AI-generated astrological content.
The 4-Layer Pipeline
Every StellaCarta report passes through four sequential layers. Each layer validates the output of the previous one before proceeding.
Layer 1: Data Integrity
The engine computes the natal chart, current transits, annual profections, and 180-day transit ranges with retrograde detection. Every position is stored with full precision. Aspect geometry is calculated with exact orbs, not rounded approximations. A Mercury-Saturn sextile at 0.10 degrees is recorded as exactly that, not "Mercury sextile Saturn."
Layer 2: Chart Physics
This is where StellaCarta departs from traditional software. The dominance scoring system evaluates every planet across seven factors:
- Essential dignity (domicile, exaltation, detriment, fall)
- Accidental dignity (angular houses, speed, direction)
- Aspect count and quality (tight aspects weighted more than wide ones)
- House rulership load (how many houses the planet rules)
- Sect alignment (day chart vs. night chart advantage)
- Speed relative to mean (fast planets carry more agency)
- Dispositor chain position (final dispositors outrank intermediaries)
Each planet receives a composite score and is assigned to a tier: Dominant, Strong, Moderate, or Background. This hierarchy prevents the AI from treating all planets as equally important, which is the single most common failure in automated chart interpretation.
Layer 3: Synthesis
Two independent AI models, Claude and GPT, receive identical structured data: the dominance ranking, dispositor chains, sect analysis, aspect tiers, and pattern quality assessments. Both models follow the same 10 hard rules that constrain their output:
- Never contradict the dominance hierarchy
- Never invent aspects not present in the data
- Never exceed 12 major claims per section
- Every claim must cite a specific chart factor with exact degree
- Confidence grades must match transit orb tightness
The models engage in a Socratic debate process, challenging each other's interpretations against the source data. A deterministic audit gate then validates every claim. Entities that fail the audit are excluded before the final synthesis.
Layer 4: Decision Intelligence
The final layer translates astrological synthesis into actionable decision frameworks. Both models generate domain-specific decision memos (career, finance, relationships, health, personal development), scenario forecasts (base case, bull case, bear case), and red team analysis that stress-tests the primary narrative.
All transit citations in this layer are validated against engine-computed 180-day ranges. If a model claims "Saturn conjunct natal Mars peaks in March 2026," the validator checks whether the engine-computed orb entry, exact date, and exit date support that claim. Citations that fail are flagged and removed.
Why This Is Different
Traditional astrology apps treat AI as a black box: birth data goes in, interpretation comes out. There is no intermediate validation, no dominance hierarchy, and no cross-model verification.
StellaCarta treats AI as one component in a larger engineering system. The Swiss Ephemeris provides ground truth. The dominance rubric provides analytical structure. The dual-model debate provides interpretive diversity. The audit gates provide quality control. The transit validator provides temporal accuracy.
The result is a report where every claim traces back to a specific planetary position, computed to astronomical precision, weighted by a transparent scoring rubric, validated by a deterministic checker, and stress-tested by an adversarial model.
That is not mysticism. That is engineering applied to an ancient observation system.