diff --git a/validate_submission.py b/validate_submission.py
new file mode 100644
index 000000000..44c51732d
--- /dev/null
+++ b/validate_submission.py
@@ -0,0 +1,263 @@
+#!/usr/bin/env python3
+"""
+Pre-submission validation for Parameter Golf records.
+
+Checks folder structure, required files, submission.json schema,
+train_gpt.py syntax, and artifact size before opening a PR.
+
+Usage:
+    python3 validate_submission.py records/track_10min_16mb/2026-03-24_MySubmission/
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import sys
+from pathlib import Path
+
+SIZE_CAP = 16_000_000  # 16 MB decimal, per challenge rules
+
+VALID_TRACKS = (
+    "records/track_10min_16mb",
+    "records/track_non_record_16mb",
+)
+
+# Fields that must be present and non-null in submission.json.
+# Each entry maps a canonical name to (accepted_aliases, expected_type).
+REQUIRED_FIELDS = {
+    "val_bpb": (["val_bpb", "mean_val_bpb"], (int, float)),
+    "name": (["name", "run_name"], str),
+    "date": (["date"], str),
+}
+
+# Fields that should be present but won't fail validation.
+RECOMMENDED_FIELDS = {
+    "author": str,
+    "github_id": str,
+    "blurb": str,
+}
+
+
+def check_folder_location(folder: Path, repo_root: Path) -> tuple[bool, str]:
+    """Verify the folder is under a valid records/ track."""
+    try:
+        rel = folder.resolve().relative_to(repo_root.resolve())
+    except ValueError:
+        return False, f"Folder is not inside the repository: {folder}"
+    rel_str = str(rel).replace(os.sep, "/")
+    for track in VALID_TRACKS:
+        if rel_str.startswith(track + "/"):
+            return True, f"Track: {rel_str.split('/')[1]}"
+    return False, (
+        f"Folder must be under {' or '.join(VALID_TRACKS)}, "
+        f"got: {rel_str}"
+    )
+
+
+def check_required_files(folder: Path) -> tuple[bool, list[str]]:
+    """Check that README.md, submission.json, and a train_gpt*.py exist."""
+    missing = []
+    if not (folder / "README.md").is_file():
+        missing.append("README.md")
+    if not (folder / "submission.json").is_file():
+        missing.append("submission.json")
+    # Accept train_gpt.py or variants like train_gpt_v5.py.
+    train_scripts = list(folder.glob("train_gpt*.py"))
+    if not train_scripts:
+        missing.append("train_gpt.py (or train_gpt_*.py)")
+    return len(missing) == 0, missing
+
+
+def check_train_log(folder: Path) -> tuple[bool, str]:
+    """Check for at least one train log file (.log, .tsv, or .txt)."""
+    logs = list(folder.glob("*.log")) + list(folder.glob("*.tsv")) + list(folder.glob("*.txt"))
+    if logs:
+        names = ", ".join(p.name for p in logs[:5])
+        extra = f" (+{len(logs) - 5} more)" if len(logs) > 5 else ""
+        return True, f"Found: {names}{extra}"
+    return False, "No .log, .tsv, or .txt file found (train log required)"
+
+
+def check_submission_json(folder: Path) -> tuple[bool, list[str]]:
+    """Validate submission.json schema."""
+    path = folder / "submission.json"
+    errors: list[str] = []
+    warnings: list[str] = []
+
+    try:
+        data = json.loads(path.read_text())
+    except json.JSONDecodeError as e:
+        return False, [f"Invalid JSON: {e}"]
+
+    if not isinstance(data, dict):
+        return False, ["submission.json must be a JSON object, not a list or scalar"]
+
+    # Required fields (with alias support).
+    for canonical, (aliases, expected_type) in REQUIRED_FIELDS.items():
+        found_key = None
+        for alias in aliases:
+            if alias in data:
+                found_key = alias
+                break
+        if found_key is None:
+            errors.append(f"Missing required field: {canonical} (also accepts: {', '.join(aliases[1:])})" if len(aliases) > 1 else f"Missing required field: {canonical}")
+        elif data[found_key] is None:
+            if isinstance(expected_type, tuple):
+                want = "/".join(t.__name__ for t in expected_type)
+            else:
+                want = expected_type.__name__
+            errors.append(f"Field '{found_key}' is null (must be {want})")
+        elif not isinstance(data[found_key], expected_type):
+            got = type(data[found_key]).__name__
+            if isinstance(expected_type, tuple):
+                want = "/".join(t.__name__ for t in expected_type)
+            else:
+                want = expected_type.__name__
+            errors.append(f"Field '{found_key}' should be {want}, got {got}")
+
+    # Recommended fields (warnings only).
+    for field, expected_type in RECOMMENDED_FIELDS.items():
+        if field not in data:
+            warnings.append(f"Recommended field missing: {field}")
+        elif data[field] is not None and not isinstance(data[field], expected_type):
+            warnings.append(f"Field '{field}' should be {expected_type.__name__}")
+
+    return len(errors) == 0, errors + [f"(warn) {w}" for w in warnings]
+
+
+def find_train_script(folder: Path) -> Path | None:
+    """Find the train_gpt*.py script in the submission folder."""
+    exact = folder / "train_gpt.py"
+    if exact.is_file():
+        return exact
+    variants = list(folder.glob("train_gpt*.py"))
+    return variants[0] if variants else None
+
+
+def check_syntax(folder: Path) -> tuple[bool, str]:
+    """Verify the training script has valid Python syntax."""
+    path = find_train_script(folder)
+    if path is None:
+        return False, "No train_gpt*.py found"
+    try:
+        compile(path.read_text(), str(path), "exec")
+        return True, f"OK ({path.name})"
+    except SyntaxError as e:
+        return False, f"Syntax error in {path.name} at line {e.lineno}: {e.msg}"
+
+
+def check_size(folder: Path) -> tuple[bool, str]:
+    """Report total folder size and warn if close to / over the 16 MB cap."""
+    total = sum(f.stat().st_size for f in folder.rglob("*") if f.is_file())
+    mb = total / 1_000_000
+    if total > SIZE_CAP:
+        return False, f"Total size {mb:.2f} MB exceeds {SIZE_CAP / 1_000_000:.0f} MB cap"
+    if total > SIZE_CAP * 0.95:
+        return True, f"Total size {mb:.2f} MB (close to {SIZE_CAP / 1_000_000:.0f} MB cap)"
+    return True, f"Total size {mb:.2f} MB"
+
+
+def find_repo_root(folder: Path) -> Path | None:
+    """Walk up from folder to find the repository root (contains README.md + records/)."""
+    current = folder.resolve()
+    for _ in range(10):
+        if (current / "README.md").is_file() and (current / "records").is_dir():
+            return current
+        parent = current.parent
+        if parent == current:
+            break
+        current = parent
+    return None
+
+
+def main() -> int:
+    if len(sys.argv) < 2:
+        print(f"Usage: {sys.argv[0]} <submission_folder>")
+        print(f"Example: {sys.argv[0]} records/track_10min_16mb/2026-03-24_MyRun/")
+        return 1
+
+    folder = Path(sys.argv[1])
+    if not folder.is_dir():
+        print(f"ERROR: Not a directory: {folder}")
+        return 1
+
+    repo_root = find_repo_root(folder)
+    if repo_root is None:
+        print("WARNING: Could not locate repository root. Skipping folder location check.")
+
+    passed = 0
+    failed = 0
+    warned = 0
+
+    checks: list[tuple[str, bool, str | list[str]]] = []
+
+    # 1. Folder location.
+    if repo_root:
+        ok, msg = check_folder_location(folder, repo_root)
+        checks.append(("Folder location", ok, msg))
+    else:
+        checks.append(("Folder location", True, "Skipped (repo root not found)"))
+
+    # 2. Required files.
+    ok, missing = check_required_files(folder)
+    if ok:
+        checks.append(("Required files", True, "README.md, submission.json, train_gpt.py"))
+    else:
+        checks.append(("Required files", False, f"Missing: {', '.join(missing)}"))
+
+    # 3. Train log.
+    ok, msg = check_train_log(folder)
+    checks.append(("Train log", ok, msg))
+
+    # 4. submission.json schema (only if file exists).
+    if (folder / "submission.json").is_file():
+        ok, messages = check_submission_json(folder)
+        detail = "; ".join(messages) if messages else "OK"
+        checks.append(("submission.json", ok, detail))
+    else:
+        checks.append(("submission.json", False, "File missing"))
+
+    # 5. Training script syntax (only if a train_gpt*.py exists).
+    if find_train_script(folder):
+        ok, msg = check_syntax(folder)
+        checks.append(("Training script syntax", ok, msg))
+    else:
+        checks.append(("Training script syntax", False, "No train_gpt*.py found"))
+
+    # 6. Size check.
+    ok, msg = check_size(folder)
+    checks.append(("Size check", ok, msg))
+
+    # Print results.
+    print(f"\nValidating: {folder}\n")
+    for name, ok, detail in checks:
+        if ok:
+            warn_parts = []
+            detail_str = detail if isinstance(detail, str) else "; ".join(detail)
+            if "(warn)" in detail_str:
+                status = "WARN"
+                warned += 1
+            else:
+                status = "PASS"
+                passed += 1
+        else:
+            status = "FAIL"
+            failed += 1
+            detail_str = detail if isinstance(detail, str) else "; ".join(detail)
+        print(f"  [{status}] {name}: {detail_str}")
+
+    print()
+    if failed > 0:
+        print(f"Result: FAIL ({failed} failed, {passed} passed, {warned} warnings)")
+        return 1
+    elif warned > 0:
+        print(f"Result: PASS with warnings ({passed} passed, {warned} warnings)")
+        return 0
+    else:
+        print(f"Result: PASS ({passed} passed)")
+        return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())