# Claim 2 — optimisation versus stochastic error

## Legacy judge claim

> GLU primary benefit lies in accelerating optimization rather than reducing stochastic error.

## Release status

**Unverified.** The retained release evidence measures only an initial empirical-NTK condition number. It contains no loss trajectory, convergence-rate comparison, stochastic-error estimate, generalisation measurement, dataset experiment, or benchmark result.

## Boundary

The Claim 1 condition-number output must not be used to infer this optimisation claim. A separately authorised and appropriately scoped experiment would be required before changing this status.

The active judge input is anchored and overrides this legacy claim set; see [`judge-evidence.md`](judge-evidence.md). This release claim remains **Unverified.**