Invalidate WebShop SPrPO training seeds 1 and 2 affected by request-seed bug

Use explicit download and catalog URLs in model card

Standardize retained training artifacts at step200 and organize ablations

Add reproducible training overhead table for main experiments

Reduce edge margins while preserving the original motivation canvas size

Center clean-rate glyphs and tighten motivation whitespace

Align motivation condition names exactly with bar centers

Equalize motivation panel widths and reduce condition typography

Fit relative-drop labels between bars and clean reference lines

Place ALFWorld perturbation labels to the right of bars

Move motivation condition labels closer to the plots

Tighten motivation layout and remove explanatory footer

Restyle introduction figure with relative-drop labels and method-inspired pastel colors

Redesign half-width motivation figure with matched Gaussian 0.20 and dropout 0.05

Update ALFWorld SAM three-seed main results

Add ALFWorld 1.5B single-layer FFN Channel-SAM step100/step200 (rho=0.005, cap=0.05) (#6)

Add qwen2_5_1_5_stable_gaussian_noise_v11_step_200_seed0 step200 LoRA adapter and run config (#5)

Add qwen2_5_1_5_stable_gaussian_noise_v9_step_200_seed0 step200 LoRA adapter and run config (#4)

Tighten main-result panel spacing in LaTeX and preview

Publish step-200 main-result figures, plotting data, and LaTeX layout

Rename shared analysis group to perturbation_effect_across_stage

Publish WebShop perturbation-geometry analysis figure and data

Normalize step100 LoRA adapter configs and add step100 to manifests

Add results/alfworld/robust_training/qwen2_5_7_gaussian_noise_step_200_seed0/checkpoints/global_step_100/actor/lora_adapter

Add results/alfworld/training/qwen2_5_7_step_200_seed0/checkpoints/global_step_100/actor/lora_adapter

Add results/webshop/training/qwen2_5_7_step_200_seed0/checkpoints/global_step_100/actor

Add WebShop 7B single-layer FFN Channel-SAM step200 (rho=0.005, cap=0.05) (#2)

Publish ALFWorld Qwen2.5-7B step200 clean-vs-gaussian evaluation results

Publish qwen2_5_7_stable_gaussian_noise_step_200_seed0 step200 LoRA adapter for collaborator ALFWorld 7B handoff

Publish qwen2_5_7_stable_gaussian_noise_step_200_seed0 step200 LoRA adapter for collaborator ALFWorld 7B handoff

Publish qwen2_5_7_stable_gaussian_noise_step_200_seed0 step200 LoRA adapter for collaborator ALFWorld 7B handoff

Publish qwen2_5_7_stable_gaussian_noise_step_200_seed0 step200 LoRA adapter for collaborator ALFWorld 7B handoff

Remove redundant motivation README

Remove redundant reward surface README

Organize reward surface and motivation figures with reproducible CSV inputs

Publish complete WebShop step200 three-seed evaluation results

Remove superseded SAM-named evaluation figure

Publish complete WebShop step200 three-seed evaluation results

Use fresh cap 0.05 SAM as official step200 baseline

Add WebShop single-layer FFN Channel-SAM step200 (rho=0.005, cap=0.05) (#1)

Remove redundant evaluation package readme and manifest

Update curated evaluation figures and reproducible data

Use Vanilla, Plain Gaussian and Stable Gaussian algorithm labels

Publish WebShop 1.5B Stable Gaussian step200 main result

Simplify model card around step200 baseline and model download table

Document unified 200-step training default for 1.5B and 7B

Publish four completed Qwen2.5-7B step-200 baseline adapters

Document four finalized step-300 evaluation baselines

Publish evaluation-only step-300 LoRA baselines; remove training archive

Archive baseline: alfworld/training/qwen2_5_1_5_step_300_seed0