---

name: score_sde_schedule description: Define continuous-time score-SDE schedules, perturbation kernels, reverse drifts, and probability-flow checks for recovery experiments.

Score SDE Schedule

Use this skill when a recovery or implementation task needs the paper's continuous-time forward perturbation and reverse-dynamics contracts without depending on the original repository. It is appropriate for VE-style reduced experiments, formula cross-checks, and small deterministic tests. Do not use it as evidence that full image-model training ran.

Inputs

• SDE family, currently ve for deterministic scripts.
• Scalar or list time values in [0, 1].
• Schedule parameters sigma_min and sigma_max.
• State and score values when computing reverse or probability-flow drift.

Outputs

• VE marginal standard deviation and perturbation records.
• Diffusion coefficient g(t).
• Reverse-SDE score drift and probability-flow score drift.
• Validation booleans for factor-of-two consistency.

Workflow

1. Validate positive schedule parameters and bounded times.
2. Compute log-linear VE sigma(t) and diffusion coefficient.
3. Perturb samples using caller-supplied deterministic noise when reproducibility matters.
4. Compute reverse drift as -g(t)^2 score and probability-flow drift as -0.5*g(t)^2 score.
5. Record formula outputs in recovery artifacts when using this skill for a proxy.

Validation

Run python scripts/score_sde_schedule.py --self-test or validate the tree with validate_skill_tree.py --run-tests.

Limitations

The deterministic script intentionally implements the VE formulas needed for bounded recovery. VP and sub-VP can be added by following the same contract, but this skill should not claim to reproduce full paper metrics by itself.