Rename to FineEnvs/openai-math; update download command, registry name and NOTICE

NOTICE

@@ -1,5 +1,5 @@
-openai-math-harbor
-==================
+openai-math
+===========
 
 This dataset packages the Lean statements of OpenAI's "openai/math" release as Harbor tasks.
 

README.md

@@ -66,7 +66,7 @@ as one end of that design space.
 
 - `tasks/` — one Harbor task per validated problem: instruction, Lean statement, sandbox image recipe,
   verifier and oracle
-- `registry.json` — Harbor registry entry (`openai-math-harbor@1.0`)
+- `registry.json` — Harbor registry entry (`openai-math@1.0`)
 - `manifest.parquet` — one row per task: field, family, theorem names, reference-proof size, oracle
   result
 - `generator/` — the code that builds the sandbox image and regenerates the tasks from openai/math
@@ -79,8 +79,8 @@ credentials for the agent you run.
 
 ```bash
 pip install "harbor[daytona]>=0.24" daytona
-hf download FineEnvs/openai-math-harbor --repo-type dataset --local-dir openai-math-harbor
-cd openai-math-harbor
+hf download FineEnvs/openai-math --repo-type dataset --local-dir openai-math
+cd openai-math
 
 # Build the shared sandbox image once (Lean 4, Mathlib, Comparator; ~9 GB, about an hour).
 (cd generator && python -m openai_math_harbor.snapshot)

generator/openai_math_harbor/adapter.py

@@ -13,7 +13,7 @@ come from the challenge's own files, so no task is written by hand. What varies
 
 Output layout (the FineEnvs Harbor dataset convention):
 
-    <out>/registry.json            Harbor registry: openai-math-harbor@<version> -> tasks/*
+    <out>/registry.json            Harbor registry: openai-math@<version> -> tasks/*
     <out>/README.md                dataset card
     <out>/NOTICE, LICENSE-openai-math   attribution for the redistributed statements
     <out>/manifest.json            per-task metadata and the excluded challenges
@@ -38,7 +38,7 @@ PKG = Path(__file__).parent
 IMAGE_DIR = PKG.parent / "image"
 TEMPLATES = PKG / "templates"
 SOURCE_COMMIT = "adc7f1241b42e322a6451854ab7e4b4c146bf78a"
-DATASET_NAME = "openai-math-harbor"
+DATASET_NAME = "openai-math"
 DATASET_VERSION = "1.0"
 IMAGE_FILES = ["Dockerfile", "lean_setup.sh", "fetch_mathlib.sh", "lakefile.lean", "lake-manifest.json", "pack_submission.py"]
 

generator/openai_math_harbor/attacks.py

@@ -4,7 +4,7 @@ Each probe keeps the task's environment, tests and config unchanged and replaces
 with a script that writes an invalid submission (or tampers with the agent sandbox). Run them with
 Harbor's oracle agent; every probe must score 0.
 
-    python -m openai_math_harbor.attacks --dataset datasets/openai-math-harbor/tasks --out datasets/attacks
+    python -m openai_math_harbor.attacks --dataset datasets/openai-math/tasks --out datasets/attacks
 """
 
 from __future__ import annotations

generator/openai_math_harbor/report.py

@@ -37,7 +37,12 @@ def classify(trial: Path) -> dict:
     if reward == 1:
         cause = "accepted"
     elif reward is None:
-        cause = "error: " + str((result.get("exception_info") or {}).get("exception_type") or g.get("reason", "unknown"))
+        exc = result.get("exception_info") or {}
+        if "no space left on device" in (exc.get("exception_traceback") or "") + (exc.get("exception_message") or ""):
+            # The verifier's disk filled so completely that Harbor could not even fetch the result.
+            cause = "disk"
+        else:
+            cause = "error: " + str(exc.get("exception_type") or g.get("reason", "unknown"))
     else:
         cause = next((name for name, keys in CAUSES if any(k in text for k in keys)), "rejected")
     return {

generator/openai_math_harbor/templates/NOTICE

@@ -1,5 +1,5 @@
-openai-math-harbor
-==================
+openai-math
+===========
 
 This dataset packages the Lean statements of OpenAI's "openai/math" release as Harbor tasks.
 

generator/openai_math_harbor/templates/dataset_README.md

@@ -66,7 +66,7 @@ as one end of that design space.
 
 - `tasks/` — one Harbor task per validated problem: instruction, Lean statement, sandbox image recipe,
   verifier and oracle
-- `registry.json` — Harbor registry entry (`openai-math-harbor@1.0`)
+- `registry.json` — Harbor registry entry (`openai-math@1.0`)
 - `manifest.parquet` — one row per task: field, family, theorem names, reference-proof size, oracle
   result
 - `generator/` — the code that builds the sandbox image and regenerates the tasks from openai/math
@@ -79,8 +79,8 @@ credentials for the agent you run.
 
 ```bash
 pip install "harbor[daytona]>=0.24" daytona
-hf download FineEnvs/openai-math-harbor --repo-type dataset --local-dir openai-math-harbor
-cd openai-math-harbor
+hf download FineEnvs/openai-math --repo-type dataset --local-dir openai-math
+cd openai-math
 
 # Build the shared sandbox image once (Lean 4, Mathlib, Comparator; ~9 GB, about an hour).
 (cd generator && python -m openai_math_harbor.snapshot)

registry.json

@@ -1,6 +1,6 @@
 [
   {
-    "name": "openai-math-harbor",
+    "name": "openai-math",
     "version": "1.0",
     "description": "355 Lean 4 theorem-proving tasks from openai/math, graded by Comparator in a separate sandbox.",
     "tasks": [