Claude Skills by VectorSpaceLab

Run a bounded sequential APT proxy loop that updates proposals after corrected atomic training.

Run an auditable reduced recovery experiment for the SBI benchmark by composing generated task, sampling, and C2ST skills.

Compute a deterministic classifier two-sample accuracy proxy for comparing posterior sample distributions.

Generate deterministic reference and approximate posterior samples for bounded simulation-based inference recovery experiments.

Build validated simulation-based inference benchmark task items with prior, simulator, observation, and reference posterior metadata.

Run bounded annealed Langevin posterior sampling from composed score functions for score-based SBI recovery.

Build and validate reduced denoising score-training traces for single-observation simulation-based inference posteriors.

Compose F-NPSE and PF-NPSE posterior scores from prior corrections and per-condition score estimates.

Compute squared MMD for posterior samples and package score-based SBI recovery-result metadata.

Build likelihood-free linear-nonlinear neural simulations and STA/firing-rate summaries for SNPE-style posterior recovery.

Fit a conditional Gaussian posterior surrogate from simulated summary/parameter pairs without likelihood evaluations.

Validate SNPE-style posterior recovery using parameter correlation, posterior predictive summary error, and mechanism checks.

Coordinate prior and posterior-guided simulation rounds for a bounded SNPE-style recovery experiment.

Select SNPE, SNLE, or SNRE for a simulation-based inference task from estimator and posterior-access requirements.

Summarize posterior samples and validate SBI uncertainty with finite-sample and reference checks.

Run a bounded source-boundary-safe recovery experiment for the sbi toolkit workflow.

Prepare simulator, prior, observation, and simulated pair contracts for simulation-based inference workflows.

Run a bounded SNPE-style conditional posterior training loop for mechanism-faithful SBI recovery.

Define VE, VP, and sub-VP SDE contracts plus reverse-time SDE and probability-flow ODE dynamics for score-based generative modeling.

Construct continuous-time denoising score matching losses and tiny optimizer-step checks for Score SDE recovery.

Run bounded predictor-corrector reverse sampling diagnostics for score-based SDE mechanisms.

Build deterministic probability-flow ODE diagnostics and one-dimensional likelihood corrections for Score SDE recovery.

Execute a bounded soft-mode Score SDE recovery experiment using generated skills and mechanism checks.

Compute TSNPE high-posterior-density truncation thresholds and rejection samples from a bounded prior.

Fit a reduced Gaussian posterior surrogate with pooled maximum-likelihood updates.

Run a bounded mechanism-faithful TSNPE proxy recovery experiment.

Compute TSNPE support inclusion and simulation-based coverage calibration diagnostics after each truncated proposal training round.

Run deterministic sampling-importance-resampling diagnostics for narrow TSNPE proposals.

Estimate validity probabilities and correct likelihood-method posterior scores for invalid simulator outputs.

Run a bounded sequential simulation-based inference loop that updates proposals from variational posterior estimates.

Apply sampling importance resampling to refine variational posterior samples using likelihood-prior weights.

Compute stable mass-covering variational objective diagnostics for SNVI-style posterior optimization.

Prepare classifier-free diffusion conditioning-dropout batches and tiny denoising training traces for reduced recovery.

Run a tiny reverse-diffusion-like sampler that exercises classifier-free guided score calls at each step.

Evaluate classifier-free guidance sweeps with proxy confidence and diversity trade-off metrics.

Compute classifier-free guided diffusion score estimates from conditional and unconditional predictions.

Select a coherence boosting alpha from cached validation likelihoods without model finetuning.

Apply the coherence boosting log-linear scoring rule to full-context and premise-free answer likelihoods.

Build and validate full-context and premise-free-context records for coherence boosting answer-selection experiments.

Evaluate coherence boosting recovery metrics and source-boundary mechanism checks for proxy experiments.

Select tokens from regular or CAD-adjusted logits using comparable greedy or top-p decoding policies.

Apply the Context-aware Decoding contrastive logit formula and report mechanism diagnostics for recovery experiments.

Build and validate context-aware decoding prompt parts with separated evidence context, query, and generation prefix.

Evaluate CAD conflict-recovery runs with exact-match metrics and mechanism-check summaries.

Compose multiple FUDGE future constraints, especially topic-word constraints, by summing weighted log probabilities.

Apply the FUDGE decoding rule by combining base candidate logits with future-discriminator probabilities and renormalizing.

Build prefix-level future-discriminator examples for FUDGE-style controlled generation from completed token sequences and sequence/target-word labels.

Evaluate reduced FUDGE topic-control recovery with topic-token metrics and explicit mechanism checks.

Reweight base language-model token probabilities with GeDi posteriors and apply GeDi cumulative-mass filtering for controlled decoding.

Compute and test GeDi's hybrid discriminative-generative objective with a deterministic tiny optimizer step.