Equation Ledger#

This ledger preserves mathematical formulas, research-state rules, and discarded-but-important ideas from the NFC pivot context. Status labels are conservative: implemented, experimental, scaffolded, theory only, archived, deferred, falsified, or needs evidence.

equation_id

equation

category

source/context

current project status

where included in final dossier

notes

EQ-001

Attention(Q,K,V)=softmax(QK^T/sqrt(d))V

operator learning

Transformer baseline

implemented baseline

this ledger, master dossier

Baseline math, not NFC novelty.

EQ-002

h_{t+1}=A(x_t)h_t+B(x_t)x_t

operator learning

Mamba/selective SSM schematic

deferred baseline

this ledger, master dossier

Mamba is not wired as a NeuroTwin backbone.

EQ-003

F_s(x,t,omega) in R^d

latent field definition

NFC primitive

experimental

constitution, master dossier

Core NFC primitive.

EQ-004

Y_m=O_m(F_s,A_s,U,epsilon_m)

observation operators

NFC observation model

experimental

constitution, master dossier

Replaces direct-fusion framing.

EQ-005

`p(Y_{1:M}

U,A_s)=int p(F

U,A_s) prod_m p(Y_m

F,A_s) dF`

generative probabilistic model

NFC generative statement

EQ-006

F(t+Delta)=Phi_{theta,Delta}(F(t),U_[t-H,t],A_s)

controlled dynamics

NFC dynamics

experimental

constitution, master dossier

Implemented only as simplified neural modules.

EQ-007

dF_t=f_theta(F_t)dt+g_theta(F_t)dU_t

controlled dynamics

Neural CDE-style update

theory only

master dossier

Preserved as future dynamics option.

EQ-008

tau partial_t F(x,t)=-F(x,t)+int_Omega K_theta(x,x',t)sigma(F(x',t))dx'+B_theta U(t)+eta(x,t)

neural-field PDE / integro-differential dynamics

NFC field equation

theory only

constitution, master dossier

Guides model hierarchy.

EQ-009

Z_{t+1}=Z_t+Delta t[-DZ_t+K_t sigma(Z_t)+BU_t]+xi_t

controlled dynamics

Discretized field dynamics

experimental

constitution, master dossier

Current implementation is a simplified form.

EQ-010

K_t approx U_t V_t^T

low-rank pair kernel

Pair-kernel approximation

implemented experimental

constitution, master dossier

Implemented as low-rank pair update.

EQ-011

M_t=softmax((U_t V_t^T)/sqrt(r)+S)

low-rank pair kernel

Pair influence

implemented experimental

constitution, master dossier

Structural prior S is optional.

EQ-012

Z'_t=Z_t+M_t Z_t W

low-rank pair kernel

Node update

implemented experimental

constitution, master dossier

Pair-Operator survives as ablation idea.

EQ-013

NPU_ij(t)=E[abs(K_ij(t)-Khat_ij(t))]

uncertainty / calibration

Neural pair uncertainty

scaffolded

ledger, master dossier

Pair uncertainty exists only in limited form.

EQ-014

nabla_w F(i,j)=sqrt(w_ij)(F_j-F_i)

graph calculus

Weighted graph derivative

theory only

constitution, master dossier

Future regularizer.

EQ-015

L=D-W

graph calculus

Graph Laplacian

theory only

constitution, master dossier

Future anatomy/graph constraint.

EQ-016

R_graph(F)=sum_(i,j in E) w_ij norm(F_i-F_j)^2 = Tr(F^T L F)

graph/temporal/spectral regularizers

Graph smoothness

theory only

constitution, master dossier

Not an implemented loss.

EQ-017

(H_HRF a)(t)=int_0^infty h(tau)a(t-tau)dtau

fMRI HRF observation

HRF convolution

experimental

constitution, master dossier

Implemented as simplified convolution.

EQ-018

Y_fMRI(p,t)=R_p H_HRF g_theta(F(.,t))+epsilon

fMRI HRF observation

fMRI operator

experimental

constitution, master dossier

Current operator is a simplified readout.

EQ-019

tilde U_t=sum_{ell=0}^L alpha_ell U_{t-ell}

observation operators

Stimulus lag adapter

experimental

master dossier

Implemented as causal averaging adapter.

EQ-020

Y_EEG(t)=L_s J_theta(F_t)+epsilon_EEG

EEG/MEG observation

EEG operator

experimental

constitution, master dossier

Current operator is simplified pooling/readout.

EQ-021

Y_MEG(t)=M_s J_theta(F_t)+epsilon_MEG

EEG/MEG observation

MEG operator

theory only

constitution, master dossier

No MEG support claim.

EQ-022

Y_{n,t} ~ Poisson(Delta t softplus(w_n^T F(x_n,t)))

spike/calcium observation

Spike observation

theory only

constitution, master dossier

Not implemented.

EQ-023

c_{n,t}=(k_Ca*r_n)(t)+epsilon

spike/calcium observation

Calcium observation

theory only

constitution, master dossier

Not implemented.

EQ-024

`p(a_t

F_t,U_t)=softmax(C pool(F_t)+D U_t)`

behavior observation

Behavior observation

scaffolded

constitution, master dossier

EQ-025

`q_phi(F_{0:T}

Y_obs,U,A_s)`

variational inference / ELBO

Variational posterior

theory only

ledger, master dossier

EQ-026

`L_ELBO=E_q[sum_{m,t} log p_theta(Y_{m,t}

F_t,A_s)] - beta KL(q_phi(F

Y,U,A_s)

p_theta(F

U,A_s))`

EQ-027

F'=AF

gauge symmetry / identifiability

Gauge ambiguity

theory only

constitution, master dossier

Explains latent non-identifiability.

EQ-028

O'_m=O_m A^{-1}

gauge symmetry / identifiability

Gauge ambiguity

theory only

constitution, master dossier

Must not overclaim latent semantics.

EQ-029

O'_m(F')=O_m(F)

gauge symmetry / identifiability

Gauge invariance

theory only

constitution, master dossier

Observation equivalence.

EQ-030

R_time=sum_t norm(F_{t+1}-F_t)^2

graph/temporal/spectral regularizers

Temporal smoothness

theory only

ledger, master dossier

Not implemented as loss.

EQ-031

R_band=sum_omega gamma_omega norm(Fhat(omega))^2

graph/temporal/spectral regularizers

Spectral regularizer

theory only

ledger, master dossier

Not implemented.

EQ-032

A_s=I+U_s V_s^T

operator learning

Low-rank subject adapter

scaffolded

ledger, master dossier

Subject adaptation exists elsewhere, not NFC proof.

EQ-033

dF/dt=f_theta(F,U,t)

stability / spectral radius

Continuous stability setup

theory only

ledger, master dossier

Future stability audit.

EQ-034

J_t=partial f_theta / partial F

stability / spectral radius

Jacobian

theory only

ledger, master dossier

Not computed.

EQ-035

Re(lambda_i(J_t))<0

stability / spectral radius

Local stability

theory only

ledger, master dossier

Not enforced.

EQ-036

rho(A)<1

stability / spectral radius

Discrete stability

theory only

ledger, master dossier

Not enforced.

EQ-037

R_stab=max(0,rho(A)-1)^2

stability / spectral radius

Stability penalty

theory only

ledger, master dossier

Future loss term.

EQ-038

L_NLL=sum_{m,t,i} (y-mu)^2/(2 sigma^2)+0.5 log sigma^2

uncertainty / calibration

Uncertainty NLL

theory only

constitution, master dossier

Current uncertainty calibration is not full NLL.

EQ-039

P[Y in C_alpha(X)] approx 1-alpha

uncertainty / calibration

Calibration condition

theory only

constitution, master dossier

Requires real calibration artifacts.

EQ-040

L_InfoNCE=-log exp(sim(z_a,z_b)/tau)/sum_{b'} exp(sim(z_a,z_b')/tau)

information theory / contrastive learning

Contrastive alignment

theory only

ledger, master dossier

Future representation audit.

EQ-041

g(F_{t+1})=K g(F_t)

Koopman / residual dynamics

Koopman form

theory only

ledger, master dossier

Future dynamics baseline.

EQ-042

F_{t+1}=K F_t+R_theta(F_t,U_t)

Koopman / residual dynamics

Koopman residual update

theory only

ledger, master dossier

Not implemented.

EQ-043

L=L_pred+lambda_R norm(R_theta)^2+lambda_K max(0,rho(K)-1)^2

Koopman / residual dynamics

Koopman residual loss

theory only

ledger, master dossier

Not implemented.

EQ-044

C=(1/T)XX^T

Riemannian EEG covariance

EEG covariance

theory only

ledger, master dossier

Future EEG baseline.

EQ-045

Chat=LL^T+epsilon I

Riemannian EEG covariance

SPD covariance head

theory only

ledger, master dossier

Not implemented.

EQ-046

d_AIRM(C1,C2)=norm(log(C1^{-1/2}C2C1^{-1/2}))_F

Riemannian EEG covariance

AIRM distance

theory only

ledger, master dossier

Not implemented.

EQ-047

T_m={k Delta_m:k in Z}

sampling lattice / Fourier / coprime windows

Sampling grid

theory only

ledger, master dossier

Future sampling stress test.

EQ-048

t in union_m T_m

sampling lattice / Fourier / coprime windows

Union of sampling grids

theory only

ledger, master dossier

Future multimodal timing model.

EQ-049

phi(t)=[sin(2 pi k t/T), cos(2 pi k t/T)]_{k=1}^K

sampling lattice / Fourier / coprime windows

Fourier features

theory only

ledger, master dossier

Not implemented.

EQ-050

L v_k=lambda_k v_k

sampling lattice / Fourier / coprime windows

Graph spectral modes

theory only

ledger, master dossier

Not implemented.

EQ-051

gcd(w,s)=1

sampling lattice / Fourier / coprime windows

Coprime window stress test

theory only

ledger, master dossier

Future split/window audit.

EQ-052

log(phi_1(t;theta)/phi_0) approx J delta mu_a(t;theta)

fNIRS optical/Rytov observation

Rytov approximation

theory only

fNIRS notes, master dossier

fNIRS is docs only.

EQ-053

phi_1(t;theta) approx phi_0 exp(J delta mu_a(t;theta))

fNIRS optical/Rytov observation

Predicted intensity

theory only

fNIRS notes, master dossier

No fNIRS implementation claim.

EQ-054

Delta OD(t;theta)=log phi_base_bar - log phi_1(t;theta)

fNIRS optical/Rytov observation

Optical density

theory only

fNIRS notes, master dossier

Docs only.

EQ-055

Y_fNIRS(t)=Delta OD(phi_0 exp(J delta mu_a(t)))+epsilon

fNIRS optical/Rytov observation

fNIRS observation

theory only

fNIRS notes, master dossier

Docs only.

EQ-056

F_hemo(t)=(h_HRF*g(F_neural))(t)

fNIRS optical/Rytov observation

Neural to hemodynamic field

theory only

fNIRS notes, master dossier

Informs future generator realism.

EQ-057

Y_fMRI(t)=R F_hemo(t)+epsilon

fMRI HRF observation

Synthetic fMRI

experimental

ledger, master dossier

Current synthetic generator approximates this.

EQ-058

Y_EEG(t)=L F_neural(t)+A_motion(t)+epsilon

EEG/MEG observation

Synthetic EEG

experimental

ledger, master dossier

Motion artifacts are future realism.

EQ-059

Q:R^d -> {0,1}^{bd}

TurboQuant/TurboVec math

TurboQuant map

deferred

TurboQuant notes, master dossier

Optional infrastructure only.

EQ-060

E norm(x-xhat)_2^2

TurboQuant/TurboVec math

MSE objective

deferred

TurboQuant notes, master dossier

Not implemented.

EQ-061

E[(<x,q>-<xhat,q>)^2]

TurboQuant/TurboVec math

Inner-product objective

deferred

TurboQuant notes, master dossier

Not implemented.

EQ-062

r=norm(v)_2, u=v/norm(v)_2

TurboQuant/TurboVec math

Vector normalization

deferred

TurboQuant notes, master dossier

Optional future adapter.

EQ-063

z=Ru

TurboQuant/TurboVec math

Random rotation

deferred

TurboQuant notes, master dossier

Optional future adapter.

EQ-064

norm(z)_2=norm(u)_2=1

TurboQuant/TurboVec math

Norm preservation

deferred

TurboQuant notes, master dossier

Optional future adapter.

EQ-065

z_i approx N(0,1/d)

TurboQuant/TurboVec math

Coordinate approximation

deferred

TurboQuant notes, master dossier

Risky at low dimension.

EQ-066

q_i=c_argmin_k abs(z_i-c_k)

TurboQuant/TurboVec math

Scalar quantization

deferred

TurboQuant notes, master dossier

Optional future adapter.

EQ-067

x=xhat_MSE+r_residual

TurboQuant/TurboVec math

Residual split

deferred

TurboQuant notes, master dossier

Optional future adapter.

EQ-068

xhat_prod=xhat_MSE+QJL(r_residual)

TurboQuant/TurboVec math

QJL residual correction

deferred

TurboQuant notes, master dossier

Optional future adapter.

EQ-069

score(q,x)=<q,x>

retrieval-kNN baseline

Inner-product score

deferred

TurboQuant notes, master dossier

Use exact numpy first.

EQ-070

I_stim={Q(e_stim,i)}_{i=1}^N

semantic duplicate audit

Stimulus index

deferred

TurboQuant notes, master dossier

Future semantic audit.

EQ-071

z_t=pool(F_t)

latent field memory

Latent summary

deferred

TurboQuant notes, master dossier

Future memory layer.

EQ-072

z_{t,k}=pool_{i in network k}(F_{t,i})

latent field memory

Network summary

deferred

TurboQuant notes, master dossier

Future audit feature.

EQ-073

N_k(q)={i_1,...,i_k}

retrieval-kNN baseline

Retrieval neighborhood

deferred

TurboQuant notes, master dossier

Optional baseline.

EQ-074

yhat_test=sum_{i in N_k(q)} w_i y_i

retrieval-kNN baseline

Retrieval-kNN prediction

deferred

TurboQuant notes, master dossier

Must not use test targets.

EQ-075

w_i=exp(tau <q,z_i>)/sum_{j in N_k(q)} exp(tau <q,z_j>)

retrieval-kNN baseline

Retrieval weights

deferred

TurboQuant notes, master dossier

Train-only labels only.

EQ-076

abs(<q,z_i>-<q,zhat_i>) <= norm(q)_2 norm(z_i-zhat_i)_2

TurboQuant/TurboVec math

Score error bound

deferred

TurboQuant notes, master dossier

Quantization risk.

EQ-077

d_min(x_test,D_train)=min_{x_i in D_train} norm(q(x_test)-q(x_i))

semantic duplicate audit

Nearest train distance

deferred

TurboQuant notes, master dossier

Future leakage audit.

EQ-078

L=sum_m lambda_m L_m+lambda_latent L_latent+lambda_graph Tr(F^T L F)+lambda_stab R_stab+lambda_unc L_cal

falsification gates

Full NFC loss

theory only

master dossier

Not implemented as one loss.

EQ-079

L_fMRI=norm(M(Yhat_fMRI-Y_fMRI))^2

fMRI HRF observation

fMRI reconstruction loss

experimental

master dossier

MSE implemented, mask semantics vary.

EQ-080

L_EEG=norm(M(Yhat_EEG-Y_EEG))^2+gamma norm(PSD(Yhat)-PSD(Y))^2

EEG/MEG observation

EEG loss

theory only

master dossier

PSD term not implemented.

EQ-081

L_spike=-sum_{n,t} log Poisson(y_{n,t};lambda_{n,t})

spike/calcium observation

Spike loss

theory only

master dossier

Not implemented.

EQ-082

L_{a->b}=norm(O_b(I_theta(y_a))-y_b)

falsification gates

Cross-modal consistency

theory only

master dossier

Field-mediated route under test.

EQ-083

T_{a->b}=O_b circ I_a

falsification gates

Commutative translation factorization

theory only

master dossier

Research framing.

EQ-084

y_a -> y_b

falsification gates

Direct translation baseline

implemented baseline

master dossier

Baseline lane.

EQ-085

y_a -> F -> y_b

falsification gates

Field-mediated translation

experimental

master dossier

NFC lane.

Status Rules#

  • Pair-Operator is archived as the main primitive and retained as an ablation.

  • NFC is the experimental architecture path.

  • fNIRS is theory/evidence only and not implemented support.

  • TurboVec/TurboQuant is optional retrieval/audit infrastructure, not the model contribution.

  • MOABB Track A is a reproducibility evidence path.

  • Algonauts/CNeuroMod fMRI is the future Track B model battlefield.

  • MDD or clinical claims are not allowed.

  • A100 experiment rules require strict 1x synthetic diagnostic before Algonauts or 6x DDP.