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, 2 insertions, 0 deletions

diff --git a/research/flossing/ptrm_selection_trm_base260410_k8_n512.summary.csv b/research/flossing/ptrm_selection_trm_base260410_k8_n512.summary.csv
new file mode 100644
index 0000000..12a5a4b
--- /dev/null
+++ b/research/flossing/ptrm_selection_trm_base260410_k8_n512.summary.csv
@@ -0,0 +1,2 @@
+deterministic/exact,deterministic/token_acc,fd_lyap,include_clean,lambda_fail_mean,lambda_mean,lambda_success_mean,lyap_min/exact,lyap_min/token_acc,mean_rollout/exact,mean_rollout/token_acc,n_samples,noise_std,oracle_pass/exact,oracle_pass/token_acc,q_fail_mean,q_max/exact,q_max/token_acc,q_mean,q_minus_0.25lambda/exact,q_minus_0.25lambda/token_acc,q_minus_0.5lambda/exact,q_minus_0.5lambda/token_acc,q_minus_1lambda/exact,q_minus_1lambda/token_acc,q_minus_2lambda/exact,q_minus_2lambda/token_acc,q_success_mean,rollout0/exact,rollout0/token_acc,rollouts

+0.79296875,0.9237557649612427,1.0,1.0,7.364749908447266,6.310528755187988,6.000424385070801,0.8828125,0.955053985118866,0.772705078125,0.9167057871818542,512.0,0.001,0.888671875,0.9605998992919922,-10.958671569824219,0.888671875,0.9566453695297241,3.255891799926758,0.888671875,0.9565731287002563,0.888671875,0.9565731287002563,0.888671875,0.9566213488578796,0.884765625,0.9557292461395264,7.437174320220947,0.76953125,0.9146653413772583,8.0