Upload results/spec_rl_eval.json with huggingface_hub

results/spec_rl_eval.json

@@ -0,0 +1,14 @@
+{
+  "note": "Executed RL-style eval of the spec_rl verifiers v1 environment against the free Laguna pinference endpoint (poolside/laguna-xs.2, reasoning_effort=none, greedy). Real measured rewards — replaces the earlier eval_local stub. Under greedy decoding DFlash output is byte-identical, so the DFlash endpoint yields the SAME reward (cheaper rollouts, zero quality cost).",
+  "harness": "verifiers v1 taskset+harness (prime eval run)",
+  "model": "poolside/laguna-xs.2",
+  "endpoint": "prime/pinference (free)",
+  "n_examples": 12,
+  "rollouts_per_example": 1,
+  "temperature": 0.0,
+  "reasoning_effort": "none",
+  "mean_reward": 0.850,
+  "std_reward": 0.338,
+  "per_rollout_reward": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.2, 1.0],
+  "reward_kind": "dense fractional unit-test pass rate in [0,1]"
+}