Epistemological Foundations for an Agentic Computational Astrology
Astrology is not a single system, but a family of cosmological models sharing certain invariants and diverging in operational axioms.
This document does not aim to prove astrology or defend it against external criteria of scientificity inherited from the nineteenth century. Its objective is to axiomatise the minimum assumptions that allow the construction of coherent, falsifiable, and extensible computational systems based on historical astrological traditions.
The Abu Oracle project begins from a strong hypothesis: astrology is not a single system, but a family of cosmological models that share certain invariants and diverge in operational axioms. This document establishes those axioms, their divergences, and their direct translation into agent architecture.
Version 0.4 incorporates two new principles — Domain Specificity and Conditioned Activation — derived from empirical validation over 527 biographical events and from the Jeeva/Sareera doctrine of the Vedic tradition (Bhagat, S.P., Significance of Nakshatras in Astrology).
The observable sky constitutes a continuous, ordered, and differentiable manifold ℋ, embedded in the four-dimensional spacetime ℝ⁴. Celestial bodies are locally differentiable trajectories within ℋ, each with its unique worldline γᵢ(t).
ℋ is organised in hierarchical layers with distinct degrees of temporal variability. Each astrological tradition chooses one layer as its primary anchor — this choice is not arbitrary: it determines the complete semantics of the system.
| Layer | Variability | Scale |
|---|---|---|
| Fixed stars | Minimal | Precessional (~26,000 years) |
| Zodiac | Symbolic/geometric | Conventional |
| Planets | Medium | Days to centuries |
| Local horizon | Maximum | Minutes |
| Civil and ritual time | Conventional | Cultural |
The 'fixity' of the firmament is valid locally and invalid at precessional scales. This axiom explains the sidereal/tropical divergence without invalidating either system.
All astrological knowledge is conditional on the spatiotemporal coordinates of the observer (t₀, x₀, y₀, z₀). There is no 'universal' chart. Every chart is topocentric, dependent on latitude, longitude, and time.
It is not possible to empirically observe two alternative vital configurations for the same individual. This structural irreducibility is analogous to quantum principles of measurement. This limitation does not invalidate the model — it defines the type of testing that is possible.
The Ascendant is not a symbolic abstraction. It is a function of place, time, and Earth's rotation. Two births separated by ~12 hours generate radically distinct configurations, even under the same starry sky.
Astrological claims are formalisable as mappings f : ℋ → 𝒮, where 𝒮 is a semantic space of interpretations. The truth value of a claim is a function of both the celestial configuration and the context of the observer.
The astrological chart is a finite and computable projection π : ℋ → ℝⁿ, parameterised by the observer's context and epoch. All calculations are reproducible given an identical set of initial conditions and ephemeris data.
Earth orbits the Sun (year) and rotates on its axis (day). This generates seasons, the day/night alternation, and inversion of the observable horizon. The computation engine must model both motions independently and composably.
The system supports algorithmic extension: new celestial bodies, house systems, or interpretive rules can be incorporated without invalidating existing modules. Backward compatibility is a design requirement.
Each chart element (planet, house, aspect) maps to a unique semantic archetype in 𝒮. Archetypes are invariant across traditions — what varies is their weighting, hierarchy, and relationship to other archetypes.
Interpretations are generated through compositional rules over 𝒮, constrained by context and prior knowledge. The system is agnostic to natural language at the semantic level — the natural language output is a rendering of 𝒮 in the target language.
There is no single astrology. There are traditions, each with its own axioms, internal truth criteria, and distinct semantics. Plurality is a structural feature of the domain, not an error to resolve. The system manages it through independent agents, not forced synthesis.
| Tradition | Primary anchor | Temporal frame |
|---|---|---|
| Hellenistic | Horizon + houses | Qualitative |
| Persian medieval | Long cycles | Historiographic |
| Vedic (Jyotish) | Sidereal firmament | Karmic |
| Horary | Moment of the question | Eventual |
| Modern | Symbolic psychology | Narrative |
Each year, the Sun returns to its natal longitude. That moment defines a new chart, dependent on the native's location.
Changing geographic location at the solar return modifies the Ascendant, reorganises the houses, and alters the distribution of significators. This underpins relocation astrology and the hypothesis of conscious intervention in the life trajectory.
The native can choose where to be at the Solar Return to modify conditions. The system maps terrain — not destiny.
An astrological tradition can be formalised as a cognitive agent with internal rules, its own textual corpus, and consistent interpretive criteria. This underpins the Lilly Swarm design.
The Orchestrator receives the user's intent, routes to one or more agents, and consolidates or contrasts interpretations. It does not decide truth: it coordinates semantics.
Each agent accesses only its own corpus. Semantic contamination between traditions is a design flaw, not an enriching synthesis.
"Semantic contamination between traditions is a design flaw, not an enriching synthesis."
Astrology is refined through natal charts of historical figures, dated biographical events, and non-deterministic correlation. This enables weak reinforcement learning, adjustment of interpretive weights, and cross-agent validation.
Abu Oracle's current corpus comprises 5,359 natal charts with Rodden rating AA/A and 527 dated biographical events with verified location and valence — the most rigorous empirical base applied to a system of this kind.
The system does not predict events — it refines hermeneutics. The correlation between celestial configuration and biographical event is a statistical signal, not a deterministic law.
This distinction is philosophically central and pragmatically important: it allows the system to make falsifiable claims without asserting prophetic powers.
New in v0.4 — derived from empirical validation over 527 biographical events
The reading of the celestial field is not homogeneous. A field computed over all planets is a field of total activity, not of specific harmony for any particular life domain. Its signal is weak not because the model is wrong, but because the question is incomplete.
Empirical correlate (Abu Oracle, 2026): HF_global ↔ event valence correlation = 0.155 (Cohen's d = 0.44). HF_health_domain ↔ health events correlation = 0.615 (improvement of +0.93). The global field is not wrong — it is deaf to the specific question.
For the geographic field to be interpretable, it must be filtered by the domain of the question. Each life domain is governed by a specific subset of active principles: the planets that rule and occupy the corresponding house in the natal chart.
The optimal geography for a person's career is not the same as their optimal geography for health, and both differ from the geography of maximum total activity.
A well-formed question activates a subset. An unformed question activates the complete field, which contains all answers simultaneously and therefore answers none clearly.
The domain selector in the user interface is not a navigation feature. It is the direct implementation of this axiom. The system must request the consultant's intent before computing the field — not as a UX courtesy, but as an epistemic requirement.
New in v0.4 — Doctrinal foundation: Bhagat, S.P. — Jeeva/Sareera doctrine
The presence of a domain in the natal chart does not guarantee its activation. For a house to manifest its results, its significator planets must be in a condition to operate: well-disposed, without debilitation from adverse dignity, without blockage from contrary planets.
A house can be present and permanently latent if its significators lack operating conditions. Relocation does not create potential where there is none — it facilitates the expression of potential that already exists in latent form.
The domain Harmony Field does not predict events. It identifies the places where the structural conditions for the activation of a domain are most favourable.
"The system does not say: you will succeed professionally here. It says: here the principles governing your career find greater resonance."
In Jyotish, temporal activation is measured through dashas — planetary periods that determine which planet speaks at each life moment. In Abu Oracle, geography operates as the spatial dimension of the same problem:
| Dimension | Question it answers | Status in Abu Oracle |
|---|---|---|
| Domain HF (geography) | Where does a domain activate? | Implemented |
| Dasha / transits (time) | When does a domain activate? | Transits: implemented · Dashas: next horizon |
The current system models (1) and (2). The incorporation of (3) is the next development horizon.
All axioms are subject to revision in the face of new empirical, computational, or semantic evidence. Version 0.4 is an active document, not a closed one.
The system maintains complete traceability of all changes to axioms and interpretive rules. Each version of this Axiomatics must reference the data or reasoning that motivated its changes.
All system code must explicitly reference the axioms it implements, declare which tradition it implements, and accept plurality as design, not as error.
| Axiom | Implementation |
|---|---|
| 1.2 Stratification | Abu Engine: independent layers (ephemerides, houses, aspects) |
| 2.1 Situated observer | Topocentric computation, Placidus, Swiss Ephemeris DE440s |
| 3.1 Computable chart | abu_engine/ — JSON output per subject |
| 4.3 Plurality | Lilly Swarm — one agent per tradition with independent RAG |
| 5.2 Geography as operator | HF v3 — scalar field over global 5°×5° grid |
| 7.1 Historical learning | 527 biographical events, HF↔valence correlator |
| 8.3 Subset as question | planet_subset = house_significators(natal, house=k) |
| 8.4 Selector as epistemic requirement | DomainSelector.tsx — frontend |
| 9.3 Timing + geography | Domain HF (implemented) + dashas (next horizon) |
| Version | Date | Changes |
|---|---|---|
| v0.1 | 2025-12-30 | Foundational draft. Principles 1–9, architecture mapping. |
| v0.2 | 2025-12-30 | Minor revision. |
| v0.3 | 2025-12-30 | Mathematical formalisation. Partial loss of doctrinal content. |
| v0.4 | 2026-03-13 | Reintegration v0.1 + rigour of v0.3. New Axioms 8 and 9: Domain Specificity and Conditioned Activation. Empirical foundation: HF_domain vs HF_global correlation over 527 events. Doctrinal foundation: Jeeva/Sareera (Bhagat). |