1SCBE
Context-bound cost physics
Encode context as geometry, measure how far a state has drifted, and make that drift cost explode.
1.1 · Encode context into a geometric space
Map inputs (prompt, tool state, metadata, time) into a structured vector \(x\), and maintain an allowed region / reference structure \(\mathcal{M}\).
1.2 · Measure deviation
\[ \Delta = d(x, \mathcal{M}) \qquad\text{or, centroid baseline:}\qquad \Delta = \lVert x-\mu \rVert_2 \]
1.3 · Amplify cost super-exponentially — the “wall”
\[ H(d, R) = R^{\,d^2} \]
For fixed \(R>1\), cost explodes as the deviation dimension \(d\) grows.
Note: this is the unbounded cost-multiplier form (root package). The shipped pipeline gate consumes a separate bounded score \(1/(1+d_H+2\,pd)\in(0,1]\); the two live in different packages and should not be substituted.
1.4 · Asymmetry objective
Design goal: \(C_A(\Delta, d)\) grows much faster than \(C_D(\Delta, d)\) — defenders pay roughly constant overhead while attackers pay exploding overhead.