Record anonymous Hub readback and TRL recipe verification

verification.json

@@ -0,0 +1,105 @@
+{
+  "passed": true,
+  "dataset_revision_verified": "a57fbba49e892199d2b0c31f68034310bf2e281c",
+  "anonymous_load_all_configurations": [
+    {
+      "configuration": "all",
+      "rows": 17929,
+      "anonymous_load_dataset": true,
+      "decoded_types_or_masks_valid": true
+    },
+    {
+      "configuration": "opencode",
+      "rows": 4825,
+      "anonymous_load_dataset": true,
+      "decoded_types_or_masks_valid": true
+    },
+    {
+      "configuration": "claude-code",
+      "rows": 4581,
+      "anonymous_load_dataset": true,
+      "decoded_types_or_masks_valid": true
+    },
+    {
+      "configuration": "codex",
+      "rows": 4078,
+      "anonymous_load_dataset": true,
+      "decoded_types_or_masks_valid": true
+    },
+    {
+      "configuration": "mini-swe-agent",
+      "rows": 4445,
+      "anonymous_load_dataset": true,
+      "decoded_types_or_masks_valid": true
+    },
+    {
+      "configuration": "lfm25_2_6b",
+      "rows": 17929,
+      "anonymous_load_dataset": true,
+      "decoded_types_or_masks_valid": true
+    },
+    {
+      "configuration": "qwen35_2b",
+      "rows": 17929,
+      "anonymous_load_dataset": true,
+      "decoded_types_or_masks_valid": true
+    }
+  ],
+  "all_remote_asset_hashes_match": true,
+  "export_roundtrip": {
+    "passed": true,
+    "counts": {
+      "conversations/claude-code": 4581,
+      "lfm25_2_6b/claude-code": 4581,
+      "conversations/codex": 4078,
+      "lfm25_2_6b/codex": 4078,
+      "conversations/mini-swe-agent": 4445,
+      "lfm25_2_6b/mini-swe-agent": 4445,
+      "conversations/opencode": 4825,
+      "lfm25_2_6b/opencode": 4825,
+      "qwen35_2b/claude-code": 4581,
+      "qwen35_2b/codex": 4078,
+      "qwen35_2b/mini-swe-agent": 4445,
+      "qwen35_2b/opencode": 4825
+    },
+    "files": 213
+  },
+  "cpu_trl_optimizer_smokes": {
+    "passed": true,
+    "checks": [
+      {
+        "configuration": "lfm25_2_6b",
+        "passed": true,
+        "mask_preserved": true,
+        "finite_loss": true,
+        "weights_changed": true,
+        "optimizer_steps": 1,
+        "test_scope": "CPU tiny GPT2 model with actual exported student IDs and labels, last 128 tokens. Not a student quality evaluation or full model GPU test."
+      },
+      {
+        "configuration": "qwen35_2b",
+        "passed": true,
+        "mask_preserved": true,
+        "finite_loss": true,
+        "weights_changed": true,
+        "optimizer_steps": 1,
+        "test_scope": "CPU tiny GPT2 model with actual exported student IDs and labels, last 128 tokens. Not a student quality evaluation or full model GPU test."
+      }
+    ]
+  },
+  "published_recipe_checks": [
+    {
+      "model": "lfm",
+      "portable_and_fast_paths_match": true,
+      "examples_checked": 8,
+      "recipe_downloaded_from_hub": true
+    },
+    {
+      "model": "qwen",
+      "portable_and_fast_paths_match": true,
+      "examples_checked": 8,
+      "recipe_downloaded_from_hub": true
+    }
+  ],
+  "note": "CPU optimizer smokes use tiny models and short slices of actual exported IDs. Full student save/resume and evaluation smokes were verified separately before production training."
+}