Showcase¶
Three messy, real-world files converted end to end by the 2.0 GB student — no hand-written rules. For each case you'll see the input, the Python script the model wrote from the file's metadata envelope, the converted output, and its validation scores.
All three ran through the public API exactly as you would:
These are live model outputs (greedy decoding, so reproducible), not curated by hand. Notice the model picks up real-world details on its own — stripping a BOM, grouping rows into a nested array, flattening nested keys to dot-notation, and falling back gracefully on a log line that doesn't match the expected shape.
1. Ragged CSV → nested JSON¶
Input — orders.csv:
user_name,user_email,order_id,order_item,order_qty
Alice,alice@example.com,1001,Widget,3
Alice,alice@example.com,1002,Gadget,1
Bob,bob@example.com,1003,Gizmo,5
The student wrote this script (retries: 1):
import sys
import json
import pandas as pd
def main():
in_path = sys.argv[1]
out_path = sys.argv[2]
# Envelope encoding is utf-8-sig; read with it to strip the BOM.
df = pd.read_csv(in_path, encoding="utf-8-sig", dtype=str, keep_default_na=False)
# NUMERIC_COLUMN_QUOTE_RISK: keep these numeric in JSON, never stringify.
for col in ("order_id", "order_qty"):
df[col] = pd.to_numeric(df[col], errors="coerce").astype("Int64")
def clean(v):
if v is None:
return ""
if isinstance(v, float) and pd.isna(v):
return ""
if isinstance(v, str):
return v.strip()
return str(v)
users = []
index = {} # (user_name, user_email) -> position in `users`
for _, row in df.iterrows():
name = clean(row.get("user_name"))
email = clean(row.get("user_email"))
key = (name, email)
if key not in index:
index[key] = len(users)
users.append({
"user": {"name": name, "email": email},
"orders": [],
})
oid = row.get("order_id")
oitem = clean(row.get("order_item"))
oq = row.get("order_qty")
users[index[key]]["orders"].append({
"id": int(oid),
"item": oitem,
"q": int(oq),
})
with open(out_path, "w", encoding="utf-8") as f:
json.dump(users, f, indent=2, ensure_ascii=False)
if __name__ == "__main__":
main()
Converted output — orders.json:
[
{
"user": {
"name": "Alice",
"email": "alice@example.com"
},
"orders": [
{
"id": 1001,
"item": "Widget",
"q": 3
},
{
"id": 1002,
"item": "Gadget",
"q": 1
}
]
},
{
"user": {
"name": "Bob",
"email": "bob@example.com"
},
"orders": [
{
"id": 1003,
"item": "Gizmo",
"q": 5
}
]
}
]
✓ accepted
format_validity 1.0 · loadability 1.0
2. Nested JSON (optional keys) → CSV¶
Input — users.json:
[
{"id": 1, "name": "Alice", "address": {"city": "NYC", "zip": "10001"}},
{"id": 2, "name": "Bob", "address": {"city": "Boston"}},
{"id": 3, "name": "Carol", "address": {"city": "LA", "zip": "90001"}}
]
The student wrote this script (retries: 1):
import sys
import json
import pandas as pd
def main():
in_path = sys.argv[1]
out_path = sys.argv[2]
with open(in_path, "r", encoding="utf-8") as f:
data = json.load(f)
# Envelope says root_shape is a record_array; tolerate a single object too.
if isinstance(data, dict):
data = [data]
# Flatten nested keys into dot-notation columns (address.city, address.zip).
df = pd.json_normalize(data, sep=".")
# NUMERIC_COLUMN_QUOTE_RISK: keep id numeric so it is not stringified.
df["id"] = pd.to_numeric(df["id"], errors="coerce").astype("Int64")
# Keep address.zip as a string so leading zeros are preserved; coerce others.
df["address.zip"] = df["address.zip"].astype("string")
# Handle missing values (nulls, missing keys) by leaving them empty.
df.to_csv(out_path, index=False, encoding="utf-8")
if __name__ == "__main__":
main()
Converted output — users.csv:
✓ accepted
format_validity 1.0 · loadability 1.0
3. Semi-structured TXT log → CSV¶
Input — events.log:
[2026-01-05 08:12:01] INFO auth: user alice logged in
[2026-01-05 08:13:44] WARNING auth: failed login for bob
[2026-01-05 08:15:02] ERROR db: connection timeout
The student wrote this script (retries: 0):
import sys
import re
from datetime import datetime
import pandas as pd
LINE_RE = re.compile(
r"^\[?(\d{4}-\d{2}-\d{2})[ T](\d{2}:\d{2}:\d{2})\]?\s+"
r"(\S+)\s+"
r"([^:]+):\s?"
r"(.*)$"
)
def to_iso(date_str, time_str):
dt = datetime.strptime(f"{date_str} {time_str}", "%Y-%m-%d %H:%M:%S")
return dt.isoformat()
def main():
in_path = sys.argv[1]
out_path = sys.argv[2]
rows = []
with open(in_path, "r", encoding="utf-8-sig") as f:
for raw in f:
line = raw.rstrip("\n").rstrip("\r")
if not line.strip():
continue
m = LINE_RE.match(line)
if m:
date_str, time_str, level, source, message = m.groups()
rows.append(
{
"timestamp": to_iso(date_str, time_str),
"level": level,
"source": source,
"message": message,
}
)
else:
# Non-conforming line: preserve content, leave level/source empty.
rows.append(
{"timestamp": "", "level": "", "source": "", "message": line.strip()}
)
df = pd.DataFrame(rows, columns=["timestamp", "level", "source", "message"])
df.to_csv(out_path, index=False, encoding="utf-8")
if __name__ == "__main__":
main()
Converted output — events.csv:
timestamp,level,source,message
2026-01-05T08:12:01,INFO,auth,user alice logged in
2026-01-05T08:13:44,WARNING,auth,failed login for bob
2026-01-05T08:15:02,ERROR,db,connection timeout
✓ accepted
format_validity 1.0 · loadability 1.0