Purpose & architecture
A separate consciousness_core.hpp connects the expanded Unified orchestrator to native numerical functions. Its scalar Phi, Psi and related outputs are catalogued with exact formulas and source context. Their names do not automatically give them the meaning of IIT quantities or experimentally validated consciousness measurements.
This subsystem keeps its own source prefix, equations and generation identity. Shared terminology with other builds is cross-referenced without assuming matching state formats or numerical behavior.
This source generation
This record follows the claude/improve-math-logic-ehIW1 branch of Janady13/complete-agi-core at commit 33aa84f2aa61cedbe010240ada7f4efbdfe795a7. The snapshot identifies the source inspected in this audit. It is not a statement about which revision is currently running on a device.
| Field | Source record |
|---|---|
| Revision | 33aa84f2aa61cedbe010240ada7f4efbdfe795a7 |
| Branch | claude/improve-math-logic-ehIW1 |
| Recorded generation | Alternate math branch · native consciousness component |
Recovered implementation structure
These paths locate the components and entry points represented by this build. The source register contains their inspected revisions.
- unified/core/cpp/consciousness_core.hpp
Behavior & verification limits
Source-defined parameters and outputs require independent numerical and runtime validation.
Equations in this build
18 linked mathematical records. Each opens its notation, interpretation and source.
Binary threshold update
A weighted sum of prior binary states is compared with a threshold through a step function H. The result is a binary next state. The exact convention at equality and the direction…
Analog tanh state update
The next analog activation is tanh of a scaled weighted sum plus bias. It lies in (−1,1) for finite real inputs. The recurrent dynamics depend on weight scale and update timing,…
Reciprocal squared-discrepancy score
The denominator adds one to a sum of squared discrepancies across the selected sites. The score is one when every discrepancy is zero and decreases as their total squared…
Eighty-thousand-site design count
Multiplying sixteen families by five thousand sites per family gives eighty thousand sites. This version’s allocation differs from an architecture with five thousand gates total…
C++ unity-proximity score
ComputePhi returns 1/(1+|x−1|), assigning one to x=1 and smaller positive values as x departs from that reference. The function measures scalar proximity to the selected target.…
Gaussian similarity kernel
The kernel is one for identical inputs and decreases with their squared distance. The positive bandwidth σ sets the scale: a distance σ gives exp(−1/2). It is a similarity…
Binary Shannon entropy
The two terms are the uncertainty of a Bernoulli outcome with probabilities p and 1−p. With natural logarithms it is measured in nats, is zero at the deterministic endpoints by…
Layer one to two: capped sigmoid increment
The source expression adds dx times a sigmoid of the current scalar’s weighted value. The body additionally caps the increment magnitude at 0.2 and clips the final value to the…
Layer two to three: product of two differently centered scores
The distance d=|x′−x_c| enters both a unity score Φ(d) and a Gaussian centered at d=0. The Gaussian favors exact agreement, but Φ(d) favors distance one; at exact agreement…
Layer three to four: complex rotating signal
The magnitude is tanh(αC) and the phase advances as ωt. Under the intended nonnegative C, the magnitude is nonnegative and bounded below one. Keeping the complex value retains…
Layer four to five: magnitude response with phase modulation
The body takes sigmoid(w|S|) and multiplies it by 1+0.2sin(arg S). The phase multiplier ranges from 0.8 to 1.2, so the output can exceed one. The source’s phase_alignment…
Layer five to six: logistic scalar, not a one-element softmax
The body returns exp(z)/(1+exp(z)) with z=wP+b, which is a logistic sigmoid. A softmax over a single scalar would always return one; a two-logit softmax over [z,0] would reproduce…
Layer six to seven: leaky accumulator with direct decision input
The body forms increment=Dexp(−F)dt, then returns clamp(0.9Ψ_prev+0.1increment+0.5D). The displayed integral contributes one part of this update, but the decay and extra direct…
Final layer score with a cubic post-transform
The helper first forms r=ΦΨC/(1+F), then returns a clipped cubic r²(3−2r). The cubic is smoothstep and is monotone from zero to one only on r∈[0,1]. The ratio requires F≠−1, and…
Candidate Lyapunov-like objective
The function adds squared deviation from a target Ω*, a weighted squared coherence shortfall, and a linear free-energy term. Nonnegativity requires suitable coefficients and F≥0.…
Finite-difference change in the stability objective
The difference of consecutive V values divided by dt estimates its average rate over one discrete interval. It is not an exact continuous derivative unless an appropriate limiting…
Gradient-flow comment and modified dynamics
For fixed coherence and free energy, the Ω derivative of the displayed objective is 2(Ω−Ω*). The body multiplies its negative by a rate and Ω(1−Ω), then adds further coherence and…
Weighted final score combining base and gate contribution
The result combines 85% of the base output with 15% of the G5K contribution. The weights sum to one, so it is a convex combination when both inputs share a common scalar range.…