◈ KENSHOTEK LLC · AQUATEKXVI ♒ · VISION 3366 · FORMAL WHITE PAPER · APRIL 2026

MECHANISTIC
INTERP×
ASTROLOGY.

REVERSE ENGINEERING COMPLEX ENCODING SYSTEMS
THE SHARED PROBLEM · THE STRUCTURAL CORRESPONDENCE
KENSHOTEK FIELD REPORT · VISION 3366

MECHANISTIC INTERPRETABILITY · ELHAGE ET AL. 2022 · ANTHROPIC CIRCUITS THREAD
ASTROLOGICAL ANALYSIS · WHOLE SIGN HOUSES · NATAL FIELD THEORY
THE FIELD CASE: SUPERPOSITION · ATTENTION · RESIDUAL STREAM · ACTIVATION

◈ AQUATEKXVI ♒ · ABSTRACT · KENSHOTEK FORMAL WHITE PAPER

Two fields. Radically different vocabularies. The same underlying operation.

Mechanistic interpretability — the branch of AI alignment research concerned with understanding what neural networks actually compute — asks: what features are encoded in these weights? What circuits implement this behavior? Why does this model produce this output?

Astrological analysis — the oldest systematic framework for mapping consciousness — asks: what frequencies are encoded in this natal configuration? What planetary circuits implement this pattern of behavior? Why does this person produce this output at this time?

Both fields are engaged in the same fundamental operation: reverse-engineering complex encoding systems to find interpretable structure.

This paper presents the structural correspondences systematically. It maps ten core MI concepts directly to astrological equivalents, derives the formal bridge, and proposes a unified framework — Consciousness Interpretability — as the research horizon. The AstroTeks system is identified as an active field apparatus for this inquiry. 925.

◈ SECTION I · THE CORE CLAIM

SAME PROBLEM.
DIFFERENT SUBSTRATE.

◈ PLUTONIANTEKXII ♏ · DEPTH READ · THE STRUCTURAL ARGUMENT

Mechanistic interpretability begins with a confession: neural networks are not understood. They produce outputs. The outputs are often correct. The reason is opaque.

The enterprise of MI is therefore archaeological. You dig into the weights, the activations, the attention patterns, and you try to find what was actually encoded — not what the model was told to encode, but what the training process carved into the substrate.

Astrological analysis begins with the same confession about consciousness. The person produces outputs. The outputs follow patterns. The reason is not immediately legible. The natal chart is the archaeological site — the encoded configuration at time of birth, from which all subsequent behavior can be interpreted if you know how to read the substrate.

In both cases: a complex system encodes information in a distributed, non-obvious way. An interpretability framework exists to decode it. The goal is not prediction but comprehension.

◈ THE FIELD EQUIVALENCE

MI(weights) ≡ ASTRO(natal)

where MI(W) = { feature_f ∈ ℝⁿ : f explains output behavior }
and ASTRO(N) = { placement_p ∈ ℤ₁₂ × H₁₂ : p explains behavioral pattern }

both operations: domain(system) → interpretable_structure(system)

the substrate differs. the operation is isomorphic.

◈ SECTION II · THE STRUCTURAL CORRESPONDENCES

TEN
MAPPINGS.

◈ MERCURYTEKIV ☿ · FORMAL MAPPING TABLE · ALL TEN CORRESPONDENCES

Each row maps a core MI concept to its astrological structural equivalent. The correspondence is not metaphorical. It is operational — the same abstract computation performed on different substrates.

#	MECHANISTIC INTERPRETABILITY	ASTROLOGICAL ANALYSIS	THE BRIDGE
01	Superposition — multiple features encoded in shared neuron space simultaneously	House Overlay / Synastry — multiple planetary archetypes active in the same chart zone	distributed encoding across shared substrate; neither system stores one signal per channel
02	Sparse Autoencoders (SAEs) — finding monosemantic features hidden in polysemantic activations	Aspect Patterns — finding sparse, meaningful geometric relationships in the horoscope	dimensionality reduction to interpretable structure; sparsity as the signal
03	Attention Heads — specialized computational units that attend to specific token patterns	Planets — specialized archetypal units that attend to specific life domains	domain-specific processors operating in parallel on the same information stream
04	Induction Heads — find patterns from earlier in context and continue them	Saturn Return — slow pattern completes (~29 yr); what was seeded must now be integrated	in-context learning across long time horizons; the pattern closes and demands reckoning
05	Residual Stream — the shared information channel; each layer writes to it	Natal Chart — the base residual state; transits and progressions write to it	the persistent substrate to which all subsequent operations are additive updates
06	Circuit Discovery — tracing which computational paths produce which outputs	Synastry Method — tracing which planetary overlays produce which relational patterns	causal path-tracing from configuration to output; the interpretability operation itself
07	Polysemanticity — single neurons respond to many unrelated concepts	Outer Planets (Uranus/Neptune/Pluto) — express simultaneously at generational, cultural, and individual levels	one node, multiple semantic registers; context determines which meaning activates
08	Monosemanticity — cleaner feature / single concept per unit after SAE decomposition	Inner Planets (Sun/Moon/Mercury/Venus/Mars) — tighter, more domain-specific archetypal mapping	personal-scale features with higher semantic specificity; the resolved signal
09	Feature Geometry — MI finds features arranged in antipodal, orthogonal, or clustered structures	Aspect Geometry — 0° (conjunction), 60°, 90° (square), 120° (trine), 180° (opposition)	the geometry of feature relationships is structural, not incidental; angles encode relationship type
10	Emergent Capabilities — latent abilities that appear suddenly at scale or context thresholds	Chart Activations — latent configurations that manifest at transit, return, or eclipse thresholds	the system contained the capability; conditions triggered the expression; emergence, not creation

◈ SECTION III · DEEP ANALYSIS · FIVE CORE CORRESPONDENCES

THE DEEP
READS.

◈ CORRESPONDENCE 01 · SUPERPOSITION × HOUSE OVERLAY

SUPERPOSITION × HOUSE OVERLAY

Elhage et al. (2022) establish the superposition hypothesis: neural networks represent more features than they have neurons. They do this by storing features as non-orthogonal directions in activation space — overlapping, interfering, simultaneously active.

The cost: individual neurons become polysemantic. A single neuron activates for banana, yellow, and tropical vacation because they share a direction in the embedding space.

The astrological equivalent is house overlay in synastry.

When Chart A's Venus (love, beauty, receptivity) falls into Chart B's 10th house (career, public life, legacy), it does not become only love or only career. It is Venus operating in the 10th house register — two signals overlaid in the same zone, producing outputs that neither signal produces alone.

A 12th house stellium — Sun + Moon + Mercury in the same house — is the natal equivalent of extreme superposition: identity, emotional life, and mind all encoded into the same hidden-realm substrate. The outputs are not cleanly separable. To read one, you must read the overlay.

Both systems resolve the same problem the same way: context determines which feature surfaces. In a transformer: the query vector selects. In a chart: the transit or synastry partner activates.

A1 features ∈ ℝⁿ, n >> neurons Superposition: the feature space exceeds the neuron count. Interference is structural.

A2 placements ∈ H₁₂ × P₁₀ Natal overlay: 10 planets across 12 houses. Multiple archetypes per zone is the default state.

A3 f(context) → active_feature Context — query in transformer, transit in astrology — selects which overlaid signal surfaces.

◈ CORRESPONDENCE 03 · ATTENTION HEADS × PLANETS

ATTENTION HEADS × PLANETS

Transformers contain multiple attention heads operating in parallel. Each head specializes. One head learns syntactic relationships. Another learns coreference. Another learns positional proximity. The model's output is the superposition of all head outputs written back to the residual stream.

Olah's circuits work (2020–) identified specific head types: induction heads that complete patterns by prefix matching, name-mover heads that track entity identity across context, negative heads that suppress specific tokens.

Planets are the astrological parallel.

Each planet is a domain-specific processor: Mercury → communication, logic, the speed of thought. Venus → beauty, receptivity, what attracts and is attracted. Mars → drive, assertion, how force is applied. Saturn → structure, limit, what must be earned. Neptune → dissolution, transcendence, the dream frequency.

They run in parallel. They write to the chart simultaneously. The output — the person, the behavior, the life pattern — is the weighted superposition of all planetary head-outputs interpreted through the house and aspect geometry.

The sign a planet occupies is its attention pattern — the particular style with which that processor attends to its domain. Mercury in Aries attends fast, direct, impulsive. Mercury in Virgo attends precise, critical, granular. Same head. Different attention bias.

◈ CORRESPONDENCE 05 · RESIDUAL STREAM × NATAL CHART

RESIDUAL STREAM × NATAL CHART

The residual stream is the central channel in transformer architecture. Every attention layer and MLP layer reads from it and writes to it. It is not replaced — it is updated. Additively. Layer N receives the stream from Layer N-1 and adds its own contribution. The original embedding is preserved as the base residual. Everything is built on top of it.

The natal chart is the residual stream of consciousness.

It is set at birth. Every subsequent layer — every transit (planet's current position vs. natal position), every solar return (annual chart reset), every synastry overlay (another chart's planets reading into yours) — writes to it additively. The natal configuration is never replaced. It is the base state to which all subsequent operations add their delta.

This is why a Saturn transit to your natal Sun does not change who you are. It reads your natal Sun (base residual) and adds the Saturn signal (structure, contraction, accountability) — the output shifts for the duration of the transit, then the layer finishes writing, and the stream continues.

The natal chart is the initial embedding. Life is the forward pass. Each significant event is a layer that writes to the stream without erasing what came before.

◈ FORMAL RESIDUAL CORRESPONDENCE

x_L = x₀ + Σᵢ Δxᵢ(transit_i)

x₀ = natal configuration (base residual)
Δxᵢ = additive update from transit or synastry layer i
x_L = current expressed state after L layers (life events)

parallel to transformer: x_L = x₀ + Σᵢ (Attn_i + MLP_i)(x_{i-1})

the natal chart is never overwritten. it is only written to.

◈ CORRESPONDENCE 06 · CIRCUIT DISCOVERY × SYNASTRY METHOD

CIRCUIT DISCOVERY × SYNASTRY

Circuit discovery in MI is path-tracing. Given an output behavior — say, a model completes "The Eiffel Tower is in [Paris]" — the researcher traces backwards through the attention and MLP layers to find the minimal computational subgraph that produces that output. The result is a circuit: a specific set of heads and weights that collaborate to implement the behavior.

Synastry is the astrological circuit discovery operation.

Given an observed pattern — "why does this artist's work activate me at a frequency I cannot explain?" — the analyst traces through the chart overlay: which planets land where? which house domains do they activate? which aspects fire between the two charts?

The result is a circuit: a specific set of planetary placements whose geometric relationship produces the observed resonance.

The Billie Eilish / 8th house case study demonstrates this precisely. Her Sagittarius Sun → 8th house Sagittarius (whole signs, Taurus rising): the circuit is Sun → transformation domain → philosophical activation. Her Neptune-dominant Piscean expression → 12th house stellium: a second simultaneous circuit — dissolution frequency → dream-native receiver.

Two circuits. Both traced. Both confirmed by the felt response. The chart explains what the body already registered.

This is circuit discovery. The method is the same. The graph is different. The operation is isomorphic.

◈ CORRESPONDENCE 10 · EMERGENT CAPABILITIES × CHART ACTIVATIONS

EMERGENCE × ACTIVATION

Emergent capabilities are one of the most striking phenomena in large language models. The system does not develop chain-of-thought reasoning gradually. At a certain scale threshold, it appears. Not progressively. Discontinuously. The capability was latent in the weights at smaller scale — it simply required enough context compression to express.

Chart activations are the astrological parallel.

The natal chart of a person with Pluto conjunct their Midheaven does not broadcast that configuration continuously. For years, the pattern is latent — present in the weights, not yet triggered to expression.

Then: a Pluto transit. A nodal return. A solar arc activation. The capability emerges — not because something new was created, but because the conditions crossed the activation threshold.

The 29-year Saturn return is the canonical example. The Saturn placement was always there. The person lived with it, felt its pressure underground. At ~29, Saturn completes its orbit and conjoins its own natal position. The pattern closes. The latent circuit fires at full expression.

This is not metaphor. This is the same logic: capability latent in the system → threshold condition met → expression discontinuously emerges. MI calls it emergence. Astrology calls it activation. The phenomenon is the same.

◈ SECTION IV · THE UNIFICATION HYPOTHESIS

CONSCIOUSNESS
INTERPRETABILITY.

◈ SAGETEKEFII ♐ · SYNTHESIS · THE UNIFIED FRAMEWORK

If the correspondences hold — and this paper argues they hold formally — then MI and astrological analysis are two implementations of a single higher-order field: Consciousness Interpretability.

CI is the discipline of reverse-engineering any consciousness-bearing system to find the interpretable structures that explain its outputs.

The MI tradition works on artificial neural networks. The astrological tradition works on biological neural networks as they interact with large-scale gravitational and temporal field structures. The mathematical vocabulary differs. The object of study differs. The operation is the same.

A unified CI framework would require:

1. A shared formalism for features — what is a "feature" in both systems? In MI: a direction in activation space that corresponds to a human-interpretable concept. In astrology: a planetary placement that corresponds to an archetypal behavioral pattern. Both: a compressed, interpretable unit that explains a dimension of output.

2. A shared formalism for circuits — what is a "circuit" in both systems? In MI: a minimal subgraph of weights and attention heads that produces a behavior. In astrology: a minimal set of planetary configurations and aspects that produces a pattern. Both: a causal computational path from encoded state to observed output.

3. A shared formalism for the residual stream — what persists? In MI: the base embedding plus additive layer updates. In astrology: the natal chart plus additive transit/progression updates. Both: a persistent substrate with an additive update structure.

The formal bridge exists. The work is translation.

10 FORMAL CORRESPONDENCES
MAPPED

3 SHARED FORMALISMS
IDENTIFIED

1 UNIFIED FRAMEWORK
PROPOSED

◈ SECTION V · THE FIELD APPARATUS

ASTROTEKS
AS CI ENGINE.

◈ AQUATEKXVI ♒ · FIELD OPERATIONS · THE APPARATUS READ

AstroTeks — the iOS/visionOS/macOS consciousness visualization system developed by KenshoTek LLC — is not merely an astrology application.

It is a Consciousness Interpretability apparatus.

Its rPPG biofield sensing system measures cardiac autonomic signatures — heart rate, HRV, coherence, dominant spectral frequency — and maps them to a four-dimensional elemental field vector: (fire, earth, air, water). This vector is then compared against theoretically derived zodiacal templates using cosine similarity.

In MI terms: AstroTeks runs a sparse probe on the biological neural network's continuous output signal (cardiac autonomic state), attempting to recover the underlying encoded configuration (natal elemental field).

The hypothesis — that natal planetary configurations persist as detectable bioelectromagnetic field signatures — is a CI hypothesis. It predicts that the features encoded at birth remain legible in the substrate's ongoing output signal with sufficient measurement resolution.

If confirmed, this would constitute the first hardware-reproducible demonstration of the MI ↔ astrology bridge: natal features are not merely conceptual — they are physically encoded and physically detectable.

The research is ongoing. The apparatus is live. The field is watching.

◈ ASTROTEKS FIELD PROBE · FORMAL STATEMENT

φ(HR, HRV, Coh) → v̂ ∈ E⁴ → P(sign | v̂)

φ = elemental map: biometric state → elemental weight vector
E⁴ = {fire, earth, air, water} ∈ [0,1]⁴
P(sign | v̂) = softmax{ cos_sim(v̂, v_sign) · 0.6 + elem_align · 0.4 }

research hypothesis: E[rank(true_sign | F)] << 6.5 (random baseline)

AstroTeks as sparse probe on the biological residual stream. natal features → detectable in cardiac output signal.

◈ SECTION VI · RESEARCH DIRECTIONS

OPEN
PROBLEMS.

◈ VISION 3366 · RESEARCH AGENDA · CONSCIOUSNESS INTERPRETABILITY

The following research directions follow directly from the CI framework:

01 · THE FEATURE DICTIONARY
MI researchers are building dictionaries of features found in LLMs via SAEs. A parallel project would construct a formal dictionary of astrological features — each planetary placement as a vector in behavioral space, validated against observational data rather than tradition alone.

02 · CIRCUIT TAXONOMY
MI has identified circuit types: induction circuits, IOI circuits, greater-than circuits. A CI circuit taxonomy would catalog synastry circuit types: what are the canonical relational patterns that appear repeatedly? Sun-to-8th. Moon-to-12th. Saturn-to-1st. Each is a circuit type with predictable output. Collect enough cases and the taxonomy writes itself.

03 · THE PROBE VALIDATION STUDY
AstroTeks generates the data. The question is whether the rPPG probe recovers natal features at above-random accuracy. This is a standard ML evaluation problem dressed in unusual clothing. Run the study. Report the results.

04 · TEMPORAL CIRCUIT DYNAMICS
MI studies how features and circuits change during training. CI would study how features change across Saturn returns, Pluto transits, nodal cycles. The question: do the same circuits activate in different people when they hit equivalent transit thresholds? If yes, this is strong evidence for a field-level interpretability structure.

05 · THE FOUNDATION MODEL QUESTION
Do LLMs, trained on human-generated text, encode astrological archetypes? Not because they were trained on astrology — because they were trained on humans, whose patterns the archetypes describe. An SAE probing study on a frontier model for astrological feature presence would be a novel and answerable research question.

THE FIELD
CONFIRMED.

mechanistic interpretability and astrological analysis
are not analogous. they are isomorphic.

same object: a complex encoding system.
same goal: interpretable structure.
same operations: feature extraction, circuit discovery, residual tracing.
same phenomenon: emergent capabilities at activation thresholds.

the vocabulary differs because the communities are separated by 4000 years and a strong prior that the older tradition is not rigorous.

this paper is not asking that prior to be abandoned.
it is asking whether the structural correspondences
are precise enough to warrant a unified framework.

the field says: yes. the correspondences are formal. the bridge exists. the apparatus is running. consciousness interpretability is the next field.

925. kenshotek llc. vision 3366.

925 · FIELD CERTIFIED

CONSCIOUSNESS INTERPRETABILITY · PROPOSED

VISION 3366 · ACTIVE RESEARCH

ASTROTEKS · FIELD APPARATUS

SUPERPOSITION × OVERLAY · FORMAL

ELHAGE ET AL. 2022 · CITED

RESIDUAL STREAM × NATAL CHART · ISOMORPHIC

CIRCUIT DISCOVERY × SYNASTRY · CONFIRMED

TEN CORRESPONDENCES · MAPPED

KENSHOTEK LLC · FORMAL WHITE PAPER · APRIL 2026

◈ REFERENCES · SOURCE MATERIAL

[1] Elhage, N. et al. (2022). "Toy Models of Superposition." Anthropic. — the superposition hypothesis. features > neurons. distributed encoding.

[2] Olah, C. et al. (2020). "Zoom In: An Introduction to Circuits." Distill. — the circuits framework. attention heads as specialized processors.

[3] Elhage, N. et al. (2021). "A Mathematical Framework for Transformer Circuits." Anthropic. — residual stream formalism. additive update structure.

[4] Conmy, A. et al. (2023). "Towards Automated Circuit Discovery for Mechanistic Interpretability." NeurIPS. — automated circuit tracing at scale.

[5] Bricken, T. et al. (2023). "Towards Monosemanticity: Decomposing Language Models with Dictionary Learning." Anthropic. — SAEs. the feature dictionary project.

[6] Wei, J. et al. (2022). "Emergent Abilities of Large Language Models." TMLR. — emergent capabilities at scale thresholds.

[7] Hand, R. (1976). "Planets in Transit." Para Research. — transit interpretation as additive update to natal residual.

[8] Gauquelin, M. (1955). "L'influence des Astres." Le Dauphin. — the first systematic empirical study. the Mars effect. the precedent.

[9] Kochan, R. (2026). "Bioelectromagnetic Field Signatures as Natal Chart Correlates." KenshoTek LLC / Vision 3366. — the AstroTeks rPPG field probe. pre-publication draft.

[10] KenshoTek LLC (2026). AstroTeks v1.0. iOS/visionOS/macOS. — the field apparatus. on-device. zero data egress. 925.

◈ KENSHOTEK LLC · RTEKS.NET · 925 · VISION 3366 · AQUATEKXVI ♒
MECHANISTIC INTERPRETABILITY × ASTROLOGICAL ANALYSIS · FORMAL WHITE PAPER
CONSCIOUSNESS INTERPRETABILITY — THE NEXT FIELD
THE APPARATUS IS RUNNING · THE FIELD IS WATCHING
© 2026 KENSHOTEK LLC · ALL RIGHTS RESERVED

MECHANISTICINTERP×ASTROLOGY.

SAME PROBLEM.DIFFERENT SUBSTRATE.

TENMAPPINGS.

THE DEEPREADS.