5 days ago
-
Card figures: regenerate from make_eval_figures.py with one shared y-label
-
Card figures: one shared y-label
-
README: arXiv link and citation
16 days ago
-
Update card figures to the paper's current versions
-
Correct evaluation decoding metadata
-
Update dataset card title
17 days ago
-
Update dataset card
-
Add Monte Carlo continuations (branches.jsonl.zst, Qwen3-4B and Qwen3-8B)
-
Add hidden-state feature shards, Qwen3-8B and the gold collection
-
Add hidden-state feature shards, Qwen3-4B
-
Add text-only traces, Qwen3-8B
-
Add text-only traces, Qwen3-4B
-
Card: mark pending uploads
-
Update dataset card; add DeepMath difficulty figure
-
Add text-only traces, Qwen3-4B (part 3)
-
Add text-only traces, Qwen3-4B (part 2)
-
Add text-only traces, Qwen3-4B
-
Add dataset card and figures
-
Add the refinement-depth table
-
Add reasoning-dynamics text features
-
Add problem-split per-seed calibration scores
-
Add rollout corpus, Qwen3-8B (reward-model fields removed)
-
Add rollout corpus, Qwen3-4B (reward-model fields removed) (part 2)
-
Add rollout corpus, Qwen3-4B (reward-model fields removed)
-
Add gold-collection labels keyed by trace MD5
-
Add layer-selection curves
-
Add probe checkpoints
-
Add label-swap scores and the rollout-search table
-
Remove anchors.pt (no shipped reader)
-
Move replay files to the repository root (remove old paths)
-
Move replay files to the repository root (copy)
-
Copy evaluation files from the evaluation dataset
-
Copy trial files from the evaluation dataset
-
Add replay files
-
initial commit
