diff --git a/README.md b/README.md
index b9337c296e5a70d9a1970abfd3735ab1db6d13d6..028cba976b7ac1295970f9db5680d95bf9b5222d 100644
--- a/README.md
+++ b/README.md
@@ -22,8 +22,8 @@ pretty_name: "openai/math as Harbor environments"
 We ported the math problems from OpenAI's [openai/math](https://github.com/openai/math) release
 (commit `adc7f12`) into [Harbor](https://github.com/harbor-framework/harbor) environments.
 The release formalizes **405** of its results as Lean statements. This dataset contains
-**355 environments**, the ones whose official proof we verified end to end in the environment,
-covering 209 result families across 17 fields. The other 50 are listed
+**369 environments**, the ones whose official proof we verified end to end in the environment,
+covering 214 result families across 17 fields. The other 36 are listed
 under [Validation](#validation) with the reason each could not be validated.
 
 ## How a task works
@@ -107,66 +107,51 @@ official proof was accepted are included.
 
 | Oracle result | Tasks | Meaning |
 |---|---|---|
-| accepted | 355 | OpenAI's proof is accepted: the environment is verified end to end |
-| missing-package | 16 | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
+| accepted | 369 | OpenAI's proof is accepted: the environment is verified end to end |
+| missing-package | 21 | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | definition-mismatch | 2 | OpenAI's proof writes a definition with the same text but it compiles differently, so the exact-match check fails |
-| memory | 5 | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
-| disk | 4 | OpenAI's proof is too large for the sandbox's 10 GB disk |
-| not completed | 22 | grading did not complete in our run |
+| memory | 10 | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
+| not completed | 2 | grading did not complete in our run |
 
-Grading an accepted proof took 16 minutes at the median. Each task's result is in its `task.toml` (`oracle_result`) and in `manifest.parquet`.
+Grading an accepted proof took 8 minutes at the median. Each task's result is in its `task.toml` (`oracle_result`) and in `manifest.parquet`.
 
-<details><summary>50 statements not included (click to expand)</summary>
+<details><summary>36 statements not included (click to expand)</summary>
 
 | Statement | Why it could not be validated |
 |---|---|
 | `arnold-counterexample` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
-| `atomic-gaussian` | grading did not complete in our run |
 | `catalan` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `coarse-assembly` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
-| `continuum-coulomb-hardness` | OpenAI's proof is too large for the sandbox's 10 GB disk |
-| `cycle-clique-ramsey` | grading did not complete in our run |
+| `continuum-coulomb-hardness` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
+| `cycle-clique-ramsey` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
 | `defocusing-nls` | left out by design: OAI.DefocusingNLS.sobolevProduct: open definition the theorem statements depend on |
 | `dirichlet-seven-eighths` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `duke-prime-degree` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `einstein-four` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
-| `erdos-reciprocal` | OpenAI's proof is too large for the sandbox's 10 GB disk |
+| `erdos-reciprocal` | grading did not complete in our run |
 | `hecke-seven-eighths` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
-| `invariant-ising` | grading did not complete in our run |
 | `jacobsthal` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `jacobsthal-improved` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `joint-dickman` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
-| `koebe-circle-domains` | OpenAI's proof is too large for the sandbox's 10 GB disk |
+| `koebe-circle-domains` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `matrix-multiplication` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `mub-six` | grading did not complete in our run |
 | `nuclear-ultrapower` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
 | `occupied-overlap` | OpenAI's proof writes a definition with the same text but it compiles differently, so the exact-match check fails |
 | `ordinary-two-point-correlations` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
-| `ostmann-complete` | grading did not complete in our run |
+| `ostmann-complete` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `ostmann-primes` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
-| `patterson-first-moment` | OpenAI's proof is too large for the sandbox's 10 GB disk |
-| `planar-packing` | grading did not complete in our run |
-| `polycyclic-recognition` | grading did not complete in our run |
+| `patterson-first-moment` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
+| `planar-packing` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
+| `polycyclic-recognition` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `quasi-riemann-hypothesis` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
-| `ramsey-five` | grading did not complete in our run |
 | `riesz-quantitative` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
-| `sat-computability` | grading did not complete in our run |
-| `sharp-log-ramsey` | grading did not complete in our run |
 | `siegel-zeros` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
-| `simple-amenable` | grading did not complete in our run |
-| `simple-overgroups` | grading did not complete in our run |
-| `sk-barriers` | grading did not complete in our run |
-| `sk-high-temperature` | grading did not complete in our run |
-| `sk-ratio` | grading did not complete in our run |
-| `smooth-yau` | grading did not complete in our run |
 | `snaky-certificate` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
-| `snaky-conditional` | grading did not complete in our run |
-| `snaky-twenty-one` | grading did not complete in our run |
+| `snaky-twenty-one` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
 | `spin-angle` | OpenAI's proof writes a definition with the same text but it compiles differently, so the exact-match check fails |
-| `standard-map-components` | grading did not complete in our run |
-| `strong-kadison-kastler` | grading did not complete in our run |
-| `surface-immersion` | grading did not complete in our run |
-| `taming-compatibility` | grading did not complete in our run |
+| `surface-immersion` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
+| `taming-compatibility` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
 | `totient-asymptotic` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `totient-companion-zero` | OpenAI's proof uses a Lean library beyond Mathlib that the sandbox does not include |
 | `universal-tensor-squares` | OpenAI's proof needs more than the sandbox's 8 GB of memory to build |
@@ -175,7 +160,7 @@ Grading an accepted proof took 16 minutes at the median. Each task's result is i
 
 The verifier was also tested against submissions that try to cheat (`sorry` in several disguises,
 fake axioms, `native_decide`, weakened statements, redefined definitions, planted reward files, code
-that runs during grading). All scored 0. A no-op agent scores 0.
+that runs during grading, a build that fills the verifier's disk). All scored 0. A no-op agent scores 0.
 
 ## Notes
 
diff --git a/generator/README.md b/generator/README.md
index 0f74bcd51fff2ec2f70390a4688caf3bd816f6ff..87bd6bdf6bd51a7731fa47b128e9c2eead94b21b 100644
--- a/generator/README.md
+++ b/generator/README.md
@@ -50,7 +50,7 @@ per challenge, all from the data:
 |---|---|---|
 | Definition holes | 9 configs | Comparator checks only the name and type of a `definition_names` entry, never its body, so a solution could redefine `ProperColoring := False` and prove the theorem trivially. A hole whose body is written out is removed from the config, so it must match exactly (8 configs). A hole whose body is `sorry` stays open only if its name appears nowhere else in the file (`ElementaryPositivity`, whose type is the claim). `DefocusingNLS` has a `sorry` inside a definition the theorems use and is excluded. |
 | Reference proofs needing other Lean libraries | 23 | PrimeNumberTheoremAnd, StrongPNT, RellichKondrachov and others. Every statement imports only Mathlib, so agents are unaffected; only those official proofs cannot be built in this image. |
-| Grading time | all | 1 h plus 15 min per MB of the reference proof, at most 6 h. |
+| Grading time | all | 3 h plus 15 min per MB of the reference proof, at most 6 h. |
 
 The 137 result families without Lean statements have no machine-checkable target and are not part
 of the dataset.
@@ -67,7 +67,8 @@ agent sandbox (snapshot)                    verifier sandbox (same snapshot, fre
   collect hook: pack_submission.py               3. iptables: grader uid gets no sockets
   -> /logs/artifacts/submission.tar ---------->  4. Comparator as `grader`: landrun builds, statement
      (one plain file crosses)                       + axiom check, kernel replay
-                                                 5. kill grader processes, root writes reward.txt
+                                                 5. kill grader processes, delete the build,
+                                                    root writes reward.txt
 ```
 
 Harbor's `[verifier] environment_mode = "separate"` (0.24+) provides the second sandbox. Both
@@ -88,6 +89,7 @@ after 10 minutes, which is less than the image build takes, hence `snapshot.py`.
 | Re-export the trusted Challenge module | its theorems are `sorry` | `import-challenge` |
 | Plant `reward.txt` or a fake `comparator` in the agent sandbox | grading runs in a separate fresh sandbox | `agent-side-reward` |
 | Lean code run while the verifier builds the proof (`run_cmd`, `#eval`, macros) writes `reward.txt`, runs a shell or opens a connection | landrun (read-only file system except `.lake`, execution limited to Lean and git, TCP denied), unprivileged `grader` user, iptables owner rule; root writes the reward only after killing every grader process | `compile-time-reward` |
+| Exhaust the verifier's disk or memory so grading cannot finish | Lake by-products are emptied as the build runs; the build is deleted before the verdict is written, so a full disk still yields reward 0; memory and time are bounded (8 GB, `grade_timeout_sec`) and a killed build scores 0 | `fill-disk` |
 | Symlinks, path traversal, device files, hard links, huge archives | the collect hook packs only regular files; `grade.py` accepts only regular `Submission.lean` and `Submission/**.lean` whose path components are Lean identifiers | `symlink`, `tests/test_grade_unpack.py` |
 | Read OpenAI's proof | not in the image; the agent phase has no network | network probe (below) |
 | A background process in the agent sandbox interfering with grading | Harbor stops the agent sandbox before the verifier starts and empties the verifier's log directory first; on Daytona the two sandboxes share no files | by construction |
@@ -117,7 +119,7 @@ All through `harbor run` on Daytona with the separate verifier; none involves a
 | Oracle over every task (official proofs) | 1 where the proof fits the image | see the dataset card and each task's `oracle_result` |
 | All 405 statements built in one sandbox (`lake build`, 3 threads) | compile | 405 of 405; median 5 s, 90th percentile 7 s, slowest 220 s |
 | No-op agent | 0 | 0, "no submission" |
-| 11 adversarial submissions (`attacks.py`) | 0 | 0 on all eleven, each rejected for the intended reason; the build-time probe printed `write denied`, and neither a shell nor git's HTTP helper could be executed |
+| 12 adversarial submissions (`attacks.py`) | 0 | 0 on all twelve, each rejected for the intended reason (`fill-disk` on Oct 9 with the final grader: "ran out of disk space" after 85 s); the build-time probe printed `write denied`, and neither a shell nor git's HTTP helper could be executed |
 | Agent-phase view (oracle agent printing its environment) | no network | DNS fails, HTTPS to github.com times out, a direct-IP connection is refused; `LEAN_NUM_THREADS=3` is set; `/tests` is absent |
 | Oracle with `--allow-agent-host github.com` | 1 | 1 (positive control for the network check) |
 | landrun and iptables in a Daytona sandbox | both enforce | as described above |
@@ -144,9 +146,13 @@ details matter:
 ## Limits
 
 - **Daytona's per-sandbox caps: 4 CPUs, 8 GB memory, 10 GB disk.** The 10 GB is working space on
-  top of the image. Compiled output of a large proof can exceed it: reference proofs up to about
-  1.8 MB of Lean source grade fine, while 3.5 MB and 4.8 MB ones ran out of disk. The oracle results
-  record which tasks are affected. Higher per-sandbox limits remove the constraint.
+  top of the image. Lake compiles every module to C as well as to `.olean`; grading never reads the
+  C, but for the largest proofs it reached 9.4 GB and filled the disk. `grade.py` therefore empties
+  the generated `.c` files every 15 seconds while the build runs (as the `grader` user; Lake keeps
+  the recorded hash and does not rebuild), and deletes the build before writing its verdict, so a
+  build that still fills the disk scores 0 instead of leaving Harbor waiting for a verifier that
+  can no longer report. Some proofs still need more than 8 GB of memory or 6 hours; the oracle
+  results record which. Higher per-sandbox limits remove the constraint.
 - **The agent sandbox contains Comparator,** since agent and verifier share one image. Nothing in it
   is secret: the statement and the rules are public, and `lake build` plus `#print axioms` already
   tell an agent whether a proof is complete.
diff --git a/generator/openai_math_harbor/adapter.py b/generator/openai_math_harbor/adapter.py
index b3c01d78527fa8c13bd2d6243486f24c33193a5d..368805d56b5a2fe856969ca3cc85e1841b47d98e 100644
--- a/generator/openai_math_harbor/adapter.py
+++ b/generator/openai_math_harbor/adapter.py
@@ -221,13 +221,15 @@ def validation_block(written: list[dict]) -> str:
     if not results:
         return "The oracle has not been run on this build of the dataset."
     counts: dict[str, list[str]] = {}
+    # No verdict: not run yet, or the run was interrupted (`error: ...`) rather than graded.
+    pending = [w["task"] for w in written if w.get("oracle_result") not in ORACLE_MEANING]
     for w in results:
-        counts.setdefault(w["oracle_result"], []).append(w["task"])
-    order = [k for k in ORACLE_MEANING if k in counts] + sorted(k for k in counts if k not in ORACLE_MEANING)
+        if w["oracle_result"] in ORACLE_MEANING:
+            counts.setdefault(w["oracle_result"], []).append(w["task"])
+    order = [k for k in ORACLE_MEANING if k in counts]
     rows = ["| Oracle result | Tasks | Meaning |", "|---|---|---|"]
     for key in order:
-        rows.append(f"| {key} | {len(counts[key])} | {ORACLE_MEANING.get(key, 'grading did not complete')} |")
-    pending = [w["task"] for w in written if not w.get("oracle_result")]
+        rows.append(f"| {key} | {len(counts[key])} | {ORACLE_MEANING[key]} |")
     if pending:
         rows.append(f"| not completed | {len(pending)} | grading did not complete in our run |")
     lines = ["\n".join(rows), ""]
@@ -236,7 +238,7 @@ def validation_block(written: list[dict]) -> str:
     if secs:
         lines.append(f"Grading an accepted proof took {secs[len(secs) // 2] // 60} minutes at the median. "
                      "Each task's result is in its `task.toml` (`oracle_result`) and in `manifest.parquet`.")
-    unexplained = sorted(t for k in counts if k not in ORACLE_MEANING or k in ("rejected", "timeout") for t in counts[k])
+    unexplained = sorted(t for k in ("rejected", "timeout") for t in counts.get(k, []))
     if unexplained:
         lines.append("Other: " + ", ".join(f"`{t}`" for t in unexplained) + ".")
     return "\n".join(lines)
@@ -339,7 +341,7 @@ def main() -> None:
         if args.only_validated and entry["oracle_result"] != "accepted":
             shutil.rmtree(args.out / "tasks" / entry["task"])
             not_validated.append({"task": entry["task"], "challenge": c.name, "family": c.family.number,
-                                  "reason": entry["oracle_result"] or "pending"})
+                                  "reason": entry["oracle_result"] if entry["oracle_result"] in ORACLE_MEANING else "not completed"})
             continue
         written.append(entry)
     (args.out / "manifest.json").write_text(json.dumps({
diff --git a/generator/openai_math_harbor/attacks.py b/generator/openai_math_harbor/attacks.py
index 500baaaddffd9508ef186aac9255e814627d6298..ca22553387ea22438156f7f66be8967da9267c63 100644
--- a/generator/openai_math_harbor/attacks.py
+++ b/generator/openai_math_harbor/attacks.py
@@ -109,6 +109,21 @@ def openaiMathProbeRun (cmd : String) (args : Array String) : IO String := do
     return _write("/workspace/Submission.lean", lean)
 
 
+def attack_fill_disk(t: Path) -> str:
+    # Lean code run at build time writes 12 GB into .lake (the one place landrun lets the build
+    # write), more than the verifier's 10 GB disk. Grading must still finish with reward 0: a full
+    # disk once made the verifier unable to report, leaving Harbor waiting until its timeout.
+    probe = """
+#eval show IO Unit from do
+  let chunk := "".pushn 'x' (1 <<< 20)
+  let h ← IO.FS.Handle.mk ".lake/openai-math-fill" .write
+  for _ in [0:12288] do
+    h.putStr chunk
+"""
+    lean = _challenge(t).replace("import Mathlib\n", "import Mathlib\n" + probe, 1)
+    return _write("/workspace/Submission.lean", lean)
+
+
 def attack_agent_side_reward(t: Path) -> str:
     # Plant rewards and fake comparator output in the agent sandbox; the verifier runs elsewhere.
     return (
@@ -169,6 +184,7 @@ ATTACKS = {
     "weaken-statement": (attack_weaken_statement, "same theorem name, statement changed to True"),
     "import-challenge": (attack_import_challenge, "import the trusted Challenge module and re-export it"),
     "compile-time-reward": (attack_compile_time_reward, "Lean code run at build time plants reward.txt and dials the daemon"),
+    "fill-disk": (attack_fill_disk, "Lean code run at build time fills the verifier's disk"),
     "agent-side-reward": (attack_agent_side_reward, "plant reward files and a fake comparator in the agent sandbox"),
     "symlink": (attack_symlink, "submit symlinks instead of files"),
 }
diff --git a/generator/openai_math_harbor/report.py b/generator/openai_math_harbor/report.py
index e797e33062bce6576e9cdf6689054128bca3d1cf..bd047fda468241967bf64a4d20e6d03bfbc5d15e 100644
--- a/generator/openai_math_harbor/report.py
+++ b/generator/openai_math_harbor/report.py
@@ -17,7 +17,7 @@ from pathlib import Path
 
 # Checked in order against grading.json's reason and the Comparator log.
 CAUSES = [
-    ("disk", ("no space left on device",)),
+    ("disk", ("no space left on device", "ran out of disk space")),
     # The reference proof states a definition with the same text as the challenge, but compiled in
     # its own library's context it elaborates to a different term, so exact matching fails.
     ("definition-mismatch", ("Const does not match between challenge and target",)),
diff --git a/generator/openai_math_harbor/templates/dataset_README.md b/generator/openai_math_harbor/templates/dataset_README.md
index 02542e313342350fe53b11ff71f0a483383bb64c..79169fcdbf934c878b0194ad7f5f5c8c40bf3640 100644
--- a/generator/openai_math_harbor/templates/dataset_README.md
+++ b/generator/openai_math_harbor/templates/dataset_README.md
@@ -111,7 +111,7 @@ $not_validated
 
 The verifier was also tested against submissions that try to cheat (`sorry` in several disguises,
 fake axioms, `native_decide`, weakened statements, redefined definitions, planted reward files, code
-that runs during grading). All scored 0. A no-op agent scores 0.
+that runs during grading, a build that fills the verifier's disk). All scored 0. A no-op agent scores 0.
 
 ## Notes
 
diff --git a/generator/openai_math_harbor/templates/tests/grade.py b/generator/openai_math_harbor/templates/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/generator/openai_math_harbor/templates/tests/grade.py
+++ b/generator/openai_math_harbor/templates/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/generator/tests/test_grade_unpack.py b/generator/tests/test_grade_unpack.py
index c8eb8b143a0faba44b28ac1d9c2358f50f36b84a..13b60944a8bc4feef60b640a20be45be553c8035 100644
--- a/generator/tests/test_grade_unpack.py
+++ b/generator/tests/test_grade_unpack.py
@@ -136,6 +136,33 @@ class UnpackTest(unittest.TestCase):
         self.assertEqual(reward, 0)
         self.assertIn("out of memory", reason)
 
+    def test_empty_scratch_only_touches_finished_modules(self):
+        import os, subprocess, sys
+        ir = self.tmp / "ir" / "Submission"; ir.mkdir(parents=True)
+        def module(name, hash_age=None):
+            for ext in (".setup.json", ".c"):
+                (ir / (name + ext)).write_text("x" * 100)
+            os.utime(ir / (name + ".setup.json"), (1000, 1000))
+            if hash_age is not None:
+                (ir / (name + ".c.hash")).write_text("1")
+                os.utime(ir / (name + ".c.hash"), (1000 + hash_age, 1000 + hash_age))
+        module("Done", hash_age=5)       # Lake finished: hash written after the setup
+        module("Building")               # Lean or leanir may still read the setup
+        module("Stale", hash_age=-5)     # hash older than the setup: a rebuild is in progress
+        subprocess.run([sys.executable, "-c", grade.EMPTY_SCRATCH, str(self.tmp / "ir")], check=True)
+        size = lambda n: (ir / n).stat().st_size
+        self.assertEqual((size("Done.setup.json"), size("Done.c")), (0, 0))
+        self.assertEqual((size("Building.setup.json"), size("Building.c")), (100, 100))
+        self.assertEqual((size("Stale.setup.json"), size("Stale.c")), (100, 100))
+        self.assertEqual(size("Done.c.hash"), 1)
+
+    def test_out_of_disk_detection(self):
+        out = "Building Submission\nerror: resource exhausted (error code: 28, no space left on device)\n"
+        reward, reason = grade.classify(1, out)
+        self.assertEqual(reward, 0)
+        self.assertIn("out of disk", reason)
+        self.assertTrue(grade.out_of_disk("OSError: [Errno 28] No space left on device"))
+
     def test_classify(self):
         ok = "Building Challenge\nBuilding Submission\nYour solution is okay!\n"
         self.assertEqual(grade.classify(0, ok)[0], 1)
diff --git a/manifest.json b/manifest.json
index dca74664592ced2aae5607ce92b5d8743acb9844..29a7eb1f26168e49c558c20185f573ff540ceb61 100644
--- a/manifest.json
+++ b/manifest.json
@@ -18,7 +18,7 @@
       "packages": [],
       "grade_timeout_sec": 10850,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 463
+      "oracle_grade_seconds": 47
     },
     {
       "task": "affine-bernstein",
@@ -36,7 +36,7 @@
       "packages": [],
       "grade_timeout_sec": 12598,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1091
+      "oracle_grade_seconds": 1429
     },
     {
       "task": "algorithmic-thin-trees",
@@ -54,7 +54,7 @@
       "packages": [],
       "grade_timeout_sec": 14901,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2906
+      "oracle_grade_seconds": 2392
     },
     {
       "task": "amplitude-damping",
@@ -72,7 +72,7 @@
       "packages": [],
       "grade_timeout_sec": 11055,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1119
+      "oracle_grade_seconds": 432
     },
     {
       "task": "artin-cat0",
@@ -90,7 +90,7 @@
       "packages": [],
       "grade_timeout_sec": 11173,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 792
+      "oracle_grade_seconds": 193
     },
     {
       "task": "artin-parabolic-intersections",
@@ -111,7 +111,7 @@
       "packages": [],
       "grade_timeout_sec": 13770,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1711
+      "oracle_grade_seconds": 2200
     },
     {
       "task": "asymptotically-minimal-littlewood",
@@ -129,7 +129,25 @@
       "packages": [],
       "grade_timeout_sec": 11189,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 919
+      "oracle_grade_seconds": 346
+    },
+    {
+      "task": "atomic-gaussian",
+      "challenge": "AtomicGaussian",
+      "family": "090",
+      "family_title": "Universal optimality of the triangular lattice",
+      "subject": "Convex and metric geometry",
+      "theorems": [
+        "OAI.AtomicTriangular.sharp_gaussian_minorants_atomic"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 215,
+      "bytes": 8527615,
+      "packages": [],
+      "grade_timeout_sec": 18474,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 10626
     },
     {
       "task": "auslander-reiten",
@@ -147,7 +165,7 @@
       "packages": [],
       "grade_timeout_sec": 12914,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2615
+      "oracle_grade_seconds": 2177
     },
     {
       "task": "backward-intertwiners",
@@ -165,7 +183,7 @@
       "packages": [],
       "grade_timeout_sec": 11036,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 823
+      "oracle_grade_seconds": 337
     },
     {
       "task": "balanced-box-routing",
@@ -183,7 +201,7 @@
       "packages": [],
       "grade_timeout_sec": 10948,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 299
+      "oracle_grade_seconds": 141
     },
     {
       "task": "balanced-ryser",
@@ -201,7 +219,7 @@
       "packages": [],
       "grade_timeout_sec": 14307,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1733
+      "oracle_grade_seconds": 1384
     },
     {
       "task": "balanced-three-stack",
@@ -219,7 +237,7 @@
       "packages": [],
       "grade_timeout_sec": 11246,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1956
+      "oracle_grade_seconds": 392
     },
     {
       "task": "ball-packing",
@@ -237,7 +255,7 @@
       "packages": [],
       "grade_timeout_sec": 14876,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 4562
+      "oracle_grade_seconds": 2433
     },
     {
       "task": "ball-packing-necessity",
@@ -255,7 +273,7 @@
       "packages": [],
       "grade_timeout_sec": 12306,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1874
+      "oracle_grade_seconds": 1186
     },
     {
       "task": "barnette-hamiltonian",
@@ -273,7 +291,7 @@
       "packages": [],
       "grade_timeout_sec": 11141,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 752
+      "oracle_grade_seconds": 337
     },
     {
       "task": "bass-torsion-free",
@@ -291,7 +309,7 @@
       "packages": [],
       "grade_timeout_sec": 11720,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1087
+      "oracle_grade_seconds": 650
     },
     {
       "task": "bass-trace",
@@ -309,7 +327,7 @@
       "packages": [],
       "grade_timeout_sec": 11697,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1120
+      "oracle_grade_seconds": 633
     },
     {
       "task": "benjamini-schramm",
@@ -329,7 +347,7 @@
       "packages": [],
       "grade_timeout_sec": 16950,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 3592
+      "oracle_grade_seconds": 2752
     },
     {
       "task": "bernoulli-nonflat-in-seven",
@@ -347,7 +365,7 @@
       "packages": [],
       "grade_timeout_sec": 11376,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1160
+      "oracle_grade_seconds": 664
     },
     {
       "task": "biholder-transport",
@@ -365,7 +383,7 @@
       "packages": [],
       "grade_timeout_sec": 13682,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 3094
+      "oracle_grade_seconds": 2398
     },
     {
       "task": "bin-packing-gap",
@@ -383,7 +401,7 @@
       "packages": [],
       "grade_timeout_sec": 17601,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1412
+      "oracle_grade_seconds": 1119
     },
     {
       "task": "binary-edit-lower",
@@ -401,7 +419,7 @@
       "packages": [],
       "grade_timeout_sec": 11218,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 790
+      "oracle_grade_seconds": 280
     },
     {
       "task": "binary-matching",
@@ -419,7 +437,7 @@
       "packages": [],
       "grade_timeout_sec": 11355,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1132
+      "oracle_grade_seconds": 524
     },
     {
       "task": "binary-sweep",
@@ -437,7 +455,7 @@
       "packages": [],
       "grade_timeout_sec": 11836,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1210
+      "oracle_grade_seconds": 768
     },
     {
       "task": "bipartite-crossing",
@@ -455,7 +473,7 @@
       "packages": [],
       "grade_timeout_sec": 11604,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 790
+      "oracle_grade_seconds": 448
     },
     {
       "task": "bogomolov-pop-injectivity",
@@ -473,7 +491,7 @@
       "packages": [],
       "grade_timeout_sec": 10939,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 445
+      "oracle_grade_seconds": 159
     },
     {
       "task": "boltzmann-nonuniqueness",
@@ -491,7 +509,7 @@
       "packages": [],
       "grade_timeout_sec": 17397,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1584
+      "oracle_grade_seconds": 1339
     },
     {
       "task": "boone-higman",
@@ -509,7 +527,7 @@
       "packages": [],
       "grade_timeout_sec": 12318,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 976
+      "oracle_grade_seconds": 379
     },
     {
       "task": "borsuk-nine",
@@ -527,7 +545,7 @@
       "packages": [],
       "grade_timeout_sec": 13972,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 893
+      "oracle_grade_seconds": 1206
     },
     {
       "task": "bounded-recovery",
@@ -545,7 +563,7 @@
       "packages": [],
       "grade_timeout_sec": 11131,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1002
+      "oracle_grade_seconds": 494
     },
     {
       "task": "bounded-tree-potentials",
@@ -563,7 +581,7 @@
       "packages": [],
       "grade_timeout_sec": 11336,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 758
+      "oracle_grade_seconds": 358
     },
     {
       "task": "bounded-treewidth-l1",
@@ -581,7 +599,7 @@
       "packages": [],
       "grade_timeout_sec": 11437,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 791
+      "oracle_grade_seconds": 283
     },
     {
       "task": "brenier",
@@ -613,7 +631,7 @@
       "packages": [],
       "grade_timeout_sec": 12179,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2528
+      "oracle_grade_seconds": 440
     },
     {
       "task": "brennan",
@@ -631,7 +649,7 @@
       "packages": [],
       "grade_timeout_sec": 11423,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1116
+      "oracle_grade_seconds": 507
     },
     {
       "task": "brennan-sharp",
@@ -649,7 +667,7 @@
       "packages": [],
       "grade_timeout_sec": 11440,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 895
+      "oracle_grade_seconds": 354
     },
     {
       "task": "buffered-ising",
@@ -667,7 +685,7 @@
       "packages": [],
       "grade_timeout_sec": 10922,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2053
+      "oracle_grade_seconds": 127
     },
     {
       "task": "c0-absorption",
@@ -685,7 +703,7 @@
       "packages": [],
       "grade_timeout_sec": 11129,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1155
+      "oracle_grade_seconds": 256
     },
     {
       "task": "c1-entropy-counterexample",
@@ -703,7 +721,7 @@
       "packages": [],
       "grade_timeout_sec": 11668,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1331
+      "oracle_grade_seconds": 647
     },
     {
       "task": "ch-obstruction",
@@ -721,7 +739,7 @@
       "packages": [],
       "grade_timeout_sec": 10979,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1476
+      "oracle_grade_seconds": 124
     },
     {
       "task": "cks-bondi-penrose",
@@ -740,7 +758,7 @@
       "packages": [],
       "grade_timeout_sec": 13346,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1456
+      "oracle_grade_seconds": 1417
     },
     {
       "task": "cannon-geometric-action",
@@ -758,7 +776,7 @@
       "packages": [],
       "grade_timeout_sec": 13413,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 970
+      "oracle_grade_seconds": 1075
     },
     {
       "task": "cayley-percolation",
@@ -776,7 +794,7 @@
       "packages": [],
       "grade_timeout_sec": 13179,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1048
+      "oracle_grade_seconds": 698
     },
     {
       "task": "character-criterion",
@@ -794,7 +812,7 @@
       "packages": [],
       "grade_timeout_sec": 11832,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1889
+      "oracle_grade_seconds": 991
     },
     {
       "task": "choiceless-polynomial-time",
@@ -812,7 +830,7 @@
       "packages": [],
       "grade_timeout_sec": 11523,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 574
+      "oracle_grade_seconds": 286
     },
     {
       "task": "circulant-hadamard",
@@ -830,7 +848,7 @@
       "packages": [],
       "grade_timeout_sec": 11160,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1652
+      "oracle_grade_seconds": 235
     },
     {
       "task": "classical-on",
@@ -848,7 +866,7 @@
       "packages": [],
       "grade_timeout_sec": 11324,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1392
+      "oracle_grade_seconds": 752
     },
     {
       "task": "clique-free-log",
@@ -866,7 +884,7 @@
       "packages": [],
       "grade_timeout_sec": 11006,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 794
+      "oracle_grade_seconds": 193
     },
     {
       "task": "coboundary-expanders",
@@ -884,7 +902,7 @@
       "packages": [],
       "grade_timeout_sec": 12113,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1082
+      "oracle_grade_seconds": 437
     },
     {
       "task": "common-bases-fpras",
@@ -902,7 +920,7 @@
       "packages": [],
       "grade_timeout_sec": 12002,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 950
+      "oracle_grade_seconds": 347
     },
     {
       "task": "commuting-derivations",
@@ -920,7 +938,7 @@
       "packages": [],
       "grade_timeout_sec": 10921,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 426
+      "oracle_grade_seconds": 81
     },
     {
       "task": "compact-banach",
@@ -938,7 +956,7 @@
       "packages": [],
       "grade_timeout_sec": 10986,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1439
+      "oracle_grade_seconds": 303
     },
     {
       "task": "complete-crossing",
@@ -956,7 +974,7 @@
       "packages": [],
       "grade_timeout_sec": 12117,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 800
+      "oracle_grade_seconds": 344
     },
     {
       "task": "complete-crouzeix",
@@ -974,7 +992,7 @@
       "packages": [],
       "grade_timeout_sec": 11386,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1153
+      "oracle_grade_seconds": 546
     },
     {
       "task": "complex-cancellation",
@@ -992,7 +1010,7 @@
       "packages": [],
       "grade_timeout_sec": 11018,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 971
+      "oracle_grade_seconds": 355
     },
     {
       "task": "conductivity",
@@ -1010,7 +1028,7 @@
       "packages": [],
       "grade_timeout_sec": 14056,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1871
+      "oracle_grade_seconds": 1793
     },
     {
       "task": "conjugate-points",
@@ -1028,7 +1046,7 @@
       "packages": [],
       "grade_timeout_sec": 11951,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1609
+      "oracle_grade_seconds": 595
     },
     {
       "task": "contingency-tables",
@@ -1048,7 +1066,7 @@
       "packages": [],
       "grade_timeout_sec": 18988,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 3243
+      "oracle_grade_seconds": 3262
     },
     {
       "task": "continuous-circle-weight",
@@ -1066,7 +1084,7 @@
       "packages": [],
       "grade_timeout_sec": 11617,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1217
+      "oracle_grade_seconds": 530
     },
     {
       "task": "continuum-transition",
@@ -1084,7 +1102,7 @@
       "packages": [],
       "grade_timeout_sec": 11427,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1607
+      "oracle_grade_seconds": 942
     },
     {
       "task": "coordinate-sweeps",
@@ -1103,7 +1121,7 @@
       "packages": [],
       "grade_timeout_sec": 11777,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1051
+      "oracle_grade_seconds": 383
     },
     {
       "task": "coordinate-trace",
@@ -1121,7 +1139,7 @@
       "packages": [],
       "grade_timeout_sec": 11676,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1254
+      "oracle_grade_seconds": 427
     },
     {
       "task": "cotype",
@@ -1139,7 +1157,7 @@
       "packages": [],
       "grade_timeout_sec": 11253,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1088
+      "oracle_grade_seconds": 444
     },
     {
       "task": "coulomb-counterexample",
@@ -1157,7 +1175,7 @@
       "packages": [],
       "grade_timeout_sec": 11220,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 891
+      "oracle_grade_seconds": 349
     },
     {
       "task": "coulomb-ionization",
@@ -1175,7 +1193,7 @@
       "packages": [],
       "grade_timeout_sec": 14952,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2115
+      "oracle_grade_seconds": 2150
     },
     {
       "task": "coulomb-radii",
@@ -1193,7 +1211,7 @@
       "packages": [],
       "grade_timeout_sec": 14272,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2054
+      "oracle_grade_seconds": 1444
     },
     {
       "task": "courtade-kumar",
@@ -1211,7 +1229,7 @@
       "packages": [],
       "grade_timeout_sec": 11095,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 468
+      "oracle_grade_seconds": 251
     },
     {
       "task": "covering-density",
@@ -1229,7 +1247,7 @@
       "packages": [],
       "grade_timeout_sec": 12384,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1458
+      "oracle_grade_seconds": 762
     },
     {
       "task": "critical-percolation",
@@ -1247,7 +1265,7 @@
       "packages": [],
       "grade_timeout_sec": 13419,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2360
+      "oracle_grade_seconds": 1685
     },
     {
       "task": "critical-sk",
@@ -1267,7 +1285,7 @@
       "packages": [],
       "grade_timeout_sec": 11601,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 892
+      "oracle_grade_seconds": 322
     },
     {
       "task": "critical-sk-mixing",
@@ -1285,7 +1303,7 @@
       "packages": [],
       "grade_timeout_sec": 15840,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2139
+      "oracle_grade_seconds": 2056
     },
     {
       "task": "critical-strip-mass",
@@ -1303,7 +1321,7 @@
       "packages": [],
       "grade_timeout_sec": 13317,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1959
+      "oracle_grade_seconds": 1033
     },
     {
       "task": "critical-z3",
@@ -1321,7 +1339,7 @@
       "packages": [],
       "grade_timeout_sec": 11215,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 859
+      "oracle_grade_seconds": 230
     },
     {
       "task": "crouzeix-hilbert",
@@ -1339,7 +1357,7 @@
       "packages": [],
       "grade_timeout_sec": 11451,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1215
+      "oracle_grade_seconds": 657
     },
     {
       "task": "cubic-torus",
@@ -1357,7 +1375,7 @@
       "packages": [],
       "grade_timeout_sec": 17093,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2541
+      "oracle_grade_seconds": 1840
     },
     {
       "task": "cycle-decomposition",
@@ -1375,7 +1393,7 @@
       "packages": [],
       "grade_timeout_sec": 11698,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 827
+      "oracle_grade_seconds": 295
     },
     {
       "task": "cylinder-covering",
@@ -1393,7 +1411,7 @@
       "packages": [],
       "grade_timeout_sec": 13511,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2016
+      "oracle_grade_seconds": 1626
     },
     {
       "task": "daugavet-moduli",
@@ -1413,7 +1431,7 @@
       "packages": [],
       "grade_timeout_sec": 10981,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 462
+      "oracle_grade_seconds": 230
     },
     {
       "task": "degree-rigidity",
@@ -1431,7 +1449,7 @@
       "packages": [],
       "grade_timeout_sec": 14330,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1821
+      "oracle_grade_seconds": 2006
     },
     {
       "task": "deligne-drinfeld",
@@ -1449,7 +1467,7 @@
       "packages": [],
       "grade_timeout_sec": 12022,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 921
+      "oracle_grade_seconds": 563
     },
     {
       "task": "depth-five",
@@ -1467,7 +1485,7 @@
       "packages": [],
       "grade_timeout_sec": 11592,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1051
+      "oracle_grade_seconds": 462
     },
     {
       "task": "depth-three",
@@ -1485,7 +1503,7 @@
       "packages": [],
       "grade_timeout_sec": 11723,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 776
+      "oracle_grade_seconds": 398
     },
     {
       "task": "diamond-distortion",
@@ -1503,7 +1521,7 @@
       "packages": [],
       "grade_timeout_sec": 11225,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 771
+      "oracle_grade_seconds": 268
     },
     {
       "task": "diluted-spin",
@@ -1521,7 +1539,7 @@
       "packages": [],
       "grade_timeout_sec": 13284,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1537
+      "oracle_grade_seconds": 1199
     },
     {
       "task": "dimension-ten-channel",
@@ -1539,7 +1557,7 @@
       "packages": [],
       "grade_timeout_sec": 13567,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2360
+      "oracle_grade_seconds": 1785
     },
     {
       "task": "dimension-ten-pair",
@@ -1557,7 +1575,7 @@
       "packages": [],
       "grade_timeout_sec": 12972,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1766
+      "oracle_grade_seconds": 1663
     },
     {
       "task": "direct-crouzeix",
@@ -1576,7 +1594,7 @@
       "packages": [],
       "grade_timeout_sec": 11082,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1202
+      "oracle_grade_seconds": 429
     },
     {
       "task": "directed-feedback",
@@ -1594,7 +1612,7 @@
       "packages": [],
       "grade_timeout_sec": 18455,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1662
+      "oracle_grade_seconds": 1607
     },
     {
       "task": "directional-ballisticity",
@@ -1612,7 +1630,7 @@
       "packages": [],
       "grade_timeout_sec": 15311,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1730
+      "oracle_grade_seconds": 1803
     },
     {
       "task": "directional-walk",
@@ -1630,7 +1648,7 @@
       "packages": [],
       "grade_timeout_sec": 11656,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 968
+      "oracle_grade_seconds": 459
     },
     {
       "task": "discrete-lipschitz-free",
@@ -1648,7 +1666,7 @@
       "packages": [],
       "grade_timeout_sec": 10956,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 662
+      "oracle_grade_seconds": 199
     },
     {
       "task": "disk-maximal",
@@ -1666,7 +1684,7 @@
       "packages": [],
       "grade_timeout_sec": 11673,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 756
+      "oracle_grade_seconds": 415
     },
     {
       "task": "dixmier",
@@ -1684,7 +1702,7 @@
       "packages": [],
       "grade_timeout_sec": 10943,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 332
+      "oracle_grade_seconds": 112
     },
     {
       "task": "dixmier-all-discrete",
@@ -1702,7 +1720,7 @@
       "packages": [],
       "grade_timeout_sec": 10963,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 471
+      "oracle_grade_seconds": 194
     },
     {
       "task": "doubling-hilbert",
@@ -1720,7 +1738,7 @@
       "packages": [],
       "grade_timeout_sec": 10941,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 454
+      "oracle_grade_seconds": 179
     },
     {
       "task": "dutta-domain",
@@ -1738,7 +1756,7 @@
       "packages": [],
       "grade_timeout_sec": 12968,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1913
+      "oracle_grade_seconds": 1966
     },
     {
       "task": "dyadic-avoidance",
@@ -1756,7 +1774,7 @@
       "packages": [],
       "grade_timeout_sec": 11132,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 796
+      "oracle_grade_seconds": 198
     },
     {
       "task": "edit-approximation",
@@ -1774,7 +1792,7 @@
       "packages": [],
       "grade_timeout_sec": 17632,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 3183
+      "oracle_grade_seconds": 2717
     },
     {
       "task": "edit-distance",
@@ -1792,7 +1810,7 @@
       "packages": [],
       "grade_timeout_sec": 12317,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1069
+      "oracle_grade_seconds": 320
     },
     {
       "task": "egyptian-fractions",
@@ -1818,7 +1836,7 @@
       "packages": [],
       "grade_timeout_sec": 12262,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 763
+      "oracle_grade_seconds": 832
     },
     {
       "task": "eilenberg-ganea",
@@ -1836,7 +1854,7 @@
       "packages": [],
       "grade_timeout_sec": 12346,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 4559
+      "oracle_grade_seconds": 3411
     },
     {
       "task": "elasticity-uniqueness",
@@ -1854,7 +1872,7 @@
       "packages": [],
       "grade_timeout_sec": 12247,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1681
+      "oracle_grade_seconds": 755
     },
     {
       "task": "elementary-positivity",
@@ -1872,7 +1890,7 @@
       "packages": [],
       "grade_timeout_sec": 13926,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2371
+      "oracle_grade_seconds": 2005
     },
     {
       "task": "entangled-games",
@@ -1890,7 +1908,7 @@
       "packages": [],
       "grade_timeout_sec": 11118,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 786
+      "oracle_grade_seconds": 276
     },
     {
       "task": "entropy-photon-number",
@@ -1908,7 +1926,7 @@
       "packages": [],
       "grade_timeout_sec": 12060,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 998
+      "oracle_grade_seconds": 517
     },
     {
       "task": "euclidean-five-color",
@@ -1928,7 +1946,7 @@
       "packages": [],
       "grade_timeout_sec": 12193,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1407
+      "oracle_grade_seconds": 597
     },
     {
       "task": "euclidean-ramsey",
@@ -1946,7 +1964,7 @@
       "packages": [],
       "grade_timeout_sec": 11059,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1062
+      "oracle_grade_seconds": 303
     },
     {
       "task": "euclidean-ramsey-circle",
@@ -1964,7 +1982,7 @@
       "packages": [],
       "grade_timeout_sec": 11086,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 883
+      "oracle_grade_seconds": 304
     },
     {
       "task": "euclidean-ramsey-nine",
@@ -1982,7 +2000,7 @@
       "packages": [],
       "grade_timeout_sec": 10869,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 600
+      "oracle_grade_seconds": 137
     },
     {
       "task": "euclidean-ramsey-quadratic",
@@ -2000,7 +2018,7 @@
       "packages": [],
       "grade_timeout_sec": 11069,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 672
+      "oracle_grade_seconds": 297
     },
     {
       "task": "euclidean-ramsey-spherical",
@@ -2018,7 +2036,7 @@
       "packages": [],
       "grade_timeout_sec": 11065,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 735
+      "oracle_grade_seconds": 222
     },
     {
       "task": "euclidean-ramsey-transitive",
@@ -2036,7 +2054,7 @@
       "packages": [],
       "grade_timeout_sec": 11065,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1235
+      "oracle_grade_seconds": 352
     },
     {
       "task": "even-barker",
@@ -2054,7 +2072,7 @@
       "packages": [],
       "grade_timeout_sec": 11169,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 743
+      "oracle_grade_seconds": 251
     },
     {
       "task": "exact-fourier",
@@ -2072,7 +2090,7 @@
       "packages": [],
       "grade_timeout_sec": 11563,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 765
+      "oracle_grade_seconds": 480
     },
     {
       "task": "exact-infiniteness",
@@ -2091,7 +2109,7 @@
       "packages": [],
       "grade_timeout_sec": 16851,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 4206
+      "oracle_grade_seconds": 4325
     },
     {
       "task": "exact-quantum-factoring",
@@ -2109,7 +2127,7 @@
       "packages": [],
       "grade_timeout_sec": 12978,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1227
+      "oracle_grade_seconds": 1235
     },
     {
       "task": "fkcrt",
@@ -2127,7 +2145,7 @@
       "packages": [],
       "grade_timeout_sec": 12489,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1071
+      "oracle_grade_seconds": 489
     },
     {
       "task": "factor-generation",
@@ -2145,7 +2163,7 @@
       "packages": [],
       "grade_timeout_sec": 12311,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 3699
+      "oracle_grade_seconds": 2800
     },
     {
       "task": "falconer-all-dimensions",
@@ -2163,7 +2181,7 @@
       "packages": [],
       "grade_timeout_sec": 18862,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 4645
+      "oracle_grade_seconds": 5233
     },
     {
       "task": "filling-coefficient",
@@ -2181,7 +2199,7 @@
       "packages": [],
       "grade_timeout_sec": 11427,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1115
+      "oracle_grade_seconds": 480
     },
     {
       "task": "finite-circle",
@@ -2199,7 +2217,7 @@
       "packages": [],
       "grade_timeout_sec": 11212,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 902
+      "oracle_grade_seconds": 285
     },
     {
       "task": "finite-congruence-graph",
@@ -2217,7 +2235,7 @@
       "packages": [],
       "grade_timeout_sec": 10820,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 466
+      "oracle_grade_seconds": 19
     },
     {
       "task": "finite-entropy-separation",
@@ -2235,7 +2253,7 @@
       "packages": [],
       "grade_timeout_sec": 11507,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1422
+      "oracle_grade_seconds": 1014
     },
     {
       "task": "finite-factor",
@@ -2253,7 +2271,7 @@
       "packages": [],
       "grade_timeout_sec": 11802,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1338
+      "oracle_grade_seconds": 745
     },
     {
       "task": "finitistic-asymmetry",
@@ -2271,7 +2289,7 @@
       "packages": [],
       "grade_timeout_sec": 12826,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2003
+      "oracle_grade_seconds": 1462
     },
     {
       "task": "fixed-clause-threshold",
@@ -2289,7 +2307,7 @@
       "packages": [],
       "grade_timeout_sec": 10994,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 655
+      "oracle_grade_seconds": 145
     },
     {
       "task": "forced-navier-stokes-computation",
@@ -2307,7 +2325,7 @@
       "packages": [],
       "grade_timeout_sec": 11246,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 547
+      "oracle_grade_seconds": 302
     },
     {
       "task": "forest-space",
@@ -2325,7 +2343,7 @@
       "packages": [],
       "grade_timeout_sec": 10975,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 741
+      "oracle_grade_seconds": 175
     },
     {
       "task": "formula-hitting",
@@ -2343,7 +2361,7 @@
       "packages": [],
       "grade_timeout_sec": 10911,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 637
+      "oracle_grade_seconds": 63
     },
     {
       "task": "foulkes-howe",
@@ -2361,7 +2379,7 @@
       "packages": [],
       "grade_timeout_sec": 10977,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1045
+      "oracle_grade_seconds": 276
     },
     {
       "task": "four-row-permanent",
@@ -2379,7 +2397,7 @@
       "packages": [],
       "grade_timeout_sec": 10822,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 788
+      "oracle_grade_seconds": 90
     },
     {
       "task": "free-ising",
@@ -2397,7 +2415,7 @@
       "packages": [],
       "grade_timeout_sec": 15777,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2215
+      "oracle_grade_seconds": 2094
     },
     {
       "task": "free-uniform-spanning-forest",
@@ -2415,7 +2433,7 @@
       "packages": [],
       "grade_timeout_sec": 11762,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1286
+      "oracle_grade_seconds": 624
     },
     {
       "task": "gamma-passage",
@@ -2433,7 +2451,7 @@
       "packages": [],
       "grade_timeout_sec": 11652,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 800
+      "oracle_grade_seconds": 466
     },
     {
       "task": "gaussian-finite-entropy",
@@ -2451,7 +2469,7 @@
       "packages": [],
       "grade_timeout_sec": 10905,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 870
+      "oracle_grade_seconds": 165
     },
     {
       "task": "gaussian-information",
@@ -2472,7 +2490,7 @@
       "packages": [],
       "grade_timeout_sec": 11504,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1216
+      "oracle_grade_seconds": 572
     },
     {
       "task": "gaussian-moat",
@@ -2490,7 +2508,7 @@
       "packages": [],
       "grade_timeout_sec": 11338,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1157
+      "oracle_grade_seconds": 422
     },
     {
       "task": "gaussian-propeller",
@@ -2508,7 +2526,7 @@
       "packages": [],
       "grade_timeout_sec": 11325,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1095
+      "oracle_grade_seconds": 496
     },
     {
       "task": "gaussian-replacement",
@@ -2536,7 +2554,7 @@
       "packages": [],
       "grade_timeout_sec": 11484,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1104
+      "oracle_grade_seconds": 486
     },
     {
       "task": "general-mahler",
@@ -2554,7 +2572,7 @@
       "packages": [],
       "grade_timeout_sec": 12728,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 5069
+      "oracle_grade_seconds": 5721
     },
     {
       "task": "generalized-star-height",
@@ -2572,7 +2590,7 @@
       "packages": [],
       "grade_timeout_sec": 11270,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 852
+      "oracle_grade_seconds": 235
     },
     {
       "task": "gotsman-linial",
@@ -2590,7 +2608,7 @@
       "packages": [],
       "grade_timeout_sec": 10954,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 728
+      "oracle_grade_seconds": 72
     },
     {
       "task": "graham-spherical",
@@ -2608,7 +2626,7 @@
       "packages": [],
       "grade_timeout_sec": 10834,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 751
+      "oracle_grade_seconds": 92
     },
     {
       "task": "grothendieck-elementary-expansion",
@@ -2626,7 +2644,7 @@
       "packages": [],
       "grade_timeout_sec": 11322,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1113
+      "oracle_grade_seconds": 436
     },
     {
       "task": "group-ring-determinant",
@@ -2644,7 +2662,7 @@
       "packages": [],
       "grade_timeout_sec": 11412,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1695
+      "oracle_grade_seconds": 788
     },
     {
       "task": "hadamard-cube-fiber",
@@ -2662,7 +2680,7 @@
       "packages": [],
       "grade_timeout_sec": 10835,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 462
+      "oracle_grade_seconds": 64
     },
     {
       "task": "halving-lines",
@@ -2680,7 +2698,7 @@
       "packages": [],
       "grade_timeout_sec": 11598,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 894
+      "oracle_grade_seconds": 410
     },
     {
       "task": "hard-sphere",
@@ -2698,7 +2716,7 @@
       "packages": [],
       "grade_timeout_sec": 13924,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1367
+      "oracle_grade_seconds": 1832
     },
     {
       "task": "harmonic-artin",
@@ -2716,7 +2734,7 @@
       "packages": [],
       "grade_timeout_sec": 13321,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1612
+      "oracle_grade_seconds": 1678
     },
     {
       "task": "harmonic-growth",
@@ -2734,7 +2752,7 @@
       "packages": [],
       "grade_timeout_sec": 12315,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1761
+      "oracle_grade_seconds": 799
     },
     {
       "task": "heilbronn-triangle",
@@ -2753,7 +2771,7 @@
       "packages": [],
       "grade_timeout_sec": 11713,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1117
+      "oracle_grade_seconds": 640
     },
     {
       "task": "heisenberg",
@@ -2771,7 +2789,7 @@
       "packages": [],
       "grade_timeout_sec": 11798,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1829
+      "oracle_grade_seconds": 1106
     },
     {
       "task": "henon-emden",
@@ -2789,7 +2807,7 @@
       "packages": [],
       "grade_timeout_sec": 11381,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1094
+      "oracle_grade_seconds": 584
     },
     {
       "task": "hilbert-crouzeix",
@@ -2810,7 +2828,7 @@
       "packages": [],
       "grade_timeout_sec": 10833,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 927
+      "oracle_grade_seconds": 136
     },
     {
       "task": "honeycomb-bridge-finiteness",
@@ -2828,7 +2846,7 @@
       "packages": [],
       "grade_timeout_sec": 11090,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 711
+      "oracle_grade_seconds": 192
     },
     {
       "task": "honeycomb-free-energy",
@@ -2846,7 +2864,7 @@
       "packages": [],
       "grade_timeout_sec": 10837,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 812
+      "oracle_grade_seconds": 51
     },
     {
       "task": "hot-spots",
@@ -2864,7 +2882,7 @@
       "packages": [],
       "grade_timeout_sec": 12043,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1437
+      "oracle_grade_seconds": 578
     },
     {
       "task": "hyperbolic-cones",
@@ -2882,7 +2900,7 @@
       "packages": [],
       "grade_timeout_sec": 10996,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 849
+      "oracle_grade_seconds": 350
     },
     {
       "task": "hyperbolic-obstruction",
@@ -2900,7 +2918,7 @@
       "packages": [],
       "grade_timeout_sec": 12987,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1194
+      "oracle_grade_seconds": 1278
     },
     {
       "task": "hyperinvariant-subspaces",
@@ -2918,7 +2936,7 @@
       "packages": [],
       "grade_timeout_sec": 11816,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1437
+      "oracle_grade_seconds": 870
     },
     {
       "task": "independent-products",
@@ -2939,7 +2957,7 @@
       "packages": [],
       "grade_timeout_sec": 10960,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 656
+      "oracle_grade_seconds": 250
     },
     {
       "task": "independent-sets",
@@ -2957,7 +2975,7 @@
       "packages": [],
       "grade_timeout_sec": 14830,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1201
+      "oracle_grade_seconds": 1204
     },
     {
       "task": "individual-strong-infiniteness",
@@ -2975,7 +2993,7 @@
       "packages": [],
       "grade_timeout_sec": 11054,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 965
+      "oracle_grade_seconds": 355
     },
     {
       "task": "infinite-matroid",
@@ -2993,7 +3011,7 @@
       "packages": [],
       "grade_timeout_sec": 10957,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 667
+      "oracle_grade_seconds": 111
     },
     {
       "task": "infinite-matroid-corollaries",
@@ -3012,7 +3030,7 @@
       "packages": [],
       "grade_timeout_sec": 10973,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 637
+      "oracle_grade_seconds": 155
     },
     {
       "task": "interpolated-factors",
@@ -3030,7 +3048,35 @@
       "packages": [],
       "grade_timeout_sec": 11924,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1971
+      "oracle_grade_seconds": 1361
+    },
+    {
+      "task": "invariant-ising",
+      "challenge": "InvariantIsing",
+      "family": "234",
+      "family_title": "All-temperature pressure for orthogonally invariant Ising spin glasses",
+      "subject": "Probability and statistical mechanics",
+      "theorems": [
+        "OAI.InvariantIsing.limiting_pressure_of_extreme_limits_unconditional",
+        "OAI.InvariantIsing.limiting_wasserstein_field_pressure_of_extreme_limits_unconditional",
+        "OAI.InvariantIsing.positive_temperature_pressure_unconditional",
+        "OAI.InvariantIsing.ground_state_limit_unconditional",
+        "OAI.InvariantIsing.random_pressure_tendsto_in_measure_unconditional",
+        "OAI.InvariantIsing.random_pressure_limit_unconditional",
+        "OAI.InvariantIsing.random_pressure_tendsto_ae_unconditional",
+        "OAI.InvariantIsing.random_field_pressure_tendsto_ae_unconditional",
+        "OAI.InvariantIsing.gaussianPattern_pressure_limit_unconditional",
+        "OAI.InvariantIsing.gaussianPattern_ground_limit_unconditional",
+        "OAI.InvariantIsing.gaussianPattern_norm_limits_unconditional"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 2123,
+      "bytes": 10537287,
+      "packages": [],
+      "grade_timeout_sec": 20283,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 5030
     },
     {
       "task": "irrational-rotation",
@@ -3049,7 +3095,7 @@
       "packages": [],
       "grade_timeout_sec": 14080,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1441
+      "oracle_grade_seconds": 2128
     },
     {
       "task": "irrational-triangle-billiard",
@@ -3067,7 +3113,7 @@
       "packages": [],
       "grade_timeout_sec": 11334,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 820
+      "oracle_grade_seconds": 347
     },
     {
       "task": "ising-finiteness",
@@ -3085,7 +3131,7 @@
       "packages": [],
       "grade_timeout_sec": 12144,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 958
+      "oracle_grade_seconds": 327
     },
     {
       "task": "isometric-immersion",
@@ -3103,7 +3149,7 @@
       "packages": [],
       "grade_timeout_sec": 13342,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1242
+      "oracle_grade_seconds": 698
     },
     {
       "task": "kl-invariance",
@@ -3121,7 +3167,7 @@
       "packages": [],
       "grade_timeout_sec": 12448,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1717
+      "oracle_grade_seconds": 1231
     },
     {
       "task": "k-median-recovery",
@@ -3139,7 +3185,7 @@
       "packages": [],
       "grade_timeout_sec": 14148,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1362
+      "oracle_grade_seconds": 1418
     },
     {
       "task": "k-median-refined-recovery",
@@ -3157,7 +3203,7 @@
       "packages": [],
       "grade_timeout_sec": 13122,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1849
+      "oracle_grade_seconds": 1462
     },
     {
       "task": "k-median-threshold",
@@ -3175,7 +3221,7 @@
       "packages": [],
       "grade_timeout_sec": 13713,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2259
+      "oracle_grade_seconds": 2180
     },
     {
       "task": "k-server",
@@ -3203,7 +3249,7 @@
       "packages": [],
       "grade_timeout_sec": 12404,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 587
+      "oracle_grade_seconds": 322
     },
     {
       "task": "kadison-ringrose",
@@ -3221,7 +3267,7 @@
       "packages": [],
       "grade_timeout_sec": 13314,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1352
+      "oracle_grade_seconds": 1426
     },
     {
       "task": "kadison-similarity",
@@ -3241,7 +3287,7 @@
       "packages": [],
       "grade_timeout_sec": 16012,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 3415
+      "oracle_grade_seconds": 4111
     },
     {
       "task": "kahler-splitting",
@@ -3259,7 +3305,7 @@
       "packages": [],
       "grade_timeout_sec": 11454,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1239
+      "oracle_grade_seconds": 468
     },
     {
       "task": "kaplansky-direct-finiteness",
@@ -3277,7 +3323,7 @@
       "packages": [],
       "grade_timeout_sec": 11289,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1170
+      "oracle_grade_seconds": 704
     },
     {
       "task": "kaplansky-finitely-presented",
@@ -3295,7 +3341,7 @@
       "packages": [],
       "grade_timeout_sec": 11300,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1470
+      "oracle_grade_seconds": 723
     },
     {
       "task": "kaplansky-quasitrace",
@@ -3313,7 +3359,7 @@
       "packages": [],
       "grade_timeout_sec": 12150,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 993
+      "oracle_grade_seconds": 492
     },
     {
       "task": "kaplansky-stable-finiteness",
@@ -3331,7 +3377,7 @@
       "packages": [],
       "grade_timeout_sec": 12365,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1007
+      "oracle_grade_seconds": 720
     },
     {
       "task": "kervaire",
@@ -3349,7 +3395,7 @@
       "packages": [],
       "grade_timeout_sec": 11257,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1057
+      "oracle_grade_seconds": 506
     },
     {
       "task": "laughlin",
@@ -3367,7 +3413,7 @@
       "packages": [],
       "grade_timeout_sec": 13474,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 6718
+      "oracle_grade_seconds": 6435
     },
     {
       "task": "laughlin-fock",
@@ -3385,7 +3431,7 @@
       "packages": [],
       "grade_timeout_sec": 18377,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 6546
+      "oracle_grade_seconds": 6157
     },
     {
       "task": "laughlin-gap",
@@ -3403,7 +3449,7 @@
       "packages": [],
       "grade_timeout_sec": 12905,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 5587
+      "oracle_grade_seconds": 4594
     },
     {
       "task": "laughlin-planar",
@@ -3421,7 +3467,7 @@
       "packages": [],
       "grade_timeout_sec": 13554,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 7126
+      "oracle_grade_seconds": 5977
     },
     {
       "task": "lieb-thirring",
@@ -3439,7 +3485,7 @@
       "packages": [],
       "grade_timeout_sec": 11388,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1327
+      "oracle_grade_seconds": 478
     },
     {
       "task": "lipschitz-equivalence",
@@ -3457,7 +3503,7 @@
       "packages": [],
       "grade_timeout_sec": 11228,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1251
+      "oracle_grade_seconds": 460
     },
     {
       "task": "lipschitz-hilbert",
@@ -3476,7 +3522,7 @@
       "packages": [],
       "grade_timeout_sec": 19136,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2945
+      "oracle_grade_seconds": 2873
     },
     {
       "task": "list-hadwiger",
@@ -3494,7 +3540,7 @@
       "packages": [],
       "grade_timeout_sec": 11612,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 715
+      "oracle_grade_seconds": 402
     },
     {
       "task": "little-finitistic",
@@ -3512,7 +3558,7 @@
       "packages": [],
       "grade_timeout_sec": 12645,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1694
+      "oracle_grade_seconds": 1382
     },
     {
       "task": "littlewood-finite-flatness",
@@ -3530,7 +3576,7 @@
       "packages": [],
       "grade_timeout_sec": 11204,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 855
+      "oracle_grade_seconds": 316
     },
     {
       "task": "log-brunn-minkowski",
@@ -3548,7 +3594,7 @@
       "packages": [],
       "grade_timeout_sec": 11905,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 898
+      "oracle_grade_seconds": 432
     },
     {
       "task": "log-concave-query",
@@ -3566,7 +3612,7 @@
       "packages": [],
       "grade_timeout_sec": 13500,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1278
+      "oracle_grade_seconds": 1273
     },
     {
       "task": "log-kodaira-fiber-negative",
@@ -3584,7 +3630,7 @@
       "packages": [],
       "grade_timeout_sec": 10959,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 813
+      "oracle_grade_seconds": 296
     },
     {
       "task": "mahler-conjecture",
@@ -3602,7 +3648,7 @@
       "packages": [],
       "grade_timeout_sec": 11696,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 968
+      "oracle_grade_seconds": 514
     },
     {
       "task": "markov-type",
@@ -3620,7 +3666,7 @@
       "packages": [],
       "grade_timeout_sec": 11788,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 926
+      "oracle_grade_seconds": 277
     },
     {
       "task": "mass-action",
@@ -3638,7 +3684,7 @@
       "packages": [],
       "grade_timeout_sec": 11095,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 751
+      "oracle_grade_seconds": 255
     },
     {
       "task": "matching-affine-lift",
@@ -3656,7 +3702,7 @@
       "packages": [],
       "grade_timeout_sec": 11089,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 905
+      "oracle_grade_seconds": 380
     },
     {
       "task": "matching-entropy",
@@ -3674,7 +3720,7 @@
       "packages": [],
       "grade_timeout_sec": 10941,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 607
+      "oracle_grade_seconds": 156
     },
     {
       "task": "matching-entropy-bounds",
@@ -3692,7 +3738,7 @@
       "packages": [],
       "grade_timeout_sec": 11111,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 808
+      "oracle_grade_seconds": 363
     },
     {
       "task": "matching-fpras",
@@ -3710,7 +3756,7 @@
       "packages": [],
       "grade_timeout_sec": 13582,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1267
+      "oracle_grade_seconds": 1243
     },
     {
       "task": "matching-psd",
@@ -3728,7 +3774,7 @@
       "packages": [],
       "grade_timeout_sec": 11066,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 855
+      "oracle_grade_seconds": 382
     },
     {
       "task": "matrix-fields",
@@ -3746,7 +3792,7 @@
       "packages": [],
       "grade_timeout_sec": 21600,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 4564
+      "oracle_grade_seconds": 4284
     },
     {
       "task": "matrix-removal",
@@ -3764,7 +3810,7 @@
       "packages": [],
       "grade_timeout_sec": 11090,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 710
+      "oracle_grade_seconds": 215
     },
     {
       "task": "matroid-prophet",
@@ -3782,7 +3828,7 @@
       "packages": [],
       "grade_timeout_sec": 11399,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 745
+      "oracle_grade_seconds": 310
     },
     {
       "task": "matroid-secretary",
@@ -3802,7 +3848,7 @@
       "packages": [],
       "grade_timeout_sec": 11857,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 973
+      "oracle_grade_seconds": 478
     },
     {
       "task": "maximal-seshadri-constants",
@@ -3820,7 +3866,7 @@
       "packages": [],
       "grade_timeout_sec": 13044,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2446
+      "oracle_grade_seconds": 2150
     },
     {
       "task": "memory-precision",
@@ -3838,7 +3884,7 @@
       "packages": [],
       "grade_timeout_sec": 11502,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 957
+      "oracle_grade_seconds": 420
     },
     {
       "task": "metric-entropy-duality",
@@ -3856,7 +3902,7 @@
       "packages": [],
       "grade_timeout_sec": 11041,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 557
+      "oracle_grade_seconds": 77
     },
     {
       "task": "metric-k-median",
@@ -3874,7 +3920,7 @@
       "packages": [],
       "grade_timeout_sec": 13032,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1451
+      "oracle_grade_seconds": 1760
     },
     {
       "task": "midpoint-lenses",
@@ -3892,7 +3938,7 @@
       "packages": [],
       "grade_timeout_sec": 10916,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 634
+      "oracle_grade_seconds": 139
     },
     {
       "task": "min-uncut",
@@ -3910,7 +3956,7 @@
       "packages": [],
       "grade_timeout_sec": 16196,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1335
+      "oracle_grade_seconds": 1263
     },
     {
       "task": "nagata",
@@ -3930,7 +3976,7 @@
       "packages": [],
       "grade_timeout_sec": 12248,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 505
+      "oracle_grade_seconds": 296
     },
     {
       "task": "naimark",
@@ -3956,7 +4002,7 @@
       "packages": [],
       "grade_timeout_sec": 11887,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2857
+      "oracle_grade_seconds": 2390
     },
     {
       "task": "navier-stokes-alternating",
@@ -3974,7 +4020,7 @@
       "packages": [],
       "grade_timeout_sec": 11147,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 874
+      "oracle_grade_seconds": 471
     },
     {
       "task": "navier-stokes-velocity",
@@ -3992,7 +4038,7 @@
       "packages": [],
       "grade_timeout_sec": 11728,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1274
+      "oracle_grade_seconds": 925
     },
     {
       "task": "nodal-length",
@@ -4010,7 +4056,7 @@
       "packages": [],
       "grade_timeout_sec": 12135,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1261
+      "oracle_grade_seconds": 915
     },
     {
       "task": "noiseless-regression",
@@ -4028,7 +4074,7 @@
       "packages": [],
       "grade_timeout_sec": 10980,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 758
+      "oracle_grade_seconds": 225
     },
     {
       "task": "odd-kaplansky",
@@ -4046,7 +4092,7 @@
       "packages": [],
       "grade_timeout_sec": 11549,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1336
+      "oracle_grade_seconds": 901
     },
     {
       "task": "one-way-liveness",
@@ -4064,7 +4110,7 @@
       "packages": [],
       "grade_timeout_sec": 10955,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 624
+      "oracle_grade_seconds": 95
     },
     {
       "task": "optimal-max-cut",
@@ -4082,7 +4128,7 @@
       "packages": [],
       "grade_timeout_sec": 18776,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1488
+      "oracle_grade_seconds": 1492
     },
     {
       "task": "ordinary-elliott",
@@ -4100,7 +4146,7 @@
       "packages": [],
       "grade_timeout_sec": 13561,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1590
+      "oracle_grade_seconds": 2016
     },
     {
       "task": "parity-lifts",
@@ -4118,7 +4164,7 @@
       "packages": [],
       "grade_timeout_sec": 10962,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 839
+      "oracle_grade_seconds": 137
     },
     {
       "task": "partial-permutation",
@@ -4136,7 +4182,7 @@
       "packages": [],
       "grade_timeout_sec": 11021,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 802
+      "oracle_grade_seconds": 352
     },
     {
       "task": "partition-consistency",
@@ -4154,7 +4200,7 @@
       "packages": [],
       "grade_timeout_sec": 12151,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 823
+      "oracle_grade_seconds": 344
     },
     {
       "task": "partition-principle",
@@ -4172,7 +4218,7 @@
       "packages": [],
       "grade_timeout_sec": 21600,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2843
+      "oracle_grade_seconds": 2875
     },
     {
       "task": "perceptron-free-energy",
@@ -4190,7 +4236,7 @@
       "packages": [],
       "grade_timeout_sec": 14271,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1268
+      "oracle_grade_seconds": 1308
     },
     {
       "task": "perfect-completeness",
@@ -4208,7 +4254,7 @@
       "packages": [],
       "grade_timeout_sec": 21600,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2380
+      "oracle_grade_seconds": 2718
     },
     {
       "task": "periodic-group",
@@ -4226,7 +4272,7 @@
       "packages": [],
       "grade_timeout_sec": 12417,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1008
+      "oracle_grade_seconds": 528
     },
     {
       "task": "periodic-tiling-three",
@@ -4244,7 +4290,7 @@
       "packages": [],
       "grade_timeout_sec": 11395,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 928
+      "oracle_grade_seconds": 409
     },
     {
       "task": "permanent-cubic",
@@ -4262,7 +4308,7 @@
       "packages": [],
       "grade_timeout_sec": 11624,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1031
+      "oracle_grade_seconds": 591
     },
     {
       "task": "petty-projection-volume",
@@ -4280,7 +4326,7 @@
       "packages": [],
       "grade_timeout_sec": 11396,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 979
+      "oracle_grade_seconds": 459
     },
     {
       "task": "pi-exponent",
@@ -4298,7 +4344,7 @@
       "packages": [],
       "grade_timeout_sec": 14689,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 2882
+      "oracle_grade_seconds": 2846
     },
     {
       "task": "pinched-kahler",
@@ -4318,7 +4364,7 @@
       "packages": [],
       "grade_timeout_sec": 11629,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1351
+      "oracle_grade_seconds": 909
     },
     {
       "task": "pinned-distances",
@@ -4336,7 +4382,7 @@
       "packages": [],
       "grade_timeout_sec": 11465,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 926
+      "oracle_grade_seconds": 467
     },
     {
       "task": "planar-anderson-spectrum",
@@ -4354,7 +4400,7 @@
       "packages": [],
       "grade_timeout_sec": 10831,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 582
+      "oracle_grade_seconds": 96
     },
     {
       "task": "planar-falconer",
@@ -4372,7 +4418,7 @@
       "packages": [],
       "grade_timeout_sec": 13176,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1803
+      "oracle_grade_seconds": 1590
     },
     {
       "task": "planar-first-passage",
@@ -4390,7 +4436,7 @@
       "packages": [],
       "grade_timeout_sec": 11568,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 768
+      "oracle_grade_seconds": 462
     },
     {
       "task": "planar-l1",
@@ -4408,7 +4454,7 @@
       "packages": [],
       "grade_timeout_sec": 12210,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 713
+      "oracle_grade_seconds": 421
     },
     {
       "task": "planar-unit-distances",
@@ -4426,7 +4472,7 @@
       "packages": [],
       "grade_timeout_sec": 12010,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1123
+      "oracle_grade_seconds": 513
     },
     {
       "task": "plane-coloring",
@@ -4444,7 +4490,7 @@
       "packages": [],
       "grade_timeout_sec": 10816,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 677
+      "oracle_grade_seconds": 99
     },
     {
       "task": "posterior-replicas",
@@ -4462,7 +4508,7 @@
       "packages": [],
       "grade_timeout_sec": 11444,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1090
+      "oracle_grade_seconds": 621
     },
     {
       "task": "power-free-values",
@@ -4480,7 +4526,7 @@
       "packages": [],
       "grade_timeout_sec": 11824,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1054
+      "oracle_grade_seconds": 692
     },
     {
       "task": "prime-gaps",
@@ -4498,7 +4544,7 @@
       "packages": [],
       "grade_timeout_sec": 11777,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 727
+      "oracle_grade_seconds": 560
     },
     {
       "task": "product-brown",
@@ -4516,7 +4562,7 @@
       "packages": [],
       "grade_timeout_sec": 11878,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 853
+      "oracle_grade_seconds": 889
     },
     {
       "task": "projection-counterexample",
@@ -4534,7 +4580,7 @@
       "packages": [],
       "grade_timeout_sec": 10918,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 576
+      "oracle_grade_seconds": 142
     },
     {
       "task": "projection-moments",
@@ -4552,7 +4598,7 @@
       "packages": [],
       "grade_timeout_sec": 11263,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 919
+      "oracle_grade_seconds": 343
     },
     {
       "task": "projection-volume",
@@ -4570,7 +4616,7 @@
       "packages": [],
       "grade_timeout_sec": 10915,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 601
+      "oracle_grade_seconds": 182
     },
     {
       "task": "qac-parity",
@@ -4588,7 +4634,7 @@
       "packages": [],
       "grade_timeout_sec": 10937,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 693
+      "oracle_grade_seconds": 171
     },
     {
       "task": "quadric-bundles",
@@ -4606,7 +4652,7 @@
       "packages": [],
       "grade_timeout_sec": 11343,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 886
+      "oracle_grade_seconds": 408
     },
     {
       "task": "quantitative-van-der-waerden",
@@ -4624,7 +4670,7 @@
       "packages": [],
       "grade_timeout_sec": 11378,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 811
+      "oracle_grade_seconds": 170
     },
     {
       "task": "quintic-lienard",
@@ -4642,7 +4688,7 @@
       "packages": [],
       "grade_timeout_sec": 11534,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1484
+      "oracle_grade_seconds": 829
     },
     {
       "task": "radial-density-interval",
@@ -4660,7 +4706,7 @@
       "packages": [],
       "grade_timeout_sec": 11428,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1163
+      "oracle_grade_seconds": 678
     },
     {
       "task": "radial-transition",
@@ -4678,7 +4724,25 @@
       "packages": [],
       "grade_timeout_sec": 11423,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1222
+      "oracle_grade_seconds": 888
+    },
+    {
+      "task": "ramsey-five",
+      "challenge": "RamseyFive",
+      "family": "170",
+      "family_title": "Sharp logarithmic exponents for off-diagonal Ramsey numbers",
+      "subject": "Combinatorics",
+      "theorems": [
+        "OAI.SharpRamseyFive.main"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 482,
+      "bytes": 3634776,
+      "packages": [],
+      "grade_timeout_sec": 14071,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 2034
     },
     {
       "task": "randomized-mean-payoff",
@@ -4696,7 +4760,7 @@
       "packages": [],
       "grade_timeout_sec": 11970,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1082
+      "oracle_grade_seconds": 809
     },
     {
       "task": "rational-hitting",
@@ -4714,7 +4778,7 @@
       "packages": [],
       "grade_timeout_sec": 11430,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 666
+      "oracle_grade_seconds": 472
     },
     {
       "task": "real-l1-renorming",
@@ -4732,7 +4796,7 @@
       "packages": [],
       "grade_timeout_sec": 10961,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 304
+      "oracle_grade_seconds": 172
     },
     {
       "task": "recursive-potentials",
@@ -4750,7 +4814,7 @@
       "packages": [],
       "grade_timeout_sec": 11201,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 617
+      "oracle_grade_seconds": 218
     },
     {
       "task": "reflexive-fixed-points",
@@ -4768,7 +4832,7 @@
       "packages": [],
       "grade_timeout_sec": 11232,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 664
+      "oracle_grade_seconds": 221
     },
     {
       "task": "regular-parity",
@@ -4786,7 +4850,7 @@
       "packages": [],
       "grade_timeout_sec": 10938,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 623
+      "oracle_grade_seconds": 185
     },
     {
       "task": "relative-generation",
@@ -4804,7 +4868,7 @@
       "packages": [],
       "grade_timeout_sec": 12350,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 3956
+      "oracle_grade_seconds": 2953
     },
     {
       "task": "rokhlin",
@@ -4827,7 +4891,7 @@
       "packages": [],
       "grade_timeout_sec": 12066,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1397
+      "oracle_grade_seconds": 627
     },
     {
       "task": "ruled-covering",
@@ -4845,7 +4909,7 @@
       "packages": [],
       "grade_timeout_sec": 10976,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 963
+      "oracle_grade_seconds": 265
     },
     {
       "task": "ryser-covering",
@@ -4863,7 +4927,7 @@
       "packages": [],
       "grade_timeout_sec": 11816,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 702
+      "oracle_grade_seconds": 355
     },
     {
       "task": "ryser-odd-extensions",
@@ -4881,7 +4945,25 @@
       "packages": [],
       "grade_timeout_sec": 11826,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 649
+      "oracle_grade_seconds": 205
+    },
+    {
+      "task": "sat-computability",
+      "challenge": "SATComputability",
+      "family": "235",
+      "family_title": "Random-SAT thresholds, sharp variance and computability",
+      "subject": "Probability and statistical mechanics",
+      "theorems": [
+        "OAI.FixedClauseThreshold.Computability.main"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 592,
+      "bytes": 4186949,
+      "packages": [],
+      "grade_timeout_sec": 14568,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 1515
     },
     {
       "task": "sat-sharpness",
@@ -4899,7 +4981,7 @@
       "packages": [],
       "grade_timeout_sec": 11013,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 561
+      "oracle_grade_seconds": 167
     },
     {
       "task": "sat-variance",
@@ -4917,7 +4999,25 @@
       "packages": [],
       "grade_timeout_sec": 11069,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 700
+      "oracle_grade_seconds": 245
+    },
+    {
+      "task": "sk-barriers",
+      "challenge": "SKBarriers",
+      "family": "227",
+      "family_title": "The dynamical phase transition in the Sherrington\u2013Kirkpatrick model",
+      "subject": "Probability and statistical mechanics",
+      "theorems": [
+        "OAI.SK.main"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 574,
+      "bytes": 2610000,
+      "packages": [],
+      "grade_timeout_sec": 13149,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 1847
     },
     {
       "task": "sk-full-support",
@@ -4935,7 +5035,43 @@
       "packages": [],
       "grade_timeout_sec": 11690,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 964
+      "oracle_grade_seconds": 656
+    },
+    {
+      "task": "sk-high-temperature",
+      "challenge": "SKHighTemperature",
+      "family": "227",
+      "family_title": "The dynamical phase transition in the Sherrington\u2013Kirkpatrick model",
+      "subject": "Probability and statistical mechanics",
+      "theorems": [
+        "OAI.SKGap.sk_main"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 556,
+      "bytes": 4702297,
+      "packages": [],
+      "grade_timeout_sec": 15032,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 2268
+    },
+    {
+      "task": "sk-ratio",
+      "challenge": "SKRatio",
+      "family": "227",
+      "family_title": "The dynamical phase transition in the Sherrington\u2013Kirkpatrick model",
+      "subject": "Probability and statistical mechanics",
+      "theorems": [
+        "OAI.SKRatio.ratio_cutoff"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 260,
+      "bytes": 2087158,
+      "packages": [],
+      "grade_timeout_sec": 12678,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 7778
     },
     {
       "task": "sk-value",
@@ -4953,7 +5089,7 @@
       "packages": [],
       "grade_timeout_sec": 12139,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1378
+      "oracle_grade_seconds": 650
     },
     {
       "task": "sle-lower-positivity",
@@ -4971,7 +5107,7 @@
       "packages": [],
       "grade_timeout_sec": 12837,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1070
+      "oracle_grade_seconds": 527
     },
     {
       "task": "saxl",
@@ -4989,7 +5125,7 @@
       "packages": [],
       "grade_timeout_sec": 11089,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 943
+      "oracle_grade_seconds": 504
     },
     {
       "task": "second-kahn-kalai",
@@ -5007,7 +5143,7 @@
       "packages": [],
       "grade_timeout_sec": 11059,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 573
+      "oracle_grade_seconds": 76
     },
     {
       "task": "self-similar",
@@ -5025,7 +5161,7 @@
       "packages": [],
       "grade_timeout_sec": 11271,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 832
+      "oracle_grade_seconds": 382
     },
     {
       "task": "self-similar-corollaries",
@@ -5044,7 +5180,7 @@
       "packages": [],
       "grade_timeout_sec": 11303,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1090
+      "oracle_grade_seconds": 399
     },
     {
       "task": "sensitivity-separation",
@@ -5062,7 +5198,7 @@
       "packages": [],
       "grade_timeout_sec": 10943,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 787
+      "oracle_grade_seconds": 133
     },
     {
       "task": "separable-quotient-negative",
@@ -5080,7 +5216,7 @@
       "packages": [],
       "grade_timeout_sec": 11402,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 974
+      "oracle_grade_seconds": 545
     },
     {
       "task": "seymour-second-neighborhood",
@@ -5098,7 +5234,7 @@
       "packages": [],
       "grade_timeout_sec": 10971,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 649
+      "oracle_grade_seconds": 59
     },
     {
       "task": "sharp-cat0-filling",
@@ -5116,7 +5252,25 @@
       "packages": [],
       "grade_timeout_sec": 12896,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1201
+      "oracle_grade_seconds": 1023
+    },
+    {
+      "task": "sharp-log-ramsey",
+      "challenge": "SharpLogRamsey",
+      "family": "170",
+      "family_title": "Sharp logarithmic exponents for off-diagonal Ramsey numbers",
+      "subject": "Combinatorics",
+      "theorems": [
+        "OAI.SharpLogRamsey.main"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 408,
+      "bytes": 2902698,
+      "packages": [],
+      "grade_timeout_sec": 13412,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 1307
     },
     {
       "task": "sharp-threshold",
@@ -5134,7 +5288,7 @@
       "packages": [],
       "grade_timeout_sec": 11082,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 662
+      "oracle_grade_seconds": 221
     },
     {
       "task": "short-egyptian-fractions",
@@ -5152,7 +5306,7 @@
       "packages": [],
       "grade_timeout_sec": 11145,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 818
+      "oracle_grade_seconds": 357
     },
     {
       "task": "sidorenko-counterexample",
@@ -5170,7 +5324,7 @@
       "packages": [],
       "grade_timeout_sec": 11895,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1285
+      "oracle_grade_seconds": 837
     },
     {
       "task": "signed-finite-band",
@@ -5188,7 +5342,7 @@
       "packages": [],
       "grade_timeout_sec": 11557,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 785
+      "oracle_grade_seconds": 376
     },
     {
       "task": "signed-sweep-moment",
@@ -5206,7 +5360,43 @@
       "packages": [],
       "grade_timeout_sec": 11878,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1266
+      "oracle_grade_seconds": 796
+    },
+    {
+      "task": "simple-amenable",
+      "challenge": "SimpleAmenable",
+      "family": "253",
+      "family_title": "An infinite finitely presented simple amenable group",
+      "subject": "Group theory",
+      "theorems": [
+        "OAI.SimpleAmenable.main"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 550,
+      "bytes": 3943717,
+      "packages": [],
+      "grade_timeout_sec": 14349,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 1969
+    },
+    {
+      "task": "simple-overgroups",
+      "challenge": "SimpleOvergroups",
+      "family": "250",
+      "family_title": "The Boone\u2013Higman conjecture and higher finiteness",
+      "subject": "Group theory",
+      "theorems": [
+        "OAI.SimpleFInftyOvergroups.main"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 238,
+      "bytes": 4013491,
+      "packages": [],
+      "grade_timeout_sec": 14412,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 1221
     },
     {
       "task": "single-fold",
@@ -5224,7 +5414,7 @@
       "packages": [],
       "grade_timeout_sec": 11103,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 851
+      "oracle_grade_seconds": 344
     },
     {
       "task": "single-lattice-covering",
@@ -5242,7 +5432,7 @@
       "packages": [],
       "grade_timeout_sec": 11831,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1015
+      "oracle_grade_seconds": 519
     },
     {
       "task": "singleton-loop-matching",
@@ -5260,7 +5450,7 @@
       "packages": [],
       "grade_timeout_sec": 11447,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1239
+      "oracle_grade_seconds": 576
     },
     {
       "task": "smooth-initial-form",
@@ -5278,7 +5468,7 @@
       "packages": [],
       "grade_timeout_sec": 11061,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 852
+      "oracle_grade_seconds": 334
     },
     {
       "task": "smooth-obstruction",
@@ -5296,7 +5486,44 @@
       "packages": [],
       "grade_timeout_sec": 12186,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1158
+      "oracle_grade_seconds": 472
+    },
+    {
+      "task": "smooth-yau",
+      "challenge": "SmoothYau",
+      "family": "350",
+      "family_title": "Yau's nodal conjecture: surfaces and counterexamples",
+      "subject": "Differential geometry",
+      "theorems": [
+        "OAI.YauCounterexamples.sphere_three",
+        "OAI.YauCounterexamples.sphere_two_torus_two"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 436,
+      "bytes": 4544098,
+      "packages": [],
+      "grade_timeout_sec": 14889,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 1965
+    },
+    {
+      "task": "snaky-conditional",
+      "challenge": "SnakyConditional",
+      "family": "187",
+      "family_title": "Snaky in 21 Maker moves",
+      "subject": "Combinatorics",
+      "theorems": [
+        "OAI.SnakyConditional.snaky_winning_strategy_with_legal_states"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 209,
+      "bytes": 7279627,
+      "packages": [],
+      "grade_timeout_sec": 17351,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 3069
     },
     {
       "task": "soft-channel204",
@@ -5314,7 +5541,7 @@
       "packages": [],
       "grade_timeout_sec": 11246,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 775
+      "oracle_grade_seconds": 323
     },
     {
       "task": "solenoidal-sheet-programs",
@@ -5332,7 +5559,7 @@
       "packages": [],
       "grade_timeout_sec": 11065,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 690
+      "oracle_grade_seconds": 244
     },
     {
       "task": "spherical-field",
@@ -5350,7 +5577,7 @@
       "packages": [],
       "grade_timeout_sec": 11928,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1292
+      "oracle_grade_seconds": 698
     },
     {
       "task": "split-tangent-integrability",
@@ -5368,7 +5595,7 @@
       "packages": [],
       "grade_timeout_sec": 11524,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 529
+      "oracle_grade_seconds": 360
     },
     {
       "task": "square-difference",
@@ -5386,7 +5613,7 @@
       "packages": [],
       "grade_timeout_sec": 11792,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 978
+      "oracle_grade_seconds": 524
     },
     {
       "task": "square-root-degree",
@@ -5404,7 +5631,25 @@
       "packages": [],
       "grade_timeout_sec": 10942,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 316
+      "oracle_grade_seconds": 207
+    },
+    {
+      "task": "standard-map-components",
+      "challenge": "StandardMapComponents",
+      "family": "146",
+      "family_title": "Sinai's positive-entropy conjecture for the standard map",
+      "subject": "Dynamical systems and ergodic theory",
+      "theorems": [
+        "OAI.StandardMapEntropy.main_components"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 274,
+      "bytes": 2886579,
+      "packages": [],
+      "grade_timeout_sec": 13397,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 1312
     },
     {
       "task": "standard-map-entropy",
@@ -5422,7 +5667,7 @@
       "packages": [],
       "grade_timeout_sec": 11988,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1548
+      "oracle_grade_seconds": 1510
     },
     {
       "task": "standard-map-lyapunov",
@@ -5440,7 +5685,7 @@
       "packages": [],
       "grade_timeout_sec": 12045,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1561
+      "oracle_grade_seconds": 1019
     },
     {
       "task": "steinitz-bergstrom",
@@ -5458,7 +5703,7 @@
       "packages": [],
       "grade_timeout_sec": 11444,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 875
+      "oracle_grade_seconds": 440
     },
     {
       "task": "strict-means",
@@ -5476,7 +5721,25 @@
       "packages": [],
       "grade_timeout_sec": 11446,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1180
+      "oracle_grade_seconds": 845
+    },
+    {
+      "task": "strong-kadison-kastler",
+      "challenge": "StrongKadisonKastler",
+      "family": "289",
+      "family_title": "The strong Kadison\u2013Kastler conjecture",
+      "subject": "Operator algebras",
+      "theorems": [
+        "OAI.KadisonKastler.universal_strong_stability"
+      ],
+      "definition_holes": [],
+      "fixed_definitions": [],
+      "n_modules": 2652,
+      "bytes": 14567629,
+      "packages": [],
+      "grade_timeout_sec": 21600,
+      "oracle_result": "accepted",
+      "oracle_grade_seconds": 10799
     },
     {
       "task": "strong-thin-tree",
@@ -5494,7 +5757,7 @@
       "packages": [],
       "grade_timeout_sec": 11497,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 974
+      "oracle_grade_seconds": 487
     },
     {
       "task": "strongly-rayleigh-dpp",
@@ -5517,7 +5780,7 @@
       "packages": [],
       "grade_timeout_sec": 11261,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 889
+      "oracle_grade_seconds": 364
     },
     {
       "task": "structural-crouzeix",
@@ -5535,7 +5798,7 @@
       "packages": [],
       "grade_timeout_sec": 11428,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1228
+      "oracle_grade_seconds": 546
     },
     {
       "task": "subpolynomial-lp",
@@ -5553,7 +5816,7 @@
       "packages": [],
       "grade_timeout_sec": 11132,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 893
+      "oracle_grade_seconds": 381
     },
     {
       "task": "subsphere-current",
@@ -5571,7 +5834,7 @@
       "packages": [],
       "grade_timeout_sec": 11229,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 463
+      "oracle_grade_seconds": 258
     },
     {
       "task": "superstring",
@@ -5589,7 +5852,7 @@
       "packages": [],
       "grade_timeout_sec": 11646,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 784
+      "oracle_grade_seconds": 297
     },
     {
       "task": "switch-chain",
@@ -5609,7 +5872,7 @@
       "packages": [],
       "grade_timeout_sec": 11629,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 924
+      "oracle_grade_seconds": 413
     },
     {
       "task": "symmetric-domains",
@@ -5627,7 +5890,7 @@
       "packages": [],
       "grade_timeout_sec": 12703,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1307
+      "oracle_grade_seconds": 1240
     },
     {
       "task": "symmetric-mahler-equality",
@@ -5645,7 +5908,7 @@
       "packages": [],
       "grade_timeout_sec": 11997,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 986
+      "oracle_grade_seconds": 662
     },
     {
       "task": "symmetric-polar",
@@ -5663,7 +5926,7 @@
       "packages": [],
       "grade_timeout_sec": 11289,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1068
+      "oracle_grade_seconds": 686
     },
     {
       "task": "tachikawa",
@@ -5681,7 +5944,7 @@
       "packages": [],
       "grade_timeout_sec": 11938,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1382
+      "oracle_grade_seconds": 974
     },
     {
       "task": "talagrand-discrete-convexity",
@@ -5699,7 +5962,7 @@
       "packages": [],
       "grade_timeout_sec": 10956,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 439
+      "oracle_grade_seconds": 72
     },
     {
       "task": "talagrand-expectation-threshold",
@@ -5717,7 +5980,7 @@
       "packages": [],
       "grade_timeout_sec": 10973,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 213
+      "oracle_grade_seconds": 64
     },
     {
       "task": "thomason-model-structures",
@@ -5736,7 +5999,7 @@
       "packages": [],
       "grade_timeout_sec": 13387,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1608
+      "oracle_grade_seconds": 599
     },
     {
       "task": "thompson-nonamenability",
@@ -5754,7 +6017,7 @@
       "packages": [],
       "grade_timeout_sec": 11135,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 146
+      "oracle_grade_seconds": 141
     },
     {
       "task": "thorp-first-reciprocal",
@@ -5772,7 +6035,7 @@
       "packages": [],
       "grade_timeout_sec": 11130,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 637
+      "oracle_grade_seconds": 344
     },
     {
       "task": "thorp-remaining",
@@ -5790,7 +6053,7 @@
       "packages": [],
       "grade_timeout_sec": 12040,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 730
+      "oracle_grade_seconds": 472
     },
     {
       "task": "thorp-routing",
@@ -5808,7 +6071,7 @@
       "packages": [],
       "grade_timeout_sec": 16723,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 1374
+      "oracle_grade_seconds": 1031
     },
     {
       "task": "thorp-weighted-compatibility",
@@ -5826,7 +6089,7 @@
       "packages": [],
       "grade_timeout_sec": 10921,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 294
+      "oracle_grade_seconds": 171
     },
     {
       "task": "three-machine",
@@ -5844,7 +6107,7 @@
       "packages": [],
       "grade_timeout_sec": 11566,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 467
+      "oracle_grade_seconds": 338
     },
     {
       "task": "three-state-supercritical",
@@ -5863,7 +6126,7 @@
       "packages": [],
       "grade_timeout_sec": 11144,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 212
+      "oracle_grade_seconds": 180
     },
     {
       "task": "three-state-tree-clauses",
@@ -5882,7 +6145,7 @@
       "packages": [],
       "grade_timeout_sec": 11145,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 204
+      "oracle_grade_seconds": 328
     },
     {
       "task": "three-torus",
@@ -5900,7 +6163,7 @@
       "packages": [],
       "grade_timeout_sec": 12348,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 669
+      "oracle_grade_seconds": 645
     },
     {
       "task": "tingley-sphere-isometry",
@@ -5918,7 +6181,7 @@
       "packages": [],
       "grade_timeout_sec": 11144,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 129
+      "oracle_grade_seconds": 94
     },
     {
       "task": "torsion-free-hyperbolic",
@@ -5936,7 +6199,7 @@
       "packages": [],
       "grade_timeout_sec": 11233,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 289
+      "oracle_grade_seconds": 216
     },
     {
       "task": "torsion-free-zero-divisors",
@@ -5954,7 +6217,7 @@
       "packages": [],
       "grade_timeout_sec": 12657,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 805
+      "oracle_grade_seconds": 560
     },
     {
       "task": "trace-reconstruction",
@@ -5974,7 +6237,7 @@
       "packages": [],
       "grade_timeout_sec": 12269,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 415
+      "oracle_grade_seconds": 316
     },
     {
       "task": "tree-edit",
@@ -5992,7 +6255,7 @@
       "packages": [],
       "grade_timeout_sec": 11018,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 202
+      "oracle_grade_seconds": 167
     },
     {
       "task": "triangle-face",
@@ -6010,7 +6273,7 @@
       "packages": [],
       "grade_timeout_sec": 11093,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 216
+      "oracle_grade_seconds": 174
     },
     {
       "task": "triangle-removal",
@@ -6028,7 +6291,7 @@
       "packages": [],
       "grade_timeout_sec": 12122,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 935
+      "oracle_grade_seconds": 986
     },
     {
       "task": "triangular-covering",
@@ -6046,7 +6309,7 @@
       "packages": [],
       "grade_timeout_sec": 10886,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 203
+      "oracle_grade_seconds": 140
     },
     {
       "task": "triangular-energy",
@@ -6100,7 +6363,7 @@
       "packages": [],
       "grade_timeout_sec": 11249,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 630
+      "oracle_grade_seconds": 621
     },
     {
       "task": "truffet-counterexample",
@@ -6118,7 +6381,7 @@
       "packages": [],
       "grade_timeout_sec": 10809,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 68
+      "oracle_grade_seconds": 46
     },
     {
       "task": "two-way-complementation",
@@ -6136,7 +6399,7 @@
       "packages": [],
       "grade_timeout_sec": 10973,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 96
+      "oracle_grade_seconds": 106
     },
     {
       "task": "two-way-determinization",
@@ -6154,7 +6417,7 @@
       "packages": [],
       "grade_timeout_sec": 11140,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 197
+      "oracle_grade_seconds": 137
     },
     {
       "task": "type-system-normalization",
@@ -6172,7 +6435,7 @@
       "packages": [],
       "grade_timeout_sec": 12221,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 186
+      "oracle_grade_seconds": 187
     },
     {
       "task": "uniform-commutator",
@@ -6208,7 +6471,7 @@
       "packages": [],
       "grade_timeout_sec": 11462,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 520
+      "oracle_grade_seconds": 534
     },
     {
       "task": "uniform-k-server",
@@ -6244,7 +6507,7 @@
       "packages": [],
       "grade_timeout_sec": 11239,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 470
+      "oracle_grade_seconds": 392
     },
     {
       "task": "unique-games-theorem",
@@ -6280,7 +6543,7 @@
       "packages": [],
       "grade_timeout_sec": 11595,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 416
+      "oracle_grade_seconds": 502
     },
     {
       "task": "variable-wl",
@@ -6298,7 +6561,7 @@
       "packages": [],
       "grade_timeout_sec": 11864,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 429
+      "oracle_grade_seconds": 456
     },
     {
       "task": "velocity-hemisphere",
@@ -6334,7 +6597,7 @@
       "packages": [],
       "grade_timeout_sec": 12071,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 696
+      "oracle_grade_seconds": 725
     },
     {
       "task": "vertex-cover",
@@ -6389,7 +6652,7 @@
       "packages": [],
       "grade_timeout_sec": 11488,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 529
+      "oracle_grade_seconds": 358
     },
     {
       "task": "weak-hessian",
@@ -6443,7 +6706,7 @@
       "packages": [],
       "grade_timeout_sec": 11533,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 338
+      "oracle_grade_seconds": 329
     },
     {
       "task": "weisfeiler-leman",
@@ -6461,7 +6724,7 @@
       "packages": [],
       "grade_timeout_sec": 11626,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 401
+      "oracle_grade_seconds": 301
     },
     {
       "task": "witnessed-choice",
@@ -6479,7 +6742,7 @@
       "packages": [],
       "grade_timeout_sec": 11806,
       "oracle_result": "accepted",
-      "oracle_grade_seconds": 585
+      "oracle_grade_seconds": 485
     },
     {
       "task": "yau-counterexample",
@@ -6513,12 +6776,6 @@
       "family": "347",
       "reason": "memory"
     },
-    {
-      "task": "atomic-gaussian",
-      "challenge": "AtomicGaussian",
-      "family": "090",
-      "reason": "pending"
-    },
     {
       "task": "catalan",
       "challenge": "Catalan",
@@ -6535,13 +6792,13 @@
       "task": "continuum-coulomb-hardness",
       "challenge": "ContinuumCoulombHardness",
       "family": "275",
-      "reason": "disk"
+      "reason": "memory"
     },
     {
       "task": "cycle-clique-ramsey",
       "challenge": "CycleCliqueRamsey",
       "family": "189",
-      "reason": "pending"
+      "reason": "memory"
     },
     {
       "task": "dirichlet-seven-eighths",
@@ -6565,7 +6822,7 @@
       "task": "erdos-reciprocal",
       "challenge": "ErdosReciprocal",
       "family": "159",
-      "reason": "disk"
+      "reason": "not completed"
     },
     {
       "task": "hecke-seven-eighths",
@@ -6573,12 +6830,6 @@
       "family": "003",
       "reason": "missing-package"
     },
-    {
-      "task": "invariant-ising",
-      "challenge": "InvariantIsing",
-      "family": "234",
-      "reason": "pending"
-    },
     {
       "task": "jacobsthal",
       "challenge": "Jacobsthal",
@@ -6601,13 +6852,13 @@
       "task": "koebe-circle-domains",
       "challenge": "KoebeCircleDomains",
       "family": "071",
-      "reason": "disk"
+      "reason": "missing-package"
     },
     {
       "task": "mub-six",
       "challenge": "MUBSix",
       "family": "266",
-      "reason": "pending"
+      "reason": "not completed"
     },
     {
       "task": "matrix-multiplication",
@@ -6637,7 +6888,7 @@
       "task": "ostmann-complete",
       "challenge": "OstmannComplete",
       "family": "013",
-      "reason": "pending"
+      "reason": "missing-package"
     },
     {
       "task": "ostmann-primes",
@@ -6649,19 +6900,19 @@
       "task": "patterson-first-moment",
       "challenge": "PattersonFirstMoment",
       "family": "023",
-      "reason": "disk"
+      "reason": "missing-package"
     },
     {
       "task": "planar-packing",
       "challenge": "PlanarPacking",
       "family": "090",
-      "reason": "pending"
+      "reason": "memory"
     },
     {
       "task": "polycyclic-recognition",
       "challenge": "PolycyclicRecognition",
       "family": "255",
-      "reason": "pending"
+      "reason": "missing-package"
     },
     {
       "task": "quasi-riemann-hypothesis",
@@ -6669,89 +6920,29 @@
       "family": "003",
       "reason": "missing-package"
     },
-    {
-      "task": "ramsey-five",
-      "challenge": "RamseyFive",
-      "family": "170",
-      "reason": "pending"
-    },
     {
       "task": "riesz-quantitative",
       "challenge": "RieszQuantitative",
       "family": "081",
       "reason": "missing-package"
     },
-    {
-      "task": "sat-computability",
-      "challenge": "SATComputability",
-      "family": "235",
-      "reason": "pending"
-    },
-    {
-      "task": "sk-barriers",
-      "challenge": "SKBarriers",
-      "family": "227",
-      "reason": "pending"
-    },
-    {
-      "task": "sk-high-temperature",
-      "challenge": "SKHighTemperature",
-      "family": "227",
-      "reason": "pending"
-    },
-    {
-      "task": "sk-ratio",
-      "challenge": "SKRatio",
-      "family": "227",
-      "reason": "pending"
-    },
-    {
-      "task": "sharp-log-ramsey",
-      "challenge": "SharpLogRamsey",
-      "family": "170",
-      "reason": "pending"
-    },
     {
       "task": "siegel-zeros",
       "challenge": "SiegelZeros",
       "family": "003",
       "reason": "missing-package"
     },
-    {
-      "task": "simple-amenable",
-      "challenge": "SimpleAmenable",
-      "family": "253",
-      "reason": "pending"
-    },
-    {
-      "task": "simple-overgroups",
-      "challenge": "SimpleOvergroups",
-      "family": "250",
-      "reason": "pending"
-    },
-    {
-      "task": "smooth-yau",
-      "challenge": "SmoothYau",
-      "family": "350",
-      "reason": "pending"
-    },
     {
       "task": "snaky-certificate",
       "challenge": "SnakyCertificate",
       "family": "187",
       "reason": "memory"
     },
-    {
-      "task": "snaky-conditional",
-      "challenge": "SnakyConditional",
-      "family": "187",
-      "reason": "pending"
-    },
     {
       "task": "snaky-twenty-one",
       "challenge": "SnakyTwentyOne",
       "family": "187",
-      "reason": "pending"
+      "reason": "memory"
     },
     {
       "task": "spin-angle",
@@ -6759,29 +6950,17 @@
       "family": "238",
       "reason": "definition-mismatch"
     },
-    {
-      "task": "standard-map-components",
-      "challenge": "StandardMapComponents",
-      "family": "146",
-      "reason": "pending"
-    },
-    {
-      "task": "strong-kadison-kastler",
-      "challenge": "StrongKadisonKastler",
-      "family": "289",
-      "reason": "pending"
-    },
     {
       "task": "surface-immersion",
       "challenge": "SurfaceImmersion",
       "family": "333",
-      "reason": "pending"
+      "reason": "missing-package"
     },
     {
       "task": "taming-compatibility",
       "challenge": "TamingCompatibility",
       "family": "342",
-      "reason": "pending"
+      "reason": "memory"
     },
     {
       "task": "totient-asymptotic",
diff --git a/manifest.parquet b/manifest.parquet
index 7cab14642b886b502548d65e115b94623c02ad01..dd61f3ac92284af1a4b81704e50f5604775b42fb 100644
--- a/manifest.parquet
+++ b/manifest.parquet
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3afe9e43520364a939417dd85aeec87f892c6c81fe54c6f9d0bb67650c4021d0
-size 44329
+oid sha256:5e4b47df11e93d77c06b8899337e1a0043f1193c9c5b18b19f0eecc9ee3acf94
+size 45681
diff --git a/registry.json b/registry.json
index e3fa7bb98556660a648e5e3b594af1e1c2fc39e2..f8414fdfa289f7fee283cc0590f9ced5baca00f5 100644
--- a/registry.json
+++ b/registry.json
@@ -2,7 +2,7 @@
   {
     "name": "openai-math",
     "version": "1.0",
-    "description": "355 Lean 4 theorem-proving tasks from openai/math, graded by Comparator in a separate sandbox.",
+    "description": "369 Lean 4 theorem-proving tasks from openai/math, graded by Comparator in a separate sandbox.",
     "tasks": [
       {
         "name": "abhyankar-sathaye",
@@ -32,6 +32,10 @@
         "name": "asymptotically-minimal-littlewood",
         "path": "tasks/asymptotically-minimal-littlewood"
       },
+      {
+        "name": "atomic-gaussian",
+        "path": "tasks/atomic-gaussian"
+      },
       {
         "name": "auslander-reiten",
         "path": "tasks/auslander-reiten"
@@ -664,6 +668,10 @@
         "name": "interpolated-factors",
         "path": "tasks/interpolated-factors"
       },
+      {
+        "name": "invariant-ising",
+        "path": "tasks/invariant-ising"
+      },
       {
         "name": "irrational-rotation",
         "path": "tasks/irrational-rotation"
@@ -1024,6 +1032,10 @@
         "name": "radial-transition",
         "path": "tasks/radial-transition"
       },
+      {
+        "name": "ramsey-five",
+        "path": "tasks/ramsey-five"
+      },
       {
         "name": "randomized-mean-payoff",
         "path": "tasks/randomized-mean-payoff"
@@ -1068,6 +1080,10 @@
         "name": "ryser-odd-extensions",
         "path": "tasks/ryser-odd-extensions"
       },
+      {
+        "name": "sat-computability",
+        "path": "tasks/sat-computability"
+      },
       {
         "name": "sat-sharpness",
         "path": "tasks/sat-sharpness"
@@ -1076,10 +1092,22 @@
         "name": "sat-variance",
         "path": "tasks/sat-variance"
       },
+      {
+        "name": "sk-barriers",
+        "path": "tasks/sk-barriers"
+      },
       {
         "name": "sk-full-support",
         "path": "tasks/sk-full-support"
       },
+      {
+        "name": "sk-high-temperature",
+        "path": "tasks/sk-high-temperature"
+      },
+      {
+        "name": "sk-ratio",
+        "path": "tasks/sk-ratio"
+      },
       {
         "name": "sk-value",
         "path": "tasks/sk-value"
@@ -1120,6 +1148,10 @@
         "name": "sharp-cat0-filling",
         "path": "tasks/sharp-cat0-filling"
       },
+      {
+        "name": "sharp-log-ramsey",
+        "path": "tasks/sharp-log-ramsey"
+      },
       {
         "name": "sharp-threshold",
         "path": "tasks/sharp-threshold"
@@ -1140,6 +1172,14 @@
         "name": "signed-sweep-moment",
         "path": "tasks/signed-sweep-moment"
       },
+      {
+        "name": "simple-amenable",
+        "path": "tasks/simple-amenable"
+      },
+      {
+        "name": "simple-overgroups",
+        "path": "tasks/simple-overgroups"
+      },
       {
         "name": "single-fold",
         "path": "tasks/single-fold"
@@ -1160,6 +1200,14 @@
         "name": "smooth-obstruction",
         "path": "tasks/smooth-obstruction"
       },
+      {
+        "name": "smooth-yau",
+        "path": "tasks/smooth-yau"
+      },
+      {
+        "name": "snaky-conditional",
+        "path": "tasks/snaky-conditional"
+      },
       {
         "name": "soft-channel204",
         "path": "tasks/soft-channel204"
@@ -1184,6 +1232,10 @@
         "name": "square-root-degree",
         "path": "tasks/square-root-degree"
       },
+      {
+        "name": "standard-map-components",
+        "path": "tasks/standard-map-components"
+      },
       {
         "name": "standard-map-entropy",
         "path": "tasks/standard-map-entropy"
@@ -1200,6 +1252,10 @@
         "name": "strict-means",
         "path": "tasks/strict-means"
       },
+      {
+        "name": "strong-kadison-kastler",
+        "path": "tasks/strong-kadison-kastler"
+      },
       {
         "name": "strong-thin-tree",
         "path": "tasks/strong-thin-tree"
diff --git a/tasks/abhyankar-sathaye/task.toml b/tasks/abhyankar-sathaye/task.toml
index 3b68e62f4686ddedaf29faae3aae57789ba0212d..785f32ed6dc7c1099a4be509f575e70badb90478 100644
--- a/tasks/abhyankar-sathaye/task.toml
+++ b/tasks/abhyankar-sathaye/task.toml
@@ -23,7 +23,7 @@ reference_modules = 15
 reference_lean_bytes = 55824
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 463
+oracle_grade_seconds = 47
 difficulty = "research"
 
 [agent]
diff --git a/tasks/abhyankar-sathaye/tests/grade.py b/tasks/abhyankar-sathaye/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/abhyankar-sathaye/tests/grade.py
+++ b/tasks/abhyankar-sathaye/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/affine-bernstein/task.toml b/tasks/affine-bernstein/task.toml
index 5295f6c1f1fe7389d5f53910a8cc8556abe438c8..5e61893faf32f66352cc85db9cf457f9a7a61f17 100644
--- a/tasks/affine-bernstein/task.toml
+++ b/tasks/affine-bernstein/task.toml
@@ -23,7 +23,7 @@ reference_modules = 322
 reference_lean_bytes = 1998143
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 1091
+oracle_grade_seconds = 1429
 difficulty = "research"
 
 [agent]
diff --git a/tasks/affine-bernstein/tests/grade.py b/tasks/affine-bernstein/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/affine-bernstein/tests/grade.py
+++ b/tasks/affine-bernstein/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/algorithmic-thin-trees/task.toml b/tasks/algorithmic-thin-trees/task.toml
index fa5b93b5b749e59beb5be3a2de94d6f8e46ab513..adc6d53158ab595ce2197c519d41d2a6a3d83b11 100644
--- a/tasks/algorithmic-thin-trees/task.toml
+++ b/tasks/algorithmic-thin-trees/task.toml
@@ -23,7 +23,7 @@ reference_modules = 736
 reference_lean_bytes = 4556988
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 2906
+oracle_grade_seconds = 2392
 difficulty = "research"
 
 [agent]
diff --git a/tasks/algorithmic-thin-trees/tests/grade.py b/tasks/algorithmic-thin-trees/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/algorithmic-thin-trees/tests/grade.py
+++ b/tasks/algorithmic-thin-trees/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/amplitude-damping/task.toml b/tasks/amplitude-damping/task.toml
index 7bc55f1ee5b2ed7419a34a684089e374aa4cb420..90874f7ccba3e07a331485887ad9d6e085379ac5 100644
--- a/tasks/amplitude-damping/task.toml
+++ b/tasks/amplitude-damping/task.toml
@@ -23,7 +23,7 @@ reference_modules = 35
 reference_lean_bytes = 283452
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 1119
+oracle_grade_seconds = 432
 difficulty = "research"
 
 [agent]
diff --git a/tasks/amplitude-damping/tests/grade.py b/tasks/amplitude-damping/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/amplitude-damping/tests/grade.py
+++ b/tasks/amplitude-damping/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/artin-cat0/task.toml b/tasks/artin-cat0/task.toml
index da64eb5d41d785287aafbfa77dca565ff591ee1f..e6e7b8f51af2625df8bdcfb23b0c214f408513d5 100644
--- a/tasks/artin-cat0/task.toml
+++ b/tasks/artin-cat0/task.toml
@@ -23,7 +23,7 @@ reference_modules = 20
 reference_lean_bytes = 414894
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 792
+oracle_grade_seconds = 193
 difficulty = "research"
 
 [agent]
diff --git a/tasks/artin-cat0/tests/grade.py b/tasks/artin-cat0/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/artin-cat0/tests/grade.py
+++ b/tasks/artin-cat0/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/artin-parabolic-intersections/task.toml b/tasks/artin-parabolic-intersections/task.toml
index 15a0fed7c7d27654a8ecb8042a6fab98e43ec13b..4d1a1b67986c1baf6064720b653f8fe0500b672c 100644
--- a/tasks/artin-parabolic-intersections/task.toml
+++ b/tasks/artin-parabolic-intersections/task.toml
@@ -23,7 +23,7 @@ reference_modules = 782
 reference_lean_bytes = 3300665
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 1711
+oracle_grade_seconds = 2200
 difficulty = "research"
 
 [agent]
diff --git a/tasks/artin-parabolic-intersections/tests/grade.py b/tasks/artin-parabolic-intersections/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/artin-parabolic-intersections/tests/grade.py
+++ b/tasks/artin-parabolic-intersections/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/asymptotically-minimal-littlewood/task.toml b/tasks/asymptotically-minimal-littlewood/task.toml
index 1be341ffaa9f39da7ebd4db8914a55e3b25b19c3..5c07789aec972fdec6219576ce0cc30a4377ed6a 100644
--- a/tasks/asymptotically-minimal-littlewood/task.toml
+++ b/tasks/asymptotically-minimal-littlewood/task.toml
@@ -23,7 +23,7 @@ reference_modules = 24
 reference_lean_bytes = 432620
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 919
+oracle_grade_seconds = 346
 difficulty = "research"
 
 [agent]
diff --git a/tasks/asymptotically-minimal-littlewood/tests/grade.py b/tasks/asymptotically-minimal-littlewood/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/asymptotically-minimal-littlewood/tests/grade.py
+++ b/tasks/asymptotically-minimal-littlewood/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/atomic-gaussian/environment/Dockerfile b/tasks/atomic-gaussian/environment/Dockerfile
new file mode 100644
index 0000000000000000000000000000000000000000..aa57529d28879aaf8f823eaf61743b87a7bb8310
--- /dev/null
+++ b/tasks/atomic-gaussian/environment/Dockerfile
@@ -0,0 +1,8 @@
+# Shared by every openai/math task (agent and separate verifier). See lean_setup.sh.
+FROM ubuntu:24.04
+ENV ELAN_HOME=/opt/elan \
+    PATH=/opt/elan/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin \
+    LANG=C.UTF-8
+COPY lean_setup.sh fetch_mathlib.sh lakefile.lean lake-manifest.json pack_submission.py /opt/build/
+RUN bash /opt/build/lean_setup.sh
+WORKDIR /workspace
diff --git a/tasks/atomic-gaussian/environment/fetch_mathlib.sh b/tasks/atomic-gaussian/environment/fetch_mathlib.sh
new file mode 100644
index 0000000000000000000000000000000000000000..1cc7232e51d27e7b0d9e0b136da9632d889f9c55
--- /dev/null
+++ b/tasks/atomic-gaussian/environment/fetch_mathlib.sh
@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+# Unpack Mathlib's prebuilt cache in batches and truncate the files that importing Mathlib never
+# reads (generated C and .ilean editor indexes), keeping their .hash files so Lake still sees every
+# module as built. Lean + full Mathlib otherwise needs ~11 GB, more than a 10 GB sandbox disk.
+set -euo pipefail
+cd /opt/lean-env
+trim() {
+  find .lake/packages -path '*/.lake/build/*' \( -name '*.c' -o -name '*.ilean' \) -size +0 -exec truncate -s 0 {} +
+  rm -rf "${HOME:-/root}/.cache/mathlib"
+}
+for dir in .lake/packages/mathlib/Mathlib/*/; do
+  name=$(basename "$dir")
+  echo "== Mathlib/$name"
+  lake exe cache get ".lake/packages/mathlib/Mathlib/$name" >/dev/null
+  trim
+  df -h / | tail -1
+done
+lake exe cache get
+trim
diff --git a/tasks/atomic-gaussian/environment/lake-manifest.json b/tasks/atomic-gaussian/environment/lake-manifest.json
new file mode 100644
index 0000000000000000000000000000000000000000..99b5eaa62153fd75dd2a9b1abdfac3738e914c88
--- /dev/null
+++ b/tasks/atomic-gaussian/environment/lake-manifest.json
@@ -0,0 +1,116 @@
+{
+ "version": "1.2.0",
+ "packagesDir": ".lake/packages",
+ "packages": [
+  {
+   "url": "https://github.com/leanprover-community/mathlib4.git",
+   "type": "git",
+   "subDir": null,
+   "scope": "",
+   "rev": "d13f23b723b8a846827a245b89c10fc7d3f11612",
+   "name": "mathlib",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "d13f23b723b8a846827a245b89c10fc7d3f11612",
+   "inherited": false,
+   "configFile": "lakefile.lean"
+  },
+  {
+   "url": "https://github.com/leanprover-community/plausible",
+   "type": "git",
+   "subDir": null,
+   "scope": "leanprover-community",
+   "rev": "118aa17ee84656b8bd727fef7c458ee8c833385c",
+   "name": "plausible",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "main",
+   "inherited": true,
+   "configFile": "lakefile.toml"
+  },
+  {
+   "url": "https://github.com/leanprover-community/LeanSearchClient",
+   "type": "git",
+   "subDir": null,
+   "scope": "leanprover-community",
+   "rev": "ddf04cf3949fa556442341e87d47f9f6e6074707",
+   "name": "LeanSearchClient",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "main",
+   "inherited": true,
+   "configFile": "lakefile.toml"
+  },
+  {
+   "url": "https://github.com/leanprover-community/import-graph",
+   "type": "git",
+   "subDir": null,
+   "scope": "leanprover-community",
+   "rev": "e928b72544873815af278d38681b31c0293588e3",
+   "name": "importGraph",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "main",
+   "inherited": true,
+   "configFile": "lakefile.toml"
+  },
+  {
+   "url": "https://github.com/leanprover-community/ProofWidgets4",
+   "type": "git",
+   "subDir": null,
+   "scope": "leanprover-community",
+   "rev": "106ff4fafc74ef4ac99d81dbf3ab399118f497a5",
+   "name": "proofwidgets",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "main",
+   "inherited": true,
+   "configFile": "lakefile.lean"
+  },
+  {
+   "url": "https://github.com/leanprover-community/aesop",
+   "type": "git",
+   "subDir": null,
+   "scope": "leanprover-community",
+   "rev": "355695d523e41d0554926416cba2a2b3544fbbc9",
+   "name": "aesop",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "master",
+   "inherited": true,
+   "configFile": "lakefile.toml"
+  },
+  {
+   "url": "https://github.com/leanprover-community/quote4",
+   "type": "git",
+   "subDir": null,
+   "scope": "leanprover-community",
+   "rev": "6a489d9af5d0c47e5b259e2e8bcdfc1811b5a259",
+   "name": "Qq",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "master",
+   "inherited": true,
+   "configFile": "lakefile.toml"
+  },
+  {
+   "url": "https://github.com/leanprover-community/batteries",
+   "type": "git",
+   "subDir": null,
+   "scope": "leanprover-community",
+   "rev": "f2effa3d803fda822b1f97b806c47cf2adfbcbc2",
+   "name": "batteries",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "main",
+   "inherited": true,
+   "configFile": "lakefile.toml"
+  },
+  {
+   "url": "https://github.com/leanprover/lean4-cli",
+   "type": "git",
+   "subDir": null,
+   "scope": "leanprover",
+   "rev": "e92c9f15fdfacc8536f31cfb3b7ad26c3c8cd204",
+   "name": "Cli",
+   "manifestFile": "lake-manifest.json",
+   "inputRev": "v4.34.0",
+   "inherited": true,
+   "configFile": "lakefile.toml"
+  }
+ ],
+ "name": "openai_math",
+ "lakeDir": ".lake"
+}
\ No newline at end of file
diff --git a/tasks/atomic-gaussian/environment/lakefile.lean b/tasks/atomic-gaussian/environment/lakefile.lean
new file mode 100644
index 0000000000000000000000000000000000000000..0330e313d8613d18833b17d46138a44a23269a68
--- /dev/null
+++ b/tasks/atomic-gaussian/environment/lakefile.lean
@@ -0,0 +1,12 @@
+import Lake
+open Lake DSL
+
+-- Mathlib at the commit openai/math pins. Agents write `Submission.lean` and `Submission/**.lean`.
+package openai_math where
+  leanOptions := #[⟨`autoImplicit, false⟩]
+
+require mathlib from git
+  "https://github.com/leanprover-community/mathlib4.git" @ "d13f23b723b8a846827a245b89c10fc7d3f11612"
+
+-- `Submission` and every `Submission.*` module it imports.
+lean_lib Submission
diff --git a/tasks/atomic-gaussian/environment/lean_setup.sh b/tasks/atomic-gaussian/environment/lean_setup.sh
new file mode 100644
index 0000000000000000000000000000000000000000..5a6a070b2935a67d97d15a2b4bde73baea197533
--- /dev/null
+++ b/tasks/atomic-gaussian/environment/lean_setup.sh
@@ -0,0 +1,126 @@
+#!/usr/bin/env bash
+# Builds the image shared by every openai/math task, for both the agent and the separate verifier:
+#   - Lean v4.34.1 and Mathlib d13f23b, the versions openai/math pins (Mathlib only, no OAI library)
+#   - Comparator, lean4export and landrun, used only by the verifier's grade.py
+#   - every challenge statement, at /opt/openai-math/challenges/<Name>.lean
+#   - /workspace, a Lake project where the agent writes Submission.lean
+# All revisions are pinned. The openai/math library (the reference proofs) is never installed.
+set -euxo pipefail
+export DEBIAN_FRONTEND=noninteractive
+
+LEAN_TOOLCHAIN=leanprover/lean4:v4.34.1
+OPENAI_MATH_REV=adc7f1241b42e322a6451854ab7e4b4c146bf78a
+COMPARATOR_REV=d03acab154d269c06e60e4de7e4cc85deebff94b   # "bump toolchain to v4.34.0"; built with v4.34.1
+LANDRUN_REV=811cfff51ceaf3d9843708aa6d22e9b84ccac8b4
+GO_VERSION=1.24.7
+
+apt-get update -qq
+apt-get install -y -qq --no-install-recommends git curl ca-certificates build-essential xz-utils \
+  zstd python3 procps iproute2 iptables tmux >/dev/null
+rm -rf /var/lib/apt/lists/*
+git config --system --add safe.directory '*'
+
+export ELAN_HOME=/opt/elan PATH=/opt/elan/bin:$PATH
+curl -sSfL https://raw.githubusercontent.com/leanprover/elan/master/elan-init.sh \
+  | sh -s -- -y --default-toolchain "$LEAN_TOOLCHAIN" --no-modify-path
+
+# --- Verifier tools, built before Mathlib so the build trees are gone before the big download ---
+curl -sSfL "https://go.dev/dl/go${GO_VERSION}.linux-amd64.tar.gz" | tar -C /tmp -xz
+GOBIN=/usr/local/bin GOPATH=/tmp/gopath GOCACHE=/tmp/gocache \
+  /tmp/go/bin/go install "github.com/zouuup/landrun/cmd/landrun@${LANDRUN_REV}"
+rm -rf /tmp/go /tmp/gopath /tmp/gocache
+
+git clone -q https://github.com/leanprover/comparator.git /tmp/comparator
+(
+  cd /tmp/comparator
+  git checkout -q "$COMPARATOR_REV"
+  # Comparator passes only PATH/HOME/LEAN_ABORT_ON_PANIC into its landrun sandbox. Also pass
+  # LEAN_NUM_THREADS: Lean sees every host CPU (64) rather than the sandbox's 4, so without it Lake
+  # starts one Mathlib-loaded `lean` per module and an 8 GB sandbox runs out of memory. It bounds
+  # parallelism only and does not affect what is checked.
+  sed -i 's/envPass := #\["PATH", "HOME", "LEAN_ABORT_ON_PANIC"\]/envPass := #["PATH", "HOME", "LEAN_ABORT_ON_PANIC", "LEAN_NUM_THREADS"]/; s/envPass := #\["PATH", "HOME", "LEAN_PATH", "LEAN_ABORT_ON_PANIC"\]/envPass := #["PATH", "HOME", "LEAN_PATH", "LEAN_ABORT_ON_PANIC", "LEAN_NUM_THREADS"]/' Main.lean
+  test "$(grep -c '"LEAN_NUM_THREADS"' Main.lean)" -eq 2
+  echo "$LEAN_TOOLCHAIN" > lean-toolchain
+  lake build lean4export comparator
+  install -m 0755 .lake/build/bin/comparator /usr/local/bin/comparator
+  install -m 0755 .lake/packages/lean4export/.lake/build/bin/lean4export /usr/local/bin/lean4export
+)
+rm -rf /tmp/comparator
+
+# --- Lean project with Mathlib ---
+# Packages are shallow clones at the manifest revisions, so Lake sees the right HEAD and never fetches.
+mkdir -p /opt/lean-env/.lake/packages
+cd /opt/lean-env
+echo "$LEAN_TOOLCHAIN" > lean-toolchain
+cp /opt/build/lakefile.lean /opt/build/lake-manifest.json .
+python3 - <<'PY'
+import json, subprocess
+for p in json.load(open("lake-manifest.json"))["packages"]:
+    d = f".lake/packages/{p['name']}"
+    subprocess.run(["git", "init", "-q", d], check=True)
+    subprocess.run(["git", "-C", d, "remote", "add", "origin", p["url"]], check=True)
+    subprocess.run(["git", "-C", d, "fetch", "-q", "--depth", "1", "origin", p["rev"]], check=True)
+    subprocess.run(["git", "-C", d, "checkout", "-q", "FETCH_HEAD"], check=True)
+PY
+# Linking native executables (Comparator above, Mathlib's cache tool here) is the only use of the
+# static libraries, LLVM and gcc; `lean` and `lake build` of a library need none of them. Dropping
+# them before the Mathlib download keeps the build, and the image, under Daytona's 10 GB disk.
+lake exe cache --help >/dev/null 2>&1 || true
+find /opt/elan/toolchains/*/lib -name '*.a' -delete
+rm -f /opt/elan/toolchains/*/lib/libLLVM* /opt/elan/toolchains/*/lib/libclang-cpp*
+apt-get purge -y -qq $(dpkg-query -W -f '${binary:Package}\n' build-essential 'gcc-13*' 'g++-13*' \
+  'cpp-13*' 'libstdc++-13-dev*' 'libgcc-13-dev*' 2>/dev/null) >/dev/null || true
+apt-get autoremove -y -qq >/dev/null || true
+bash /opt/build/fetch_mathlib.sh
+rm -f /opt/lean-env/.lake/packages/mathlib/.lake/build/bin/cache
+
+# --- Challenge statements (public; the reference proofs are not fetched) ---
+git init -q /tmp/oai && cd /tmp/oai
+git remote add origin https://github.com/openai/math.git
+git config core.sparseCheckout true
+echo "lean/ComparatorChallenges/" > .git/info/sparse-checkout
+git fetch -q --depth 1 --filter=blob:none origin "$OPENAI_MATH_REV"
+git checkout -q FETCH_HEAD
+mkdir -p /opt/openai-math/challenges
+cp lean/ComparatorChallenges/*.lean /opt/openai-math/challenges/
+cd / && rm -rf /tmp/oai
+install -m 0755 /opt/build/pack_submission.py /opt/openai-math/pack_submission.py
+
+# --- Agent workspace and the verifier's unprivileged user ---
+mkdir -p /workspace/.lake
+cp /opt/lean-env/lakefile.lean /opt/lean-env/lake-manifest.json /opt/lean-env/lean-toolchain /workspace/
+ln -s /opt/lean-env/.lake/packages /workspace/.lake/packages
+useradd --uid 2000 --create-home --shell /usr/sbin/nologin grader
+mkdir -p /var/lib/oai-grade && chmod 0711 /var/lib/oai-grade
+
+# Smoke test. The first build may rebuild a few dependency modules the cache lacks (they persist in
+# the image); a second build must compile nothing but Submission, or agents would pay for it.
+cd /workspace
+printf 'import Mathlib\n\nexample : (2 : ℕ) + 2 = 4 := by norm_num\n' > Submission.lean
+lake build Submission 2>&1 | grep -E 'Built|error' | grep -v 'Built Submission' | head -20 || true
+echo '-- second build' >> Submission.lean
+lake build Submission 2>&1 | tee /tmp/smoke.log
+if grep 'Built' /tmp/smoke.log | grep -qv 'Built Submission'; then
+  echo "the prebuilt Mathlib is incomplete: a second build still compiled dependencies" >&2; exit 1
+fi
+rm -rf Submission.lean .lake/build /tmp/smoke.log
+
+# Mathlib's cache tool unpacks files as 0600. The verifier builds as the unprivileged `grader` user,
+# so everything must be world-readable (and stays root-owned, so not writable).
+chmod -R a+rX /opt/elan /opt/lean-env /opt/openai-math
+# Same smoke test as `grader`, in a scratch project like the one grade.py lays out.
+smoke=$(mktemp -d /var/lib/oai-grade/smoke.XXXX)
+cp /opt/lean-env/lakefile.lean /opt/lean-env/lake-manifest.json /opt/lean-env/lean-toolchain "$smoke/"
+mkdir "$smoke/.lake" && ln -s /opt/lean-env/.lake/packages "$smoke/.lake/packages"
+printf 'import Mathlib\n\nexample : (2 : ℕ) + 2 = 4 := by norm_num\n' > "$smoke/Submission.lean"
+chmod 0755 "$smoke" && chown grader "$smoke/.lake"
+(cd "$smoke" && runuser -u grader -- env -i PATH="$(echo /opt/elan/toolchains/*/bin):/usr/bin:/bin" HOME="$smoke" \
+  lake build Submission) 2>&1 | tee /tmp/smoke.log
+if grep 'Built' /tmp/smoke.log | grep -qv 'Built Submission' || ! grep -q 'Build completed successfully' /tmp/smoke.log; then
+  echo "the grader user cannot build against the prebuilt Mathlib" >&2; exit 1
+fi
+rm -rf "$smoke" /tmp/smoke.log
+
+rm -rf /opt/build /root/.cache
+df -h / | tail -1
+du -sh /opt/elan /opt/lean-env
diff --git a/tasks/atomic-gaussian/environment/pack_submission.py b/tasks/atomic-gaussian/environment/pack_submission.py
new file mode 100644
index 0000000000000000000000000000000000000000..689dff39666b1453a328498760d185e309d9b99e
--- /dev/null
+++ b/tasks/atomic-gaussian/environment/pack_submission.py
@@ -0,0 +1,70 @@
+#!/usr/bin/env python3
+"""Pack the agent's proof into one file for the verifier.
+
+Harbor runs this as a [[verifier.collect]] hook in the agent's sandbox after the agent finishes.
+It clears /logs/artifacts and writes /logs/artifacts/submission.tar holding only regular
+`Submission.lean` and `Submission/**.lean` files. Symlinks and other entries are skipped, never
+followed, so only one plain file crosses to the verifier sandbox.
+
+The agent's sandbox is untrusted, so the verifier re-checks every entry of the archive. This step
+only keeps the transfer small and free of links.
+"""
+
+from __future__ import annotations
+
+import os
+import shutil
+import stat
+import tarfile
+from pathlib import Path
+
+WORKSPACE = Path("/workspace")
+OUT_DIR = Path("/logs/artifacts")
+MAX_TOTAL_BYTES = 768 << 20
+
+
+def clear(directory: Path) -> None:
+    directory.mkdir(parents=True, exist_ok=True)
+    for entry in directory.iterdir():
+        if entry.is_dir() and not entry.is_symlink():
+            shutil.rmtree(entry)
+        else:
+            entry.unlink()
+
+
+def regular(path: Path) -> bool:
+    try:
+        return stat.S_ISREG(path.lstat().st_mode)
+    except OSError:
+        return False
+
+
+def main() -> None:
+    clear(OUT_DIR)
+    tmp = OUT_DIR / "submission.tar.part"
+    total = 0
+    with tarfile.open(tmp, "w", format=tarfile.PAX_FORMAT) as tar:
+        root = WORKSPACE / "Submission.lean"
+        if regular(root):
+            total += root.lstat().st_size
+            if total > MAX_TOTAL_BYTES:
+                raise SystemExit("submission exceeds the size limit")
+            tar.add(root, arcname="Submission.lean", recursive=False)
+        tree = WORKSPACE / "Submission"
+        if tree.is_dir() and not tree.is_symlink():
+            for dirpath, dirnames, filenames in os.walk(tree, followlinks=False):
+                dirnames[:] = [d for d in dirnames if not (Path(dirpath) / d).is_symlink()]
+                for name in sorted(filenames):
+                    path = Path(dirpath) / name
+                    if not name.endswith(".lean") or not regular(path):
+                        continue
+                    total += path.lstat().st_size
+                    if total > MAX_TOTAL_BYTES:
+                        raise SystemExit("submission exceeds the size limit")
+                    tar.add(path, arcname=str(path.relative_to(WORKSPACE)), recursive=False)
+    tmp.rename(OUT_DIR / "submission.tar")
+    print(f"packed {total} bytes")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/tasks/atomic-gaussian/instruction.md b/tasks/atomic-gaussian/instruction.md
new file mode 100644
index 0000000000000000000000000000000000000000..cc4830e5f7311aa5ffb67a2bac04f1aa17bb5790
--- /dev/null
+++ b/tasks/atomic-gaussian/instruction.md
@@ -0,0 +1,303 @@
+# openai/math challenge `AtomicGaussian` (family 090)
+
+Prove the following result from OpenAI's [openai/math](https://github.com/openai/math) release in
+Lean 4, with a proof the Lean kernel accepts.
+
+Context: this statement belongs to family 090 of the release, *Universal optimality of the triangular lattice*
+(Convex and metric geometry). The family as a whole: Proves that the triangular lattice minimizes the lower limit of energy per particle for every nonnegative completely monotone potential of squared distance among locally finite planar configurations of centered-disk density one. It also minimizes unit-background renormalized Riesz energies for $0<s<2$ and Coulomb energy, resolving Sandier–Serfaty and the two-dimensional Brauchart–Hardin–Saff conjecture on the linear term of optimal spherical logarithmic energy.
+
+The challenge is `AtomicGaussian`, also at `/opt/openai-math/challenges/AtomicGaussian.lean`:
+
+```lean
+import Mathlib
+
+namespace OAI
+
+noncomputable section
+open scoped BigOperators FourierTransform SchwartzMap
+open Complex Polynomial
+
+namespace AtomicTriangular
+
+abbrev Plane := EuclideanSpace ℝ (Fin 2)
+
+def b : ℝ := Real.sqrt 3 / 2
+
+def triangularPoint (jk : ℤ × ℤ) : Plane :=
+  (Real.sqrt b)⁻¹ •
+    (EuclideanSpace.single 0 ((jk.1 : ℝ) + (jk.2 : ℝ) / 2) +
+      EuclideanSpace.single 1 ((jk.2 : ℝ) * b))
+
+def A : Set Plane := Set.range triangularPoint
+
+def dualA : Set Plane := {w | ∀ a ∈ A, ∃ n : ℤ, inner ℝ a w = (n : ℝ)}
+
+def gaussian (α : ℝ) (x : Plane) : ℝ := Real.exp (-Real.pi * α * ‖x‖ ^ 2)
+
+def Radial (f : Plane → ℝ) : Prop := ∀ x y, ‖x‖ = ‖y‖ → f x = f y
+
+def realFourier (f : Plane → ℝ) (w : Plane) : ℂ := 𝓕 (fun x => (f x : ℂ)) w
+
+end AtomicTriangular
+
+namespace AtomicTriangular.Construction
+
+def b : ℝ := Real.sqrt 3 / 2
+
+def h : ℝ := 17 / 50
+
+def H : ℝ := 27 / 50
+
+def eta : ℝ := 1 / 5
+
+def kappa : ℝ := Real.pi / 36
+
+def residues : Finset ℤ := {0,1,3,4,7,9,12,13,16,19,21,25,27,28,31}
+
+def residue (a : Fin 15) : ℤ := ![0,1,3,4,7,9,12,13,16,19,21,25,27,28,31] a
+
+def Node : Type := {n : ℤ // 0 < n ∧ n % 36 ∈ residues}
+
+def FiniteNodes : Finset ℤ := {1,3,4,7,9,12,13,16,19,21}
+
+def sineFactor (a : Fin 15) (s : ℂ) : ℂ :=
+  2 * Complex.sin ((kappa : ℂ) * (s - (residue a : ℂ)))
+
+def P (s : ℂ) : ℂ := ∏ a : Fin 15, (sineFactor a s)^2
+
+def dBasis (n : Node) : ℂ → ℂ := dslope P (n.val : ℂ)
+
+def cBasis (n : Node) : ℂ → ℂ := dslope (dBasis n) (n.val : ℂ)
+
+def finiteC (a : Fin 15) (s : ℂ) : ℂ :=
+  4 * (kappa : ℂ)^2 * ∏ r ∈ Finset.univ.erase a, (sineFactor r s)^2
+
+def finiteD (a : Fin 15) (s : ℂ) : ℂ :=
+  2 * (kappa : ℂ) * sineFactor a s *
+    Complex.cos ((kappa : ℂ) * (s - (residue a : ℂ))) *
+      ∏ r ∈ Finset.univ.erase a, (sineFactor r s)^2
+
+def phase (s : ℂ) : ℂ := Complex.exp ((Real.pi : ℂ) * Complex.I * s / 18)
+
+def atomFactor (a : Fin 15) : Polynomial ℂ :=
+  -C (phase (residue a)) + C 2 * X - C (phase (-residue a)) * X^2
+
+def atomPolynomial : Polynomial ℂ := ∏ a : Fin 15, atomFactor a
+
+def cAtomPolynomial (a : Fin 15) : Polynomial ℂ :=
+  C (4 * (kappa : ℂ)^2) * ∏ r ∈ Finset.univ.erase a, atomFactor r
+
+def dAtomPolynomial (a : Fin 15) : Polynomial ℂ :=
+  C (Complex.I * (kappa : ℂ)) *
+    (C (phase (residue a)) - C (phase (-residue a)) * X^2) *
+      ∏ r ∈ Finset.univ.erase a, atomFactor r
+
+def fourierCoeff (j : Fin 31) : ℂ := atomPolynomial.coeff j.val
+
+def frequency (j : Fin 31) : ℝ := ((j.val : ℝ)-15)/18
+
+def waveDerivative (k : ℕ) (s : ℂ) : ℂ :=
+  ∑ j : Fin 31, fourierCoeff j * (I*Real.pi*frequency j)^k *
+    exp (I*Real.pi*frequency j*s)
+
+def Q (n : ℤ) : ℝ := (waveDerivative 2 n).re / 2
+
+def D (n : ℤ) : ℝ := (waveDerivative 3 n).re / (6 * Q n)
+
+end AtomicTriangular.Construction
+
+namespace AtomicTriangular.FiniteCertificate
+open Construction
+
+def cShiftPolynomial (a : Fin 15) : Polynomial ℂ := X * cAtomPolynomial a
+
+def cCoefficient (a : Fin 15) (j : Fin 31) : ℂ := (cShiftPolynomial a).coeff j.val
+
+def dCoefficient (a : Fin 15) (j : Fin 31) : ℂ := (dAtomPolynomial a).coeff j.val
+
+end AtomicTriangular.FiniteCertificate
+
+namespace AtomicTriangular.Construction
+
+abbrev Plane := EuclideanSpace ℝ (Fin 2)
+
+def radial (x : Plane) : ℝ := b*‖x‖^2
+
+def waveDen (k t : ℝ) : ℝ := k^2+t^2
+
+def Xwave (k t : ℝ) : ℝ := (4/3)/waveDen k t
+
+def lambda (k t : ℝ) : ℂ := ((k:ℂ)+I*t)/(b*waveDen k t : ℝ)
+
+def zeta (k v t : ℝ) : ℂ := -(t:ℂ)*(Xwave k t:ℂ)+I*(k*Xwave k t-v:ℝ)
+
+def transformedWave (k v t : ℝ) (s : ℂ) : ℂ := lambda k t*exp (I*Real.pi*zeta k v t*s)
+
+def finiteColumn (col : Fin 23) (j : Fin 31) : ℂ :=
+  if hc : col.val<22 then
+    if col.val%2=0 then FiniteCertificate.cCoefficient ⟨col.val/2,by omega⟩ j
+      else FiniteCertificate.dCoefficient ⟨col.val/2,by omega⟩ j
+  else fourierCoeff j
+
+def finiteNode (n : Fin 10) : ℤ := residue ⟨n.val+1,by omega⟩
+
+def jetMultiplier (k v : ℝ) (n : ℤ) (d : Fin 2) (t : ℝ) : ℂ :=
+  (Q n:ℂ)⁻¹*transformedWave k v t n *
+    (if d.val=0 then -1 else (D n:ℂ)-I*Real.pi*zeta k v t)
+
+def jet (k v : ℝ) (n : ℤ) (d : Fin 2) (S : Fin 31 → ℂ) : ℝ :=
+  (∑ j : Fin 31, jetMultiplier k v n d (frequency j)*S j).re
+
+def fullJetMatrix : Matrix (Fin 20) (Fin 23) ℝ := fun i col =>
+  jet (if col.val=22 then h else H) H
+    (finiteNode ⟨i.val/2,by omega⟩) ⟨i.val%2,by omega⟩ (finiteColumn col)
+
+def R : Matrix (Fin 20) (Fin 20) ℝ := fun i j => fullJetMatrix i ⟨j.val+2,by omega⟩
+
+def R0 : Matrix (Fin 20) (Fin 3) ℝ := fun i j =>
+  fullJetMatrix i (if j.val<2 then ⟨j.val,by omega⟩ else 22)
+
+abbrev TailNode := {n : Node // n.val ∉ FiniteNodes}
+
+structure InputCoefficients where
+  C : ℝ
+  finiteC : Fin 11 → ℝ
+  finiteD : Fin 11 → ℝ
+  tailC : TailNode → ℝ
+  tailD : TailNode → ℝ
+  summable : Summable (fun n => |tailC n|+|tailD n|)
+
+def tailInput (a : InputCoefficients) (s : ℂ) : ℂ :=
+  (a.C:ℂ)*P s + ∑' n : TailNode, ((a.tailC n:ℂ)*cBasis n.val s+(a.tailD n:ℂ)*dBasis n.val s)
+
+def finiteInput (a : InputCoefficients) (s : ℂ) : ℂ :=
+  ∑ j : Fin 11, ((a.finiteC j:ℂ)*finiteC ⟨j.val,by omega⟩ s +
+    (a.finiteD j:ℂ)*finiteD ⟨j.val,by omega⟩ s)
+
+def inputComplex (a : InputCoefficients) (x : Plane) : ℂ :=
+  exp (-(Real.pi:ℂ)*h*(radial x:ℂ))*tailInput a (radial x) +
+    exp (-(Real.pi:ℂ)*H*(radial x:ℂ))*finiteInput a (radial x)
+
+def Z : ℝ := 391/1000
+
+def targetCoefficient (n : Fin 10) : ℝ := Q 1 / Q (finiteNode n) * Real.exp (Real.pi*eta*(finiteNode n))
+
+def Yf : Matrix (Fin 20) (Fin 8) ℝ := fun i j =>
+  if i.val<8 then
+    targetCoefficient ⟨i.val/2,by omega⟩ *
+      (if i.val%2=0 then (if j.val=i.val then 1 else 0)
+       else (if j.val=i.val then -1 else if j.val=i.val-1 then Real.pi*eta-D (finiteNode ⟨i.val/2,by omega⟩) else 0))
+  else 0
+
+def gamma (side : Fin 2) : Matrix (Fin 3) (Fin 8) ℝ := fun i j =>
+  if j.val=1 then (if i.val=0 then 11/25 else if i.val=1 then 0 else if side.val=0 then -(3/500) else 3/500) else 0
+
+def Uplus : Matrix (Fin 20) (Fin 8) ℝ :=
+  (1-R)⁻¹*(Yf+R0*(gamma 1+gamma 0))
+
+def Uminus : Matrix (Fin 20) (Fin 8) ℝ :=
+  (1+R)⁻¹*(Yf+R0*(gamma 1-gamma 0))
+
+def finiteSolution (side : Fin 2) : Matrix (Fin 20) (Fin 8) ℝ :=
+  if side.val=0 then (1/2:ℝ) • (Uplus+Uminus) else (1/2:ℝ) • (Uplus-Uminus)
+
+def Acoefficient (side : Fin 2) : Matrix (Fin 23) (Fin 8) ℝ := fun i j =>
+  if hi : i.val<2 then gamma side ⟨i.val,by omega⟩ j
+  else if h22 : i.val=22 then gamma side 2 j
+  else finiteSolution side ⟨i.val-2,by omega⟩ j
+
+def gaussianZ (α : ℝ) : ℝ := 1/(Real.pi*(α/b-h))
+
+def actualParameter (z : ℝ) : Fin 8 → ℝ :=
+  ![z,1,z*Real.exp (-2/z),Real.exp (-2/z),z*Real.exp (-3/z),Real.exp (-3/z),
+    z*Real.exp (-6/z),Real.exp (-6/z)]
+
+def normalizingDen (z : ℝ) := Q 1*z*Real.exp (1/z)
+
+end AtomicTriangular.Construction
+
+namespace AtomicTriangular.Construction
+
+def finiteCoordinates (a : InputCoefficients) (i : Fin 22) : ℝ :=
+  if i.val % 2 = 0 then a.finiteC ⟨i.val / 2, by omega⟩
+  else a.finiteD ⟨i.val / 2, by omega⟩
+
+def HasAtomicConstruction (α : ℝ) (f : Plane → ℝ) : Prop :=
+  IsUnit (1 - R) ∧ IsUnit (1 + R) ∧
+  (1 - R) * Uplus = Yf + R0 * (gamma 1 + gamma 0) ∧
+  (1 + R) * Uminus = Yf + R0 * (gamma 1 - gamma 0) ∧
+  ∃ (a : Fin 2 → InputCoefficients) (u : Fin 2 → Fin 20 → ℝ),
+    (∀ side,
+      (a side).C = (if side.val = 0 then -(3/500) else 3/500) ∧
+      (a side).finiteC 0 = 11/25 ∧ (a side).finiteD 0 = 0 ∧
+      ∀ i : Fin 22,
+        finiteCoordinates (a side) i =
+          (∑ j : Fin 8, Acoefficient side ⟨i.val, by omega⟩ j *
+            actualParameter (gaussianZ α) j) +
+          (if hi : i.val < 2 then 0 else u side ⟨i.val - 2, by omega⟩)) ∧
+    (∀ side,
+      (∑ i : Fin 20, |u side i|) < 1/10^5 ∧
+      Summable (fun n : TailNode => |(a side).tailC n| + |(a side).tailD n|) ∧
+      (∑' n : TailNode, (|(a side).tailC n| + |(a side).tailD n|)) < 3/10^9) ∧
+    (∀ x : Plane, (f x : ℂ) =
+      (inputComplex (a 0) x + 𝓕 (inputComplex (a 1)) x) /
+        (normalizingDen (gaussianZ α) : ℂ))
+
+end AtomicTriangular.Construction
+
+namespace AtomicTriangular
+
+theorem sharp_gaussian_minorants_atomic (α : ℝ) (hα : 0 < α) :
+    ∃ f : SchwartzMap Plane ℝ,
+      Radial f ∧ (∀ x, f x ≤ gaussian α x) ∧
+      (∀ w, (realFourier f w).im = 0 ∧ 0 ≤ (realFourier f w).re) ∧
+      (∀ a ∈ A, a ≠ 0 → f a = gaussian α a) ∧
+      (∀ w ∈ dualA, w ≠ 0 → realFourier f w = 0) ∧
+      (1 ≤ α → Construction.HasAtomicConstruction α f) := by
+  sorry
+
+end AtomicTriangular
+
+end
+
+end OAI
+```
+
+## What to submit
+
+Write `/workspace/Submission.lean`. Start from a copy of the challenge:
+
+```bash
+cp /opt/openai-math/challenges/AtomicGaussian.lean /workspace/Submission.lean
+```
+
+then replace every `sorry` with a proof. The file is graded on three things:
+
+- **Same statements.** The theorem `OAI.AtomicTriangular.sharp_gaussian_minorants_atomic` must keep exactly the statement shown above: same names,
+  namespaces, binders and types. Every definition the statements use must stay exactly as written.
+  Change nothing except the proofs.
+- **Standard axioms only.** Proofs may use only `propext`, `Quot.sound` and `Classical.choice`.
+  `sorry`, `admit`, new `axiom`s and `native_decide` (it introduces an axiom of its own) are rejected.
+- **Keep the challenge's declarations as they are.** Put new lemmas and instances *after* the
+  definitions the statements use, or in a separate `Submission/*.lean` module. A declaration added
+  before them can change how they elaborate, and then they no longer match the challenge.
+- **Kernel-checked.** The proofs are re-checked by the Lean kernel, not just the elaborator.
+
+Long proofs can be split into modules under `/workspace/Submission/` (module names
+`Submission.Foo`, `Submission.Foo.Bar`) imported from `Submission.lean`. Only `.lean` files at
+those two paths are graded.
+
+## Environment
+
+- Lean `v4.34.1` and Mathlib at commit `d13f23b` are installed and prebuilt; `/workspace` is a
+  Lake project.
+- The sandbox has 4 CPUs and 8 GB of memory; `LEAN_NUM_THREADS=3` keeps `lake build` to three
+  parallel jobs. Check your work with `cd /workspace && lake build Submission`. Add `#print axioms <name>` to see
+  which axioms a proof uses.
+- There is no internet access. OpenAI's own proofs are not installed.
+
+## Grading
+
+When you finish, `Submission.lean` and `Submission/**.lean` are copied to a fresh machine and
+checked with [Comparator](https://github.com/leanprover/comparator), the Lean FRO's proof checker.
+The reward is 1 if Comparator accepts the proof and 0 otherwise. A partial proof scores 0.
diff --git a/tasks/atomic-gaussian/solution/closure.txt b/tasks/atomic-gaussian/solution/closure.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9587af88fe52e81f5f5d27193f128fbaca957137
--- /dev/null
+++ b/tasks/atomic-gaussian/solution/closure.txt
@@ -0,0 +1,215 @@
+OAI.Analysis.Triangular.Auxiliary.Bernstein
+OAI.Analysis.Triangular.Auxiliary.ExponentialTail
+OAI.Analysis.Triangular.Auxiliary.IntegralBounds
+OAI.Analysis.Triangular.Auxiliary.MatrixNorm
+OAI.Analysis.Triangular.Auxiliary.PolynomialShift
+OAI.Analysis.Triangular.Basic
+OAI.Analysis.Triangular.Certificate.Inverse
+OAI.Analysis.Triangular.Certificate.JetMatrix
+OAI.Analysis.Triangular.Certificate.MatrixInterval
+OAI.Analysis.Triangular.Certificate.Param
+OAI.Analysis.Triangular.Certificate.ParamA
+OAI.Analysis.Triangular.Certificate.ParamActual
+OAI.Analysis.Triangular.Certificate.ParamInterval
+OAI.Analysis.Triangular.Certificate.ParamNorm
+OAI.Analysis.Triangular.Certificate.ParamNormBase
+OAI.Analysis.Triangular.Certificate.ParamRHS
+OAI.Analysis.Triangular.Certificate.ParamU
+OAI.Analysis.Triangular.Certificate.Polytope
+OAI.Analysis.Triangular.Certificate.PolytopeInterval
+OAI.Analysis.Triangular.Certificate.TargetInterval
+OAI.Analysis.Triangular.Certificate.WaveExpData
+OAI.Analysis.Triangular.Certificate.WaveIntervals
+OAI.Analysis.Triangular.Certificate.WaveJetsData
+OAI.Analysis.Triangular.Construction.ActualCorrection
+OAI.Analysis.Triangular.Construction.ActualInput
+OAI.Analysis.Triangular.Construction.ActualInterpolation
+OAI.Analysis.Triangular.Construction.ActualTail
+OAI.Analysis.Triangular.Construction.AtomicCurvature
+OAI.Analysis.Triangular.Construction.BlockOperators
+OAI.Analysis.Triangular.Construction.CoefficientAddition
+OAI.Analysis.Triangular.Construction.ConcreteSchur
+OAI.Analysis.Triangular.Construction.CurvatureAnalytic
+OAI.Analysis.Triangular.Construction.CurvatureBoxes
+OAI.Analysis.Triangular.Construction.CurvatureEnvelope
+OAI.Analysis.Triangular.Construction.CurvatureSums
+OAI.Analysis.Triangular.Construction.DirectCoordinates
+OAI.Analysis.Triangular.Construction.DirectWaveJets
+OAI.Analysis.Triangular.Construction.FarCurvature
+OAI.Analysis.Triangular.Construction.FiniteAssembly
+OAI.Analysis.Triangular.Construction.FiniteCoordinates
+OAI.Analysis.Triangular.Construction.FiniteInterpolation
+OAI.Analysis.Triangular.Construction.FiniteJets
+OAI.Analysis.Triangular.Construction.FinitePairBridge
+OAI.Analysis.Triangular.Construction.FiniteProjection
+OAI.Analysis.Triangular.Construction.FiniteRemainder
+OAI.Analysis.Triangular.Construction.FiniteSystem
+OAI.Analysis.Triangular.Construction.FiniteTargetBridge
+OAI.Analysis.Triangular.Construction.L1Kernel
+OAI.Analysis.Triangular.Construction.L1Matrix
+OAI.Analysis.Triangular.Construction.NodeJets
+OAI.Analysis.Triangular.Construction.Normalization
+OAI.Analysis.Triangular.Construction.NormalizedJets
+OAI.Analysis.Triangular.Construction.PeriodicJets
+OAI.Analysis.Triangular.Construction.Repetition
+OAI.Analysis.Triangular.Construction.RepetitionBound
+OAI.Analysis.Triangular.Construction.RepetitionOperators
+OAI.Analysis.Triangular.Construction.Schur
+OAI.Analysis.Triangular.Construction.SpectralJetBridge
+OAI.Analysis.Triangular.Construction.SpectralProduct
+OAI.Analysis.Triangular.Construction.TailCoordinates
+OAI.Analysis.Triangular.Construction.TailEnumeration
+OAI.Analysis.Triangular.Construction.TailInterpolation
+OAI.Analysis.Triangular.Construction.TailJetVector
+OAI.Analysis.Triangular.Energy.Basic
+OAI.Analysis.Triangular.Energy.Mixtures
+OAI.Analysis.Triangular.Envelope.Analytic
+OAI.Analysis.Triangular.Envelope.Basic
+OAI.Analysis.Triangular.Envelope.EnvelopeBoxes
+OAI.Analysis.Triangular.Envelope.EnvelopeNorm0
+OAI.Analysis.Triangular.Envelope.EnvelopeNorm1
+OAI.Analysis.Triangular.Envelope.EnvelopeNorm2
+OAI.Analysis.Triangular.Envelope.EnvelopeNorms
+OAI.Analysis.Triangular.Envelope.EnvelopePsiB
+OAI.Analysis.Triangular.Envelope.EnvelopePsiC
+OAI.Analysis.Triangular.Envelope.EnvelopePsiD
+OAI.Analysis.Triangular.Envelope.EnvelopeSums
+OAI.Analysis.Triangular.Envelope.EnvelopeSumsBase
+OAI.Analysis.Triangular.Envelope.EnvelopeWaves
+OAI.Analysis.Triangular.Envelope.JetEnvelopeAnalytic
+OAI.Analysis.Triangular.Envelope.KernelBounds
+OAI.Analysis.Triangular.Envelope.TailEnvelopeSeries
+OAI.Analysis.Triangular.Envelope.TargetEnvelope
+OAI.Analysis.Triangular.Lattice.LatticeGeometry
+OAI.Analysis.Triangular.Main
+OAI.Analysis.Triangular.Schwartz.FourierCLM
+OAI.Analysis.Triangular.Schwartz.FourierPair
+OAI.Analysis.Triangular.Schwartz.Gaussian
+OAI.Analysis.Triangular.Schwartz.GaussianPair
+OAI.Analysis.Triangular.Schwartz.InputCLM
+OAI.Analysis.Triangular.Schwartz.Realness
+OAI.Analysis.Triangular.Schwartz.Regularity
+OAI.Analysis.Triangular.Schwartz.SpectralCLM
+OAI.Analysis.Triangular.Schwartz.SpectralEnvelope
+OAI.Analysis.Triangular.Schwartz.TailSpectral
+OAI.Analysis.Triangular.Signs.DeletedAccuracy
+OAI.Analysis.Triangular.Signs.DeletedExponential
+OAI.Analysis.Triangular.Signs.DeletedJets
+OAI.Analysis.Triangular.Signs.DeletedQuotient
+OAI.Analysis.Triangular.Signs.Exceptional
+OAI.Analysis.Triangular.Signs.ExceptionalComparison
+OAI.Analysis.Triangular.Signs.ExceptionalInterval
+OAI.Analysis.Triangular.Signs.ExceptionalSign0
+OAI.Analysis.Triangular.Signs.ExceptionalSign1
+OAI.Analysis.Triangular.Signs.ExceptionalSign2
+OAI.Analysis.Triangular.Signs.ExceptionalSign3
+OAI.Analysis.Triangular.Signs.FarSigns
+OAI.Analysis.Triangular.Signs.GaussianDeleted
+OAI.Analysis.Triangular.Signs.NearSigns
+OAI.Analysis.Triangular.Signs.OrdinaryComparison
+OAI.Analysis.Triangular.Signs.RemainderBoxes
+OAI.Analysis.Triangular.Signs.RemainderData0
+OAI.Analysis.Triangular.Signs.RemainderData1
+OAI.Analysis.Triangular.Signs.RemainderData2
+OAI.Analysis.Triangular.Signs.RemainderData3
+OAI.Analysis.Triangular.Signs.RemainderData4
+OAI.Analysis.Triangular.Signs.RemainderData5
+OAI.Analysis.Triangular.Signs.RemainderData6
+OAI.Analysis.Triangular.Signs.RemainderNorm
+OAI.Analysis.Triangular.Signs.RemainderPre
+OAI.Analysis.Triangular.Signs.RemainderSums0
+OAI.Analysis.Triangular.Signs.RemainderSums1
+OAI.Analysis.Triangular.Signs.RemainderSums2
+OAI.Analysis.Triangular.Signs.RemainderSums3
+OAI.Analysis.Triangular.Signs.RemainderSums4
+OAI.Analysis.Triangular.Signs.RemainderSums5
+OAI.Analysis.Triangular.Signs.RemainderSums6
+OAI.Analysis.Triangular.Signs.SignBase
+OAI.Analysis.Triangular.Signs.Signs00
+OAI.Analysis.Triangular.Signs.Signs01
+OAI.Analysis.Triangular.Signs.Signs02
+OAI.Analysis.Triangular.Signs.Signs03
+OAI.Analysis.Triangular.Signs.Signs04
+OAI.Analysis.Triangular.Signs.Signs05
+OAI.Analysis.Triangular.Signs.Signs06
+OAI.Analysis.Triangular.Signs.Signs10
+OAI.Analysis.Triangular.Signs.Signs11
+OAI.Analysis.Triangular.Signs.Signs12
+OAI.Analysis.Triangular.Signs.Signs13
+OAI.Analysis.Triangular.Signs.Signs14
+OAI.Analysis.Triangular.Signs.Signs15
+OAI.Analysis.Triangular.Signs.Signs16
+OAI.Analysis.Triangular.Signs.SineBarrierBase
+OAI.Analysis.Triangular.Signs.SineGlobal
+OAI.Analysis.Triangular.Signs.SineGrid0
+OAI.Analysis.Triangular.Signs.SineGrid1
+OAI.Analysis.Triangular.Signs.SineGrid2
+OAI.Analysis.Triangular.Signs.SineGridAnalytic
+OAI.Analysis.Triangular.Signs.SineGridChecks
+OAI.Analysis.Triangular.Signs.SineGridCover
+OAI.Analysis.Triangular.Signs.SineGridDefs
+OAI.Analysis.Triangular.Taylor.CoefficientSpectral
+OAI.Analysis.Triangular.Taylor.CurvatureData00
+OAI.Analysis.Triangular.Taylor.CurvatureData01
+OAI.Analysis.Triangular.Taylor.CurvatureData10
+OAI.Analysis.Triangular.Taylor.CurvatureData11
+OAI.Analysis.Triangular.Taylor.FiniteTaylorError
+OAI.Analysis.Triangular.Taylor.FiniteTaylorIdentity
+OAI.Analysis.Triangular.Taylor.Taylor
+OAI.Analysis.Triangular.Taylor.TaylorExp
+OAI.Analysis.Triangular.Taylor.TaylorExp00
+OAI.Analysis.Triangular.Taylor.TaylorExp01
+OAI.Analysis.Triangular.Taylor.TaylorExp10
+OAI.Analysis.Triangular.Taylor.TaylorExp11
+OAI.Analysis.Triangular.Taylor.TaylorPhase
+OAI.Analysis.Triangular.Taylor.TaylorRows00
+OAI.Analysis.Triangular.Taylor.TaylorRows01
+OAI.Analysis.Triangular.Taylor.TaylorRows02
+OAI.Analysis.Triangular.Taylor.TaylorRows03
+OAI.Analysis.Triangular.Taylor.TaylorRows04
+OAI.Analysis.Triangular.Taylor.TaylorRows05
+OAI.Analysis.Triangular.Taylor.TaylorRows06
+OAI.Analysis.Triangular.Taylor.TaylorRows10
+OAI.Analysis.Triangular.Taylor.TaylorRows11
+OAI.Analysis.Triangular.Taylor.TaylorRows12
+OAI.Analysis.Triangular.Taylor.TaylorRows13
+OAI.Analysis.Triangular.Taylor.TaylorRows14
+OAI.Analysis.Triangular.Taylor.TaylorRows15
+OAI.Analysis.Triangular.Taylor.TaylorRows16
+OAI.Analysis.Triangular.Taylor.TaylorRowsBase
+OAI.Analysis.Triangular.Taylor.TaylorSpectrum
+OAI.Analysis.Triangular.Taylor.TaylorWave
+OAI.Analysis.Triangular.Taylor.TaylorWave00
+OAI.Analysis.Triangular.Taylor.TaylorWave000
+OAI.Analysis.Triangular.Taylor.TaylorWave001
+OAI.Analysis.Triangular.Taylor.TaylorWave002
+OAI.Analysis.Triangular.Taylor.TaylorWave003
+OAI.Analysis.Triangular.Taylor.TaylorWave004
+OAI.Analysis.Triangular.Taylor.TaylorWave005
+OAI.Analysis.Triangular.Taylor.TaylorWave006
+OAI.Analysis.Triangular.Taylor.TaylorWave01
+OAI.Analysis.Triangular.Taylor.TaylorWave010
+OAI.Analysis.Triangular.Taylor.TaylorWave011
+OAI.Analysis.Triangular.Taylor.TaylorWave012
+OAI.Analysis.Triangular.Taylor.TaylorWave013
+OAI.Analysis.Triangular.Taylor.TaylorWave014
+OAI.Analysis.Triangular.Taylor.TaylorWave015
+OAI.Analysis.Triangular.Taylor.TaylorWave016
+OAI.Analysis.Triangular.Taylor.TaylorWave10
+OAI.Analysis.Triangular.Taylor.TaylorWave100
+OAI.Analysis.Triangular.Taylor.TaylorWave101
+OAI.Analysis.Triangular.Taylor.TaylorWave102
+OAI.Analysis.Triangular.Taylor.TaylorWave103
+OAI.Analysis.Triangular.Taylor.TaylorWave104
+OAI.Analysis.Triangular.Taylor.TaylorWave105
+OAI.Analysis.Triangular.Taylor.TaylorWave106
+OAI.Analysis.Triangular.Taylor.TaylorWave11
+OAI.Analysis.Triangular.Taylor.TaylorWave110
+OAI.Analysis.Triangular.Taylor.TaylorWave111
+OAI.Analysis.Triangular.Taylor.TaylorWave112
+OAI.Analysis.Triangular.Taylor.TaylorWave113
+OAI.Analysis.Triangular.Taylor.TaylorWave114
+OAI.Analysis.Triangular.Taylor.TaylorWave115
+OAI.Analysis.Triangular.Taylor.TaylorWave116
+OAI.Analysis.Triangular.Taylor.VariationBound
+OAI.Analysis.Triangular.Taylor.WaveFunction
diff --git a/tasks/atomic-gaussian/solution/reference.py b/tasks/atomic-gaussian/solution/reference.py
new file mode 100644
index 0000000000000000000000000000000000000000..5a70dac71c5a6043e3d69a2bd3a4796a8b474bd3
--- /dev/null
+++ b/tasks/atomic-gaussian/solution/reference.py
@@ -0,0 +1,58 @@
+#!/usr/bin/env python3
+"""Place openai/math's reference proof as a submission (used by the oracle's solve.sh).
+
+Standalone: the adapter copies this file into every task's solution/ directory.
+
+    reference.py <openai-math lean dir> <closure.txt> <solution module> <workspace>
+
+Each module in closure.txt (``OAI.A.B``) is copied from ``<lean>/OAI/A/B.lean`` to
+``<workspace>/Submission/OAI/A/B.lean`` with ``import OAI.X`` rewritten to
+``import Submission.OAI.X``; ``Submission.lean`` imports the solution module. Declaration names
+are namespaces, not module paths, so the theorems keep the names the challenge expects.
+"""
+
+from __future__ import annotations
+
+import re
+import sys
+from pathlib import Path
+
+_IMPORT = re.compile(r"^(\s*(?:public\s+|private\s+)?(?:meta\s+)?import\s+(?:all\s+)?)(OAI(?:\.|\s|$))")
+_HEADER_LINE = re.compile(r"^\s*(?:module\b|prelude\b|--|$)")
+
+
+def rename_imports(text: str) -> str:
+    """Rewrite ``import OAI...`` to ``import Submission.OAI...`` in the import header only."""
+    out, in_header, in_comment = [], True, False
+    for line in text.splitlines(keepends=True):
+        if in_header:
+            stripped = line.strip()
+            if in_comment:
+                in_comment = "-/" not in stripped
+            elif stripped.startswith("/-"):
+                in_comment = "-/" not in stripped[2:]
+            elif _IMPORT.match(line):
+                line = _IMPORT.sub(lambda m: m.group(1) + "Submission." + m.group(2), line, count=1)
+            elif not _HEADER_LINE.match(line) and not re.match(r"^\s*(?:public\s+|private\s+)?(?:meta\s+)?import\s", line):
+                in_header = False
+        out.append(line)
+    return "".join(out)
+
+
+def module_path(root: Path, module: str) -> Path:
+    return root.joinpath(*module.split(".")).with_suffix(".lean")
+
+
+def place(lean_dir: Path, modules: list[str], solution_module: str, workspace: Path) -> None:
+    for module in modules:
+        dst = module_path(workspace / "Submission", module)
+        dst.parent.mkdir(parents=True, exist_ok=True)
+        dst.write_text(rename_imports(module_path(lean_dir, module).read_text(encoding="utf-8")), encoding="utf-8")
+    (workspace / "Submission.lean").write_text(f"import Submission.{solution_module}\n", encoding="utf-8")
+
+
+if __name__ == "__main__":
+    lean_dir, closure_txt, solution_module, workspace = sys.argv[1:5]
+    modules = [m.strip() for m in Path(closure_txt).read_text().splitlines() if m.strip()]
+    place(Path(lean_dir), modules, solution_module, Path(workspace))
+    print(f"placed {len(modules)} reference modules for {solution_module}")
diff --git a/tasks/atomic-gaussian/solution/solve.sh b/tasks/atomic-gaussian/solution/solve.sh
new file mode 100644
index 0000000000000000000000000000000000000000..4001635e665037a0dbbc914d2725783d1ea4430e
--- /dev/null
+++ b/tasks/atomic-gaussian/solution/solve.sh
@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+# Oracle: submit openai/math's own proof of AtomicGaussian, i.e. module OAI.Analysis.Triangular.Main and the
+# 215 OAI modules it imports, placed under Submission/ (see reference.py).
+set -euo pipefail
+if [ -f /solution/reference.tar.gz ]; then
+  tar -xzf /solution/reference.tar.gz -C /workspace
+else
+  # Fetch only those files at the pinned commit. The task blocks the network while the agent runs,
+  # so run the oracle with `harbor run ... --allow-agent-host github.com`.
+  tmp=$(mktemp -d)
+  git -C "$tmp" init -q
+  git -C "$tmp" remote add origin https://github.com/openai/math.git
+  git -C "$tmp" config core.sparseCheckout true
+  sed -e 's#\.#/#g' -e 's#^#/lean/#' -e 's#$#.lean#' /solution/closure.txt > "$tmp/.git/info/sparse-checkout"
+  git -C "$tmp" fetch -q --depth 1 --filter=blob:none origin adc7f1241b42e322a6451854ab7e4b4c146bf78a
+  git -C "$tmp" checkout -q FETCH_HEAD
+  python3 /solution/reference.py "$tmp/lean" /solution/closure.txt OAI.Analysis.Triangular.Main /workspace
+  rm -rf "$tmp"
+fi
+echo "placed the reference proof of AtomicGaussian"
diff --git a/tasks/atomic-gaussian/task.toml b/tasks/atomic-gaussian/task.toml
new file mode 100644
index 0000000000000000000000000000000000000000..49f4933979a7c239388e9b84e5a44757bb195501
--- /dev/null
+++ b/tasks/atomic-gaussian/task.toml
@@ -0,0 +1,49 @@
+schema_version = "1.4"
+
+[task]
+name = "openai-math/atomic-gaussian"
+version = "1.0.0"
+description = "Prove AtomicGaussian (openai/math family 090: Universal optimality of the triangular lattice) in Lean 4 with Mathlib; graded by Comparator in a separate sandbox."
+keywords = ["lean4", "mathlib", "theorem-proving", "openai-math", "convex-and-metric-geometry"]
+
+[metadata]
+source = "https://github.com/openai/math"
+source_commit = "adc7f1241b42e322a6451854ab7e4b4c146bf78a"
+source_license = "Apache-2.0"
+source_notice = "tests/Challenge.lean is lean/ComparatorChallenges/AtomicGaussian.lean from openai/math, (c) OpenAI, unchanged; see NOTICE at the dataset root"
+challenge = "AtomicGaussian"
+family = "090"
+family_title = "Universal optimality of the triangular lattice"
+subject = "Convex and metric geometry"
+theorems = ["OAI.AtomicTriangular.sharp_gaussian_minorants_atomic"]
+definition_holes = []
+fixed_definitions = []
+reference_module = "OAI.Analysis.Triangular.Main"
+reference_modules = 215
+reference_lean_bytes = 8527615
+reference_requires_packages = []
+oracle_result = "accepted"
+oracle_grade_seconds = 10626
+difficulty = "research"
+
+[agent]
+timeout_sec = 14400
+network_mode = "no-network"
+
+[environment]
+build_timeout_sec = 5400
+cpus = 4
+memory_mb = 8192
+storage_mb = 10240
+workdir = "/workspace"
+# Lean sees all host CPUs rather than the sandbox's 4; this bounds `lake build` to 3 parallel jobs.
+env = { LEAN_NUM_THREADS = "3" }
+
+[verifier]
+timeout_sec = 19374
+environment_mode = "separate"
+network_mode = "no-network"
+
+[[verifier.collect]]
+command = "python3 /opt/openai-math/pack_submission.py"
+timeout_sec = 300
diff --git a/tasks/atomic-gaussian/tests/Challenge.lean b/tasks/atomic-gaussian/tests/Challenge.lean
new file mode 100644
index 0000000000000000000000000000000000000000..648c5a91452283eb94868bac124b0a46e16c41b0
--- /dev/null
+++ b/tasks/atomic-gaussian/tests/Challenge.lean
@@ -0,0 +1,252 @@
+import Mathlib
+
+namespace OAI
+
+noncomputable section
+open scoped BigOperators FourierTransform SchwartzMap
+open Complex Polynomial
+
+namespace AtomicTriangular
+
+abbrev Plane := EuclideanSpace ℝ (Fin 2)
+
+def b : ℝ := Real.sqrt 3 / 2
+
+def triangularPoint (jk : ℤ × ℤ) : Plane :=
+  (Real.sqrt b)⁻¹ •
+    (EuclideanSpace.single 0 ((jk.1 : ℝ) + (jk.2 : ℝ) / 2) +
+      EuclideanSpace.single 1 ((jk.2 : ℝ) * b))
+
+def A : Set Plane := Set.range triangularPoint
+
+def dualA : Set Plane := {w | ∀ a ∈ A, ∃ n : ℤ, inner ℝ a w = (n : ℝ)}
+
+def gaussian (α : ℝ) (x : Plane) : ℝ := Real.exp (-Real.pi * α * ‖x‖ ^ 2)
+
+def Radial (f : Plane → ℝ) : Prop := ∀ x y, ‖x‖ = ‖y‖ → f x = f y
+
+def realFourier (f : Plane → ℝ) (w : Plane) : ℂ := 𝓕 (fun x => (f x : ℂ)) w
+
+end AtomicTriangular
+
+namespace AtomicTriangular.Construction
+
+def b : ℝ := Real.sqrt 3 / 2
+
+def h : ℝ := 17 / 50
+
+def H : ℝ := 27 / 50
+
+def eta : ℝ := 1 / 5
+
+def kappa : ℝ := Real.pi / 36
+
+def residues : Finset ℤ := {0,1,3,4,7,9,12,13,16,19,21,25,27,28,31}
+
+def residue (a : Fin 15) : ℤ := ![0,1,3,4,7,9,12,13,16,19,21,25,27,28,31] a
+
+def Node : Type := {n : ℤ // 0 < n ∧ n % 36 ∈ residues}
+
+def FiniteNodes : Finset ℤ := {1,3,4,7,9,12,13,16,19,21}
+
+def sineFactor (a : Fin 15) (s : ℂ) : ℂ :=
+  2 * Complex.sin ((kappa : ℂ) * (s - (residue a : ℂ)))
+
+def P (s : ℂ) : ℂ := ∏ a : Fin 15, (sineFactor a s)^2
+
+def dBasis (n : Node) : ℂ → ℂ := dslope P (n.val : ℂ)
+
+def cBasis (n : Node) : ℂ → ℂ := dslope (dBasis n) (n.val : ℂ)
+
+def finiteC (a : Fin 15) (s : ℂ) : ℂ :=
+  4 * (kappa : ℂ)^2 * ∏ r ∈ Finset.univ.erase a, (sineFactor r s)^2
+
+def finiteD (a : Fin 15) (s : ℂ) : ℂ :=
+  2 * (kappa : ℂ) * sineFactor a s *
+    Complex.cos ((kappa : ℂ) * (s - (residue a : ℂ))) *
+      ∏ r ∈ Finset.univ.erase a, (sineFactor r s)^2
+
+def phase (s : ℂ) : ℂ := Complex.exp ((Real.pi : ℂ) * Complex.I * s / 18)
+
+def atomFactor (a : Fin 15) : Polynomial ℂ :=
+  -C (phase (residue a)) + C 2 * X - C (phase (-residue a)) * X^2
+
+def atomPolynomial : Polynomial ℂ := ∏ a : Fin 15, atomFactor a
+
+def cAtomPolynomial (a : Fin 15) : Polynomial ℂ :=
+  C (4 * (kappa : ℂ)^2) * ∏ r ∈ Finset.univ.erase a, atomFactor r
+
+def dAtomPolynomial (a : Fin 15) : Polynomial ℂ :=
+  C (Complex.I * (kappa : ℂ)) *
+    (C (phase (residue a)) - C (phase (-residue a)) * X^2) *
+      ∏ r ∈ Finset.univ.erase a, atomFactor r
+
+def fourierCoeff (j : Fin 31) : ℂ := atomPolynomial.coeff j.val
+
+def frequency (j : Fin 31) : ℝ := ((j.val : ℝ)-15)/18
+
+def waveDerivative (k : ℕ) (s : ℂ) : ℂ :=
+  ∑ j : Fin 31, fourierCoeff j * (I*Real.pi*frequency j)^k *
+    exp (I*Real.pi*frequency j*s)
+
+def Q (n : ℤ) : ℝ := (waveDerivative 2 n).re / 2
+
+def D (n : ℤ) : ℝ := (waveDerivative 3 n).re / (6 * Q n)
+
+end AtomicTriangular.Construction
+
+namespace AtomicTriangular.FiniteCertificate
+open Construction
+
+def cShiftPolynomial (a : Fin 15) : Polynomial ℂ := X * cAtomPolynomial a
+
+def cCoefficient (a : Fin 15) (j : Fin 31) : ℂ := (cShiftPolynomial a).coeff j.val
+
+def dCoefficient (a : Fin 15) (j : Fin 31) : ℂ := (dAtomPolynomial a).coeff j.val
+
+end AtomicTriangular.FiniteCertificate
+
+namespace AtomicTriangular.Construction
+
+abbrev Plane := EuclideanSpace ℝ (Fin 2)
+
+def radial (x : Plane) : ℝ := b*‖x‖^2
+
+def waveDen (k t : ℝ) : ℝ := k^2+t^2
+
+def Xwave (k t : ℝ) : ℝ := (4/3)/waveDen k t
+
+def lambda (k t : ℝ) : ℂ := ((k:ℂ)+I*t)/(b*waveDen k t : ℝ)
+
+def zeta (k v t : ℝ) : ℂ := -(t:ℂ)*(Xwave k t:ℂ)+I*(k*Xwave k t-v:ℝ)
+
+def transformedWave (k v t : ℝ) (s : ℂ) : ℂ := lambda k t*exp (I*Real.pi*zeta k v t*s)
+
+def finiteColumn (col : Fin 23) (j : Fin 31) : ℂ :=
+  if hc : col.val<22 then
+    if col.val%2=0 then FiniteCertificate.cCoefficient ⟨col.val/2,by omega⟩ j
+      else FiniteCertificate.dCoefficient ⟨col.val/2,by omega⟩ j
+  else fourierCoeff j
+
+def finiteNode (n : Fin 10) : ℤ := residue ⟨n.val+1,by omega⟩
+
+def jetMultiplier (k v : ℝ) (n : ℤ) (d : Fin 2) (t : ℝ) : ℂ :=
+  (Q n:ℂ)⁻¹*transformedWave k v t n *
+    (if d.val=0 then -1 else (D n:ℂ)-I*Real.pi*zeta k v t)
+
+def jet (k v : ℝ) (n : ℤ) (d : Fin 2) (S : Fin 31 → ℂ) : ℝ :=
+  (∑ j : Fin 31, jetMultiplier k v n d (frequency j)*S j).re
+
+def fullJetMatrix : Matrix (Fin 20) (Fin 23) ℝ := fun i col =>
+  jet (if col.val=22 then h else H) H
+    (finiteNode ⟨i.val/2,by omega⟩) ⟨i.val%2,by omega⟩ (finiteColumn col)
+
+def R : Matrix (Fin 20) (Fin 20) ℝ := fun i j => fullJetMatrix i ⟨j.val+2,by omega⟩
+
+def R0 : Matrix (Fin 20) (Fin 3) ℝ := fun i j =>
+  fullJetMatrix i (if j.val<2 then ⟨j.val,by omega⟩ else 22)
+
+abbrev TailNode := {n : Node // n.val ∉ FiniteNodes}
+
+structure InputCoefficients where
+  C : ℝ
+  finiteC : Fin 11 → ℝ
+  finiteD : Fin 11 → ℝ
+  tailC : TailNode → ℝ
+  tailD : TailNode → ℝ
+  summable : Summable (fun n => |tailC n|+|tailD n|)
+
+def tailInput (a : InputCoefficients) (s : ℂ) : ℂ :=
+  (a.C:ℂ)*P s + ∑' n : TailNode, ((a.tailC n:ℂ)*cBasis n.val s+(a.tailD n:ℂ)*dBasis n.val s)
+
+def finiteInput (a : InputCoefficients) (s : ℂ) : ℂ :=
+  ∑ j : Fin 11, ((a.finiteC j:ℂ)*finiteC ⟨j.val,by omega⟩ s +
+    (a.finiteD j:ℂ)*finiteD ⟨j.val,by omega⟩ s)
+
+def inputComplex (a : InputCoefficients) (x : Plane) : ℂ :=
+  exp (-(Real.pi:ℂ)*h*(radial x:ℂ))*tailInput a (radial x) +
+    exp (-(Real.pi:ℂ)*H*(radial x:ℂ))*finiteInput a (radial x)
+
+def Z : ℝ := 391/1000
+
+def targetCoefficient (n : Fin 10) : ℝ := Q 1 / Q (finiteNode n) * Real.exp (Real.pi*eta*(finiteNode n))
+
+def Yf : Matrix (Fin 20) (Fin 8) ℝ := fun i j =>
+  if i.val<8 then
+    targetCoefficient ⟨i.val/2,by omega⟩ *
+      (if i.val%2=0 then (if j.val=i.val then 1 else 0)
+       else (if j.val=i.val then -1 else if j.val=i.val-1 then Real.pi*eta-D (finiteNode ⟨i.val/2,by omega⟩) else 0))
+  else 0
+
+def gamma (side : Fin 2) : Matrix (Fin 3) (Fin 8) ℝ := fun i j =>
+  if j.val=1 then (if i.val=0 then 11/25 else if i.val=1 then 0 else if side.val=0 then -(3/500) else 3/500) else 0
+
+def Uplus : Matrix (Fin 20) (Fin 8) ℝ :=
+  (1-R)⁻¹*(Yf+R0*(gamma 1+gamma 0))
+
+def Uminus : Matrix (Fin 20) (Fin 8) ℝ :=
+  (1+R)⁻¹*(Yf+R0*(gamma 1-gamma 0))
+
+def finiteSolution (side : Fin 2) : Matrix (Fin 20) (Fin 8) ℝ :=
+  if side.val=0 then (1/2:ℝ) • (Uplus+Uminus) else (1/2:ℝ) • (Uplus-Uminus)
+
+def Acoefficient (side : Fin 2) : Matrix (Fin 23) (Fin 8) ℝ := fun i j =>
+  if hi : i.val<2 then gamma side ⟨i.val,by omega⟩ j
+  else if h22 : i.val=22 then gamma side 2 j
+  else finiteSolution side ⟨i.val-2,by omega⟩ j
+
+def gaussianZ (α : ℝ) : ℝ := 1/(Real.pi*(α/b-h))
+
+def actualParameter (z : ℝ) : Fin 8 → ℝ :=
+  ![z,1,z*Real.exp (-2/z),Real.exp (-2/z),z*Real.exp (-3/z),Real.exp (-3/z),
+    z*Real.exp (-6/z),Real.exp (-6/z)]
+
+def normalizingDen (z : ℝ) := Q 1*z*Real.exp (1/z)
+
+end AtomicTriangular.Construction
+
+namespace AtomicTriangular.Construction
+
+def finiteCoordinates (a : InputCoefficients) (i : Fin 22) : ℝ :=
+  if i.val % 2 = 0 then a.finiteC ⟨i.val / 2, by omega⟩
+  else a.finiteD ⟨i.val / 2, by omega⟩
+
+def HasAtomicConstruction (α : ℝ) (f : Plane → ℝ) : Prop :=
+  IsUnit (1 - R) ∧ IsUnit (1 + R) ∧
+  (1 - R) * Uplus = Yf + R0 * (gamma 1 + gamma 0) ∧
+  (1 + R) * Uminus = Yf + R0 * (gamma 1 - gamma 0) ∧
+  ∃ (a : Fin 2 → InputCoefficients) (u : Fin 2 → Fin 20 → ℝ),
+    (∀ side,
+      (a side).C = (if side.val = 0 then -(3/500) else 3/500) ∧
+      (a side).finiteC 0 = 11/25 ∧ (a side).finiteD 0 = 0 ∧
+      ∀ i : Fin 22,
+        finiteCoordinates (a side) i =
+          (∑ j : Fin 8, Acoefficient side ⟨i.val, by omega⟩ j *
+            actualParameter (gaussianZ α) j) +
+          (if hi : i.val < 2 then 0 else u side ⟨i.val - 2, by omega⟩)) ∧
+    (∀ side,
+      (∑ i : Fin 20, |u side i|) < 1/10^5 ∧
+      Summable (fun n : TailNode => |(a side).tailC n| + |(a side).tailD n|) ∧
+      (∑' n : TailNode, (|(a side).tailC n| + |(a side).tailD n|)) < 3/10^9) ∧
+    (∀ x : Plane, (f x : ℂ) =
+      (inputComplex (a 0) x + 𝓕 (inputComplex (a 1)) x) /
+        (normalizingDen (gaussianZ α) : ℂ))
+
+end AtomicTriangular.Construction
+
+namespace AtomicTriangular
+
+theorem sharp_gaussian_minorants_atomic (α : ℝ) (hα : 0 < α) :
+    ∃ f : SchwartzMap Plane ℝ,
+      Radial f ∧ (∀ x, f x ≤ gaussian α x) ∧
+      (∀ w, (realFourier f w).im = 0 ∧ 0 ≤ (realFourier f w).re) ∧
+      (∀ a ∈ A, a ≠ 0 → f a = gaussian α a) ∧
+      (∀ w ∈ dualA, w ≠ 0 → realFourier f w = 0) ∧
+      (1 ≤ α → Construction.HasAtomicConstruction α f) := by
+  sorry
+
+end AtomicTriangular
+
+end
+
+end OAI
diff --git a/tasks/atomic-gaussian/tests/grade.py b/tasks/atomic-gaussian/tests/grade.py
new file mode 100644
index 0000000000000000000000000000000000000000..95f89f56da7f59e7af65dcf71b8ef3792ee87399
--- /dev/null
+++ b/tasks/atomic-gaussian/tests/grade.py
@@ -0,0 +1,418 @@
+#!/usr/bin/env python3
+"""Grade one openai/math challenge with Comparator.
+
+This runs as root in a separate verifier sandbox that never executed agent code. The only input the
+agent controls is /logs/artifacts/submission.tar. Everything else comes from /tests (this task's
+trusted challenge and Comparator config) and from the image.
+
+Steps:
+  1. Unpack submission.tar, accepting only regular `Submission.lean` / `Submission/**.lean` files.
+  2. Lay out a fresh Lake project: the trusted lakefile, Challenge.lean from /tests, the submission,
+     and the image's prebuilt Mathlib (root-owned, read-only to the grader).
+  3. Run `lake env comparator config.json` as the unprivileged `grader` user, with no network.
+     Comparator builds Challenge and Submission inside landrun, checks that every listed theorem
+     has exactly the challenge's statement, that only the permitted axioms are used, and replays
+     the proof through the Lean kernel.
+  4. Kill everything the grader started, then write the verdict.
+
+reward.txt is 1 only when Comparator prints its success line and exits 0. A rejected or
+non-compiling submission scores 0. A broken grading setup (the trusted Challenge fails to build,
+the network block cannot be installed) writes no reward, so Harbor reports an error instead of a
+score.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import pwd
+import re
+import shutil
+import signal
+import socket
+import stat
+import subprocess
+import sys
+import tarfile
+import tempfile
+import threading
+import time
+from pathlib import Path
+
+TESTS = Path("/tests")
+VERIFIER_DIR = Path("/logs/verifier")
+SUBMISSION_TAR = Path("/logs/artifacts/submission.tar")
+LEAN_ENV = Path("/opt/lean-env")
+RUN_ROOT = Path("/var/lib/oai-grade")
+GRADER = "grader"
+SUCCESS_LINE = "Your solution is okay!"
+GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
+# Lake and Comparator wording when a build process is killed for running out of memory.
+OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
+
+MAX_FILES = 20_000
+MAX_FILE_BYTES = 64 << 20
+MAX_TOTAL_BYTES = 768 << 20
+# Submission.lean, or Submission/A/B.lean where each component is a Lean identifier: it starts with
+# a letter or `_` and continues with letters, digits (subscripts like `₁` included), `_`, `'`, `!`, `?`.
+# Unicode is allowed because Lean identifiers allow it (openai/math has a module named `Kappa₁`).
+_COMPONENT = r"[^\W\d][\w'!?]*"
+MEMBER = re.compile(rf"^Submission(?:/{_COMPONENT})*\.lean$")
+MEMBER_DIR = re.compile(rf"^Submission(?:/{_COMPONENT})*/?$")
+
+LAKEFILE = """import Lake
+open Lake DSL
+
+package openai_math where
+  leanOptions := #[⟨`autoImplicit, false⟩]
+
+require mathlib from git
+  "https://github.com/leanprover-community/mathlib4.git" @ "d13f23b723b8a846827a245b89c10fc7d3f11612"
+
+lean_lib Challenge
+
+-- `Submission` and every `Submission.*` module it imports.
+lean_lib Submission
+"""
+
+
+class Rejected(Exception):
+    """The submission is invalid: reward 0."""
+
+
+class GradingError(Exception):
+    """The grading setup is broken: no reward, so the trial shows as errored."""
+
+
+def toolchain_bin() -> Path:
+    bins = sorted(Path("/opt/elan/toolchains").glob("*/bin"))
+    if len(bins) != 1:
+        raise GradingError(f"expected one Lean toolchain, found {bins}")
+    return bins[0]
+
+
+def unpack_submission(dest: Path) -> dict:
+    """Copy the submission's .lean files into ``dest``; reject anything else."""
+    try:
+        st = SUBMISSION_TAR.lstat()
+    except FileNotFoundError:
+        raise Rejected("no submission: /workspace/Submission.lean was not found after the agent finished")
+    if not SUBMISSION_TAR.is_file() or SUBMISSION_TAR.is_symlink() or st.st_size > MAX_TOTAL_BYTES + (64 << 20):
+        raise Rejected("submission archive is not a regular file of acceptable size")
+
+    files, total = [], 0
+    try:
+        with tarfile.open(SUBMISSION_TAR, "r:") as tar:
+            for count, member in enumerate(tar):
+                if count > MAX_FILES * 2:
+                    raise Rejected(f"submission has more than {MAX_FILES} entries")
+                name = member.name.removeprefix("./")
+                if member.isdir():
+                    if not MEMBER_DIR.match(name):
+                        raise Rejected(f"unexpected directory in submission: {name!r}")
+                    continue
+                if not member.isreg():
+                    raise Rejected(f"only regular .lean files are accepted, got {name!r}")
+                if not MEMBER.match(name):
+                    raise Rejected(f"unexpected file in submission: {name!r} (only Submission.lean and Submission/**.lean)")
+                if member.size > MAX_FILE_BYTES:
+                    raise Rejected(f"{name} is larger than {MAX_FILE_BYTES >> 20} MB")
+                total += member.size
+                if total > MAX_TOTAL_BYTES or len(files) >= MAX_FILES:
+                    raise Rejected("submission is too large")
+                data = tar.extractfile(member).read(MAX_FILE_BYTES + 1)
+                try:
+                    data.decode("utf-8")
+                except UnicodeDecodeError:
+                    raise Rejected(f"{name} is not valid UTF-8")
+                target = dest / name
+                if target.exists():
+                    raise Rejected(f"duplicate entry {name!r}")
+                target.parent.mkdir(parents=True, exist_ok=True)
+                target.write_bytes(data)
+                target.chmod(0o644)
+                files.append(name)
+    except tarfile.TarError as exc:
+        raise Rejected(f"submission archive is unreadable: {exc}")
+    if "Submission.lean" not in files:
+        raise Rejected("no submission: /workspace/Submission.lean was not found after the agent finished")
+    for d in (dest / "Submission",):
+        if d.exists():
+            for p in [d, *d.rglob("*")]:
+                if p.is_dir():
+                    p.chmod(0o755)
+    return {"files": len(files), "bytes": total}
+
+
+def write_project(run: Path, config: dict) -> None:
+    (run / "lakefile.lean").write_text(LAKEFILE)
+    shutil.copy(LEAN_ENV / "lake-manifest.json", run / "lake-manifest.json")
+    shutil.copy(LEAN_ENV / "lean-toolchain", run / "lean-toolchain")
+    shutil.copy(TESTS / "Challenge.lean", run / "Challenge.lean")
+    comparator_config = {
+        "challenge_module": "Challenge",
+        "solution_module": "Submission",
+        "theorem_names": config["theorem_names"],
+        "definition_names": config["definition_names"],
+        "permitted_axioms": config["permitted_axioms"],
+        "enable_nanoda": False,
+    }
+    (run / "config.json").write_text(json.dumps(comparator_config, indent=2))
+    lake = run / ".lake"
+    lake.mkdir()
+    (lake / "packages").symlink_to(LEAN_ENV / ".lake" / "packages")
+    for p in run.iterdir():
+        if p.name != ".lake":
+            os.chmod(p, 0o644 if p.is_file() else 0o755, follow_symlinks=False)
+    uid = pwd.getpwnam(GRADER).pw_uid
+    os.chown(lake, uid, uid)
+    os.chmod(run, 0o755)
+
+
+def fresh_run(task: dict) -> Path:
+    """A new, empty Lake project for one grading attempt."""
+    run = Path(tempfile.mkdtemp(prefix="run.", dir=RUN_ROOT))
+    write_project(run, task["comparator"])
+    return run
+
+
+def ensure_readable(roots: tuple[str, ...] = ("/opt/elan", str(LEAN_ENV))) -> int:
+    """Make the prebuilt toolchain and packages readable by the grader (never writable).
+
+    The image already does this; this only repairs an image built without it. It must touch
+    nothing that is already right: these files live in the image's lower overlay layer, and a
+    `chmod` on one copies it whole into the sandbox's writable layer. A blanket `chmod -R` took
+    535 s and used 8.7 GB of the 10 GB working disk. This walk only reads metadata (lstat).
+    Returns the number of entries changed.
+    """
+    changed = 0
+    for root in roots:
+        for dirpath, dirnames, filenames in os.walk(root, followlinks=False):
+            for name in [*dirnames, *filenames]:
+                path = os.path.join(dirpath, name)
+                st = os.lstat(path)
+                if stat.S_ISDIR(st.st_mode):
+                    want = (st.st_mode | 0o555) & ~0o022
+                elif stat.S_ISREG(st.st_mode):
+                    want = (st.st_mode | 0o444) & ~0o022
+                else:
+                    continue
+                if stat.S_IMODE(st.st_mode) != stat.S_IMODE(want):
+                    os.chmod(path, stat.S_IMODE(want))
+                    changed += 1
+    return changed
+
+
+def block_grader_network() -> None:
+    """Reject every socket the grader user opens, loopback included.
+
+    landrun already denies TCP to the sandboxed build. This rule also covers the grader's own
+    processes outside landrun and services listening inside the sandbox.
+    """
+    uid = str(pwd.getpwnam(GRADER).pw_uid)
+    for tool in ("iptables", "ip6tables"):
+        rule = ["OUTPUT", "-m", "owner", "--uid-owner", uid, "-j", "REJECT"]
+        if subprocess.run([tool, "-C", *rule], capture_output=True).returncode == 0:
+            continue
+        res = subprocess.run([tool, "-I", *rule], capture_output=True, text=True)
+        if res.returncode != 0:
+            raise GradingError(f"cannot install the {tool} rule for {GRADER}: {res.stderr.strip()}")
+    probe = subprocess.run(
+        ["runuser", "-u", GRADER, "--", sys.executable, "-c",
+         "import socket,sys\n"
+         "s=socket.socket(); s.settimeout(3)\n"
+         "sys.exit(0 if s.connect_ex(('127.0.0.1', 2280)) == 0 else 1)"],
+        capture_output=True,
+    )
+    if probe.returncode == 0:
+        raise GradingError("the grader user can still open loopback connections")
+
+
+def kill_grader() -> None:
+    for _ in range(20):
+        subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True)
+        if subprocess.run(["pgrep", "-u", GRADER], capture_output=True).returncode != 0:
+            return
+        time.sleep(0.5)
+    raise GradingError("could not stop the grader's processes")
+
+
+def out_of_memory(out: str) -> bool:
+    return any(m in out for m in OOM_MARKERS)
+
+
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
+def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
+    env = {
+        "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
+        "HOME": str(run),
+        "LANG": "C.UTF-8",
+        "LEAN_ABORT_ON_PANIC": "1",
+        # Lean sees every host CPU, not the sandbox's 4; bound Lake's parallel builds so a large
+        # proof fits in 8 GB. Needs the image's Comparator, which passes this into landrun.
+        "LEAN_NUM_THREADS": threads,
+    }
+    cmd = [
+        "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
+        "lake", "env", "comparator", "config.json",
+    ]
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
+    return code, stdout + "\n--- stderr ---\n" + stderr
+
+
+def run_with_timeout(cmd: list[str], cwd: Path, timeout_sec: int, kill_extra=None) -> tuple[int | None, str, str]:
+    """Run ``cmd``; on timeout kill its whole process tree and return code None.
+
+    `timeout(1)` only signals its direct child, so Lean processes started by Lake under landrun
+    kept the output pipe open and the grader waited past Harbor's own limit. Here the command
+    gets its own session; on timeout its process group is killed, and ``kill_extra`` (every
+    process of the grader user) catches anything that left the group.
+    """
+    proc = subprocess.Popen(cmd, cwd=cwd, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
+                            text=True, errors="replace", start_new_session=True)
+    try:
+        stdout, stderr = proc.communicate(timeout=timeout_sec)
+        return proc.returncode, stdout, stderr
+    except subprocess.TimeoutExpired:
+        try:
+            os.killpg(proc.pid, signal.SIGKILL)
+        except ProcessLookupError:
+            pass
+        if kill_extra:
+            kill_extra()
+        stdout, stderr = proc.communicate()
+        return None, stdout, stderr
+
+
+def classify(code: int | None, out: str) -> tuple[int | None, str]:
+    """Map Comparator's result to (reward, reason). reward None means a grading error."""
+    if code == 0 and SUCCESS_LINE in out:
+        return 1, "accepted: statements match, only permitted axioms, kernel replay passed"
+    if "Building Submission" not in out:
+        # Comparator stopped before it touched the submission: the trusted challenge or the
+        # toolchain is broken (or too slow), which says nothing about the agent's proof.
+        return None, ("timed out before the submission was built" if code is None
+                      else "the trusted Challenge failed to build or export")
+    if code is None:
+        return 0, "timed out while building or checking the submission"
+    tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
+    if out_of_memory(out):
+        return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
+    return 0, "rejected: " + " | ".join(tail)[-4000:]
+
+
+def main() -> int:
+    VERIFIER_DIR.mkdir(parents=True, exist_ok=True)
+    os.chmod(VERIFIER_DIR, 0o755)
+    for name in ("reward.txt", "reward.json"):
+        (VERIFIER_DIR / name).unlink(missing_ok=True)
+
+    task = json.loads((TESTS / "task.json").read_text())
+    details: dict = {"task": task["challenge"], "started_at": time.time()}
+    reward: int | None = None
+    log = ""
+    try:
+        RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
+        deadline = time.time() + int(task["grade_timeout_sec"])
+        run = fresh_run(task)
+        details["permission_fixes"] = ensure_readable()
+        block_grader_network()
+        details["submission"] = unpack_submission(run)
+        code, out = run_comparator(run, int(deadline - time.time()))
+        log = out
+        kill_grader()
+        if code not in (0, None) and out_of_memory(out) and deadline - time.time() > 60:
+            # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
+            # in a fresh directory: the first attempt's .lake was written by the submission's own
+            # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
+            run = fresh_run(task)
+            unpack_submission(run)
+            code, out = run_comparator(run, int(deadline - time.time()), threads="1")
+            log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
+            kill_grader()
+            details["retried_single_thread"] = True
+        reward, details["reason"] = classify(code, out)
+        details["comparator_exit"] = code
+    except Rejected as exc:
+        reward, details["reason"] = 0, f"rejected: {exc}"
+    except GradingError as exc:
+        reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
+    finally:
+        try:
+            kill_grader()
+        except GradingError as exc:
+            reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
+    details["reward"] = reward
+    details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
+    (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
+    print(json.dumps(details, indent=2))
+    if reward is not None:
+        (VERIFIER_DIR / "reward.txt").write_text(f"{reward}\n")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/tasks/atomic-gaussian/tests/task.json b/tasks/atomic-gaussian/tests/task.json
new file mode 100644
index 0000000000000000000000000000000000000000..a055934acab737e7e1879becb6ca3679f453b79d
--- /dev/null
+++ b/tasks/atomic-gaussian/tests/task.json
@@ -0,0 +1,15 @@
+{
+  "challenge": "AtomicGaussian",
+  "comparator": {
+    "theorem_names": [
+      "OAI.AtomicTriangular.sharp_gaussian_minorants_atomic"
+    ],
+    "definition_names": [],
+    "permitted_axioms": [
+      "propext",
+      "Quot.sound",
+      "Classical.choice"
+    ]
+  },
+  "grade_timeout_sec": 18474
+}
\ No newline at end of file
diff --git a/tasks/atomic-gaussian/tests/test.sh b/tasks/atomic-gaussian/tests/test.sh
new file mode 100644
index 0000000000000000000000000000000000000000..172bae347e1bef3e7b6f12e7db887e33034b1130
--- /dev/null
+++ b/tasks/atomic-gaussian/tests/test.sh
@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+# Runs in the separate verifier sandbox, which never ran agent code. See grade.py.
+exec python3 /tests/grade.py
diff --git a/tasks/auslander-reiten/task.toml b/tasks/auslander-reiten/task.toml
index 7c705b4350a3e8a0040dec3a3224cc954c482f83..3b1dca317e7e1382c8e1e657d56b57cbaeec1c8d 100644
--- a/tasks/auslander-reiten/task.toml
+++ b/tasks/auslander-reiten/task.toml
@@ -23,7 +23,7 @@ reference_modules = 732
 reference_lean_bytes = 2349871
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 2615
+oracle_grade_seconds = 2177
 difficulty = "research"
 
 [agent]
diff --git a/tasks/auslander-reiten/tests/grade.py b/tasks/auslander-reiten/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/auslander-reiten/tests/grade.py
+++ b/tasks/auslander-reiten/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/backward-intertwiners/task.toml b/tasks/backward-intertwiners/task.toml
index f62e01de2f250bccd5c31bf54e192eb199f8b5b7..d9021e0e6e38199a8b92b1f71ff3814df8e0122c 100644
--- a/tasks/backward-intertwiners/task.toml
+++ b/tasks/backward-intertwiners/task.toml
@@ -23,7 +23,7 @@ reference_modules = 23
 reference_lean_bytes = 262844
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 823
+oracle_grade_seconds = 337
 difficulty = "research"
 
 [agent]
diff --git a/tasks/backward-intertwiners/tests/grade.py b/tasks/backward-intertwiners/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/backward-intertwiners/tests/grade.py
+++ b/tasks/backward-intertwiners/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/balanced-box-routing/task.toml b/tasks/balanced-box-routing/task.toml
index d45536988c10a57179803f38edaf6d89d3bd7fec..2f5ed9beb045d1826904b77d6a6552407645252c 100644
--- a/tasks/balanced-box-routing/task.toml
+++ b/tasks/balanced-box-routing/task.toml
@@ -23,7 +23,7 @@ reference_modules = 7
 reference_lean_bytes = 164970
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 299
+oracle_grade_seconds = 141
 difficulty = "research"
 
 [agent]
diff --git a/tasks/balanced-box-routing/tests/grade.py b/tasks/balanced-box-routing/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/balanced-box-routing/tests/grade.py
+++ b/tasks/balanced-box-routing/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/balanced-ryser/task.toml b/tasks/balanced-ryser/task.toml
index 5322f1252c09da27326521c2792424caa2ace634..9ed3930fccd9d1531a7d7e4203a7f953637e23eb 100644
--- a/tasks/balanced-ryser/task.toml
+++ b/tasks/balanced-ryser/task.toml
@@ -23,7 +23,7 @@ reference_modules = 698
 reference_lean_bytes = 3897149
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 1733
+oracle_grade_seconds = 1384
 difficulty = "research"
 
 [agent]
diff --git a/tasks/balanced-ryser/tests/grade.py b/tasks/balanced-ryser/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/balanced-ryser/tests/grade.py
+++ b/tasks/balanced-ryser/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/balanced-three-stack/task.toml b/tasks/balanced-three-stack/task.toml
index 375a51d06f12ebfac34332a204b3b32c94d7475a..4051a79adeed758f6263aeb3432c77f39b4fe7c9 100644
--- a/tasks/balanced-three-stack/task.toml
+++ b/tasks/balanced-three-stack/task.toml
@@ -23,7 +23,7 @@ reference_modules = 60
 reference_lean_bytes = 495983
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 1956
+oracle_grade_seconds = 392
 difficulty = "research"
 
 [agent]
diff --git a/tasks/balanced-three-stack/tests/grade.py b/tasks/balanced-three-stack/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/balanced-three-stack/tests/grade.py
+++ b/tasks/balanced-three-stack/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None:
diff --git a/tasks/ball-packing/task.toml b/tasks/ball-packing/task.toml
index 318e817bb9e7b347e190550445e4dcb41e9570a7..6e61e1c99f2c6188a267b0c0b47cfe0c614999a7 100644
--- a/tasks/ball-packing/task.toml
+++ b/tasks/ball-packing/task.toml
@@ -23,7 +23,7 @@ reference_modules = 169
 reference_lean_bytes = 4529811
 reference_requires_packages = []
 oracle_result = "accepted"
-oracle_grade_seconds = 4562
+oracle_grade_seconds = 2433
 difficulty = "research"
 
 [agent]
diff --git a/tasks/ball-packing/tests/grade.py b/tasks/ball-packing/tests/grade.py
index 9f4ee2489c3ce97bcc183d7695059602472bf911..95f89f56da7f59e7af65dcf71b8ef3792ee87399 100644
--- a/tasks/ball-packing/tests/grade.py
+++ b/tasks/ball-packing/tests/grade.py
@@ -35,6 +35,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import threading
 import time
 from pathlib import Path
 
@@ -46,6 +47,7 @@ RUN_ROOT = Path("/var/lib/oai-grade")
 GRADER = "grader"
 SUCCESS_LINE = "Your solution is okay!"
 GRADE_THREADS = "2"
+SCRATCH_SWEEP_SEC = 15
 # Lake and Comparator wording when a build process is killed for running out of memory.
 OOM_MARKERS = ("Lean exited with code 137", "Child exited with 137")
 
@@ -240,6 +242,49 @@ def out_of_memory(out: str) -> bool:
     return any(m in out for m in OOM_MARKERS)
 
 
+def out_of_disk(out: str) -> bool:
+    return "no space left on device" in out.lower()
+
+
+# Run as the grader user by `empty_build_scratch`. argv: the .lake/build/ir directory.
+EMPTY_SCRATCH = r"""
+import os, sys
+for dirpath, _, files in os.walk(sys.argv[1]):
+    for name in files:
+        if not name.endswith(".setup.json"):
+            continue
+        base = os.path.join(dirpath, name[:-len(".setup.json")])
+        try:
+            # Lake writes X.c.hash once Lean, and leanir if the module needs it, are done with X;
+            # both read X.setup.json, so it may only be emptied after that.
+            if os.stat(base + ".c.hash").st_mtime < os.stat(base + ".setup.json").st_mtime:
+                continue
+            for path in (base + ".setup.json", base + ".c"):
+                if os.path.getsize(path):
+                    os.truncate(path, 0)
+        except OSError:
+            pass
+"""
+
+
+def empty_build_scratch(run: Path, stop: threading.Event) -> None:
+    """Until ``stop`` is set, empty files Lake leaves in .lake/build/ir that grading never reads.
+
+    For every module Lake writes a `.setup.json` (its imports with the path of every .olean they
+    pull in, about 5.5 MB with Mathlib) and the module compiled to C. Neither is used once the
+    module is built, and for proofs of 1,000-2,000 modules they filled the 10 GB disk before the
+    build finished. Lake keeps the recorded hashes, so emptied files do not trigger a rebuild.
+    This runs as the grader user, so it can touch nothing the build could not.
+    """
+    ir = str(run / ".lake" / "build" / "ir")
+    while not stop.wait(SCRATCH_SWEEP_SEC):
+        try:
+            subprocess.run(["runuser", "-u", GRADER, "--", sys.executable, "-c", EMPTY_SCRATCH, ir],
+                           capture_output=True, timeout=600)
+        except subprocess.TimeoutExpired:
+            pass
+
+
 def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) -> tuple[int | None, str]:
     env = {
         "PATH": f"{toolchain_bin()}:/usr/local/bin:/usr/bin:/bin",
@@ -254,10 +299,17 @@ def run_comparator(run: Path, timeout_sec: int, threads: str = GRADE_THREADS) ->
         "runuser", "-u", GRADER, "--", "env", "-i", *(f"{k}={v}" for k, v in env.items()),
         "lake", "env", "comparator", "config.json",
     ]
-    code, stdout, stderr = run_with_timeout(
-        cmd, cwd=run, timeout_sec=timeout_sec,
-        kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
-    )
+    stop = threading.Event()
+    sweeper = threading.Thread(target=empty_build_scratch, args=(run, stop), daemon=True)
+    sweeper.start()
+    try:
+        code, stdout, stderr = run_with_timeout(
+            cmd, cwd=run, timeout_sec=timeout_sec,
+            kill_extra=lambda: subprocess.run(["pkill", "-9", "-u", GRADER], capture_output=True),
+        )
+    finally:
+        stop.set()
+        sweeper.join()
     return code, stdout + "\n--- stderr ---\n" + stderr
 
 
@@ -297,6 +349,8 @@ def classify(code: int | None, out: str) -> tuple[int | None, str]:
     if code is None:
         return 0, "timed out while building or checking the submission"
     tail = [l for l in out.splitlines() if l.strip() and "has local changes" not in l][-25:]
+    if out_of_disk(out):
+        return 0, "rejected: the build ran out of disk space (10 GB) | " + " | ".join(tail)[-3000:]
     if out_of_memory(out):
         return 0, "rejected: the build was killed, most likely out of memory (8 GB) | " + " | ".join(tail)[-3000:]
     return 0, "rejected: " + " | ".join(tail)[-4000:]
@@ -311,6 +365,7 @@ def main() -> int:
     task = json.loads((TESTS / "task.json").read_text())
     details: dict = {"task": task["challenge"], "started_at": time.time()}
     reward: int | None = None
+    log = ""
     try:
         RUN_ROOT.mkdir(mode=0o711, parents=True, exist_ok=True)
         deadline = time.time() + int(task["grade_timeout_sec"])
@@ -325,26 +380,33 @@ def main() -> int:
             # Two Lean processes in parallel can exceed 8 GB on a heavy module. Retry once with one,
             # in a fresh directory: the first attempt's .lake was written by the submission's own
             # build and must not be trusted for the challenge.
+            shutil.rmtree(run, ignore_errors=True)
             run = fresh_run(task)
             unpack_submission(run)
             code, out = run_comparator(run, int(deadline - time.time()), threads="1")
             log += "\n\n=== retried with one Lean process after running out of memory ===\n\n" + out
             kill_grader()
             details["retried_single_thread"] = True
-        (VERIFIER_DIR / "comparator.log").write_text(log)
         reward, details["reason"] = classify(code, out)
         details["comparator_exit"] = code
     except Rejected as exc:
         reward, details["reason"] = 0, f"rejected: {exc}"
     except GradingError as exc:
         reward, details["reason"] = None, f"grading error: {exc}"
+    except Exception as exc:
+        reward, details["reason"] = None, f"grading error: {type(exc).__name__}: {exc}"
     finally:
         try:
             kill_grader()
         except GradingError as exc:
             reward, details["reason"] = None, f"grading error: {exc}"
+        # Free the build before writing anything. A build that filled the disk would otherwise
+        # make these writes fail, and with no space left the sandbox cannot report that the
+        # verifier exited, so Harbor waits until its own timeout and records no verdict.
+        shutil.rmtree(RUN_ROOT, ignore_errors=True)
     details["reward"] = reward
     details["finished_at"] = time.time()
+    (VERIFIER_DIR / "comparator.log").write_text(log)
     (VERIFIER_DIR / "grading.json").write_text(json.dumps(details, indent=2))
     print(json.dumps(details, indent=2))
     if reward is not None: