The Geometry of Failure: Terminal Measurement, Concentration Barriers, and the Limits of Linear Intervention in Neural Networks

Executive Summary

This consolidated paper unifies six previously separate works tracing the failure of direction-space intervention in neural networks. The arc begins with a practical tool that works (shaped noise breaks 100% of repetition loops) and ends with a mathematical theorem explaining why it cannot be made selective (the concentration barrier bounds selectivity at k/d_eff). The collected finding is a dissociation: classification and intervention operate under different constraints.

Shaped Noise Injection (former AI-8)

Shaped noise along INLP domain directions breaks 100% of repetition loops at 3B and 7B with near-perfect token uniqueness (0.99+). Domain-specific entropy reductions reach 6.1% for legal at 7B. But cross-domain selectivity fails: targeting medical reduces legal entropy by 10.7% while reducing medical by only 1.9%. All correction attempts fail -- scalar cancellation, subspace decomposition, and optimal linear correction via matrix inversion. The response matrix is invertible (determinant -84.5, condition 21.2), the linear algebra is clean, and the optimal weights are computable. They simply do not produce the predicted effects. Root cause: the terminal measurement problem -- the response matrix characterizes the system's input-output mapping but cannot invert nonlinear transformations at intermediate layers.

Layer-Resolved Response Tensor (former AI-9)

Maps the selectivity profile at nine sampled transformer layers in Qwen-2.5 7B (28 layers total). Mean selectivity peaks modestly at intermediate layers 7-10 (mean ~0.5) and declines toward both input and output. Strong domain asymmetry: code and science exhibit positive selectivity; medical and legal exhibit negative selectivity across all layers. No layer exceeds the concentration barrier bound of k/d_eff ~ 1.8.

Spectral Geometry (former AI-10)

Measures amplification spectrum of INLP domain directions through the layer Jacobian using finite-difference Jacobian-vector products. INLP directions are not preferentially amplified (INLP/random ratio 0.99 +/- 0.05). PCA-INLP alignment is near-random at intermediate layers but increases toward terminal layers where selectivity is lowest. The forward pass is an isotropic amplifier treating domain directions as generic directions in activation space.

The Concentration Barrier (former AI-11)

Proves that maximum achievable selectivity from k domain directions is bounded by k/d_eff. Measures effective dimensionality at every transformer layer: d_eff ranges 4.7-26.0 (mean 19.2) for last-token extraction. Mean-pooled d_eff collapses to 1.0 at layers 3-25, revealing previously undocumented representation anisotropy invisible to position-aware measurements. INLP variance fraction increases from 1.3% at early layers to 12.5% at terminal, yet the concentration barrier holds at all 28 layers.

SR Channel Capacity (former AI-12)

Quantifies information-theoretic content of domain-selective noise injection via KL divergence across a sigma sweep at optimal injection layer (layer 10). All four domains exhibit inverted-U KL profiles confirming stochastic resonance in information space. Total KL peaks at 14-15 bits but domain-specific component is small: +1.3 bits (medical), +1.9 bits (legal), -1.0 bits (code), -1.5 bits (science). Consistent with theoretical bound C <= log2(1 + k^2/d_eff) = 2.24 bits.

The Activation Geometry Program (former AI-13)

Program synthesis tracing the dependency graph across twelve papers. The program began with reducing training compute and arrived at the terminal measurement limit. The collected finding: classification and intervention operate under different constraints. High-dimensional activation spaces guarantee that fixed linear directions capture only a small fraction of the computation they describe. The geometry INLP discovers is real, but causal pathways are distributed across dimensions no fixed subspace can isolate. This is not an engineering limitation -- it is a property of high-dimensional nonlinear computation.

Key Findings

• Repetition loop breaking: 100% escape rate at 3B and 7B with near-perfect token uniqueness (0.99+)
• Cross-domain selectivity fails: Targeting one domain affects non-target domains comparably or more strongly; all corrections fail
• Terminal measurement limit: Response matrix characterizes but cannot invert nonlinear mixing at intermediate layers
• Selectivity peaks at intermediate layers: Layers 7-10 show highest selectivity (~0.5), declining toward input and output
• Isotropic amplification: Forward pass treats INLP directions identically to random directions (ratio 0.99)
• Concentration barrier theorem: Maximum selectivity bounded by k/d_eff -- proved analytically, verified at all 28 layers
• Information bound: Domain-specific SR channel capacity ~2 bits against ~15 bits total, consistent with geometric bound
• Classification-intervention dissociation: Classification and intervention operate under fundamentally different constraints

Superseded Papers

This paper consolidates and supersedes:

• AI-8: Shaped Noise Injection -- domain precision, loop breaking, and the terminal measurement limit
• AI-9: Layer-Resolved Response Tensor -- where domain selectivity lives in the forward pass
• AI-10: Spectral Geometry of the Forward Pass -- how INLP directions interact with layer Jacobians
• AI-11: The Concentration Barrier -- effective dimensionality bounds domain selectivity
• AI-12: Channel Capacity of Domain-Specific Stochastic Resonance
• AI-13: The Activation Geometry Program -- twelve papers on the mathematical structure of neural network representations

Key References