rrm workspace: TRM/HRM/SRM code, Maze dataset, dynamical-analysis pipelineHEADmain

Curated export for clone-and-run Maze training (2x A6000) + diagnostics.
trm/hrm pretrain.py carry trajectory-augmentation code (backward-compatible).
Heavy artifacts (checkpoints/wandb/npz) gitignored; see PROVENANCE.md.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

1 files changed, 27 insertions, 0 deletions

diff --git a/research/flossing/late_perturb_robustness/smoke_baseline.meta.json b/research/flossing/late_perturb_robustness/smoke_baseline.meta.json
new file mode 100644
index 0000000..3ceb2d3
--- /dev/null
+++ b/research/flossing/late_perturb_robustness/smoke_baseline.meta.json
@@ -0,0 +1,27 @@
+{
+  "args": {
+    "batch_size": 8,
+    "ckpt_name": "step_58590",
+    "ckpt_root": "/home/yurenh2/rrm/trm/checkpoints/Sudoku-extreme-1k-aug-1000-ACT-torch/pretrain_mlp_t_sudoku_official_gbs768_repro",
+    "label": "smoke_baseline",
+    "n_samples": 16,
+    "noise_distribution": "gaussian",
+    "out_prefix": "research/flossing/late_perturb_robustness/smoke_baseline",
+    "perturb": "both",
+    "perturb_afters": "0,8",
+    "rollouts": 2,
+    "seed": 20260606,
+    "sigmas": "0,0.01"
+  },
+  "config_global_batch_size": 768,
+  "data_path": "/home/yurenh2/rrm/data/sudoku-extreme-1k-aug-1000",
+  "n_samples": 16,
+  "perturb_afters": [
+    0,
+    8
+  ],
+  "sigmas": [
+    0.0,
+    0.01
+  ]
+}
\ No newline at end of file