ARLIT — Scale-Invariant Effective Information Lab (Claim Pack, Fixed)

Dataset Mode Dimension d Mass m IR floor k_min λ (phi⁴) Lattice N (2D) Preset

Scale Window (Regime) Λ_min Λ_max Samples N Train fraction Safe Mode (limits)

Randomness Seed

Deterministic results for noise & sim when using a fixed seed.

Bootstrap CIs Enable bootstrap B (resamples) CI levels

Resamples train/test Λ with replacement; reports CIs for s and ARLIT# (test).

Mass Generalization Fit mass m_fit Eval masses m′ (comma)

Controls

Cursor & Overlays Show plot cursor Cursor speed The animated cursor traverses Λ and highlights I and C at that scale.

Theoretical Note (sufficient conditions)

If log I(Λ) is locally affine in log Λ (bounded curvature), a power-law Z(Λ)=Λ^s fitted on train Λ minimizes the mean square slope of log C(Λ) on test Λ. For the free Gaussian QFI (θ=ln m), the source is exact: F_Q(Λ) = 0.5 m^4 * I_free_scalar(d,m,Λ). Low curvature ⇒ low ARLIT#.

Results + CIs

Validation Lab

Validation Outputs

Unit Tests

Λ: — I: — C: —

I(Λ)

C(Λ) = Z·I

Train/Test Split