FT-V13C-KILL

Adding latent-diffusion and entropy-bonus regularizers to v1.3 improves convergence over v1.2 on the same corpus.

Char-level 1.36M-param model, Tiny Shakespeare, v1.3b regularizer stack plus new terms.

• ◆Regularizer stack composition

• →Run killed at 3,500 steps; v1.2's clean convergence (loss 0.0109 over 5 epochs) was not reproduced under the new stack.

• ·Filed as a failure in the progress report; the response was the v1.4 redesign (BPE 4096, 9M params) rather than more tuning.

Hypothesis not supported at this scale. Failure redirected the roadmap to v1.4.

v1.4 held-out generalization evals.