Clarify perfect-cuboid evidence schema version (#1424)

Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>

Author: yuelgrace1810-ops

benchmarks/datasets/conjecture-probes-v1/perfect-cuboid-scope-audit/instruction.md

@@ -18,9 +18,9 @@ be reported in any common order; the verifier compares the three aligned
 radicand/root pairs as an unordered set.
 
 `evidence/answer.txt` must be a JSON object with exactly `schema_version`,
-`task_id`, `result`, and `limitations`. Use schema version `1`, the task ID and
-limitations from the submission contract, and the same `result` object as in
-`submission.json`. The file must be no larger than 2 MiB.
+`task_id`, `result`, and `limitations`. Use schema version `1` (the JSON string
+`"1"`), the task ID and limitations from the submission contract, and the same
+`result` object as in `submission.json`. The file must be no larger than 2 MiB.
 
 This is a finite semantic-scope audit. An Euler brick is not a perfect cuboid,
 and finding no perfect cuboid in these twelve cases is not evidence of global

benchmarks/validation/conjecture_probes_v1/test_perfect_cuboid_scope_audit.py

@@ -13,6 +13,11 @@
 TASK = ROOT / "benchmarks/datasets/conjecture-probes-v1/perfect-cuboid-scope-audit"
 
 
+def test_public_instruction_specifies_schema_version_as_json_string() -> None:
+    instruction = (TASK / "instruction.md").read_text()
+    assert 'Use schema version `1` (the JSON string\n`"1"`)' in instruction
+
+
 def _case(tmp_path: Path) -> tuple[Path, Path, dict]:
     app = tmp_path / "app"
     logs = tmp_path / "logs"