Problem

to_script(minimal=False) produces a script that runs fine and looks correct, but the schema it rebuilds is unusable — every validate() call on it raises TypeError: 'dict' object is not callable, including on data the original schema accepts:

import pandas as pd
from pandera.pandas import DataFrameSchema, Column, Check

schema = DataFrameSchema({"a": Column(int)}, checks=[Check.gt(0)])
ns = {}
exec(schema.to_script(minimal=False), ns)
rt = ns["schema"]

print(rt.checks)
# [{'min_value': 0, 'options': {'check_name': 'greater_than', ...}}]   <- dicts, not Checks
print(schema == rt)
# False

schema.validate(pd.DataFrame({"a": [5]}))   # passes, as it should
rt.validate(pd.DataFrame({"a": [5]}))       # SchemaError: Error while executing check
                                            # function: TypeError("'dict' object is not callable")

Column-level checks round-trip correctly, so the two paths disagree:

s2 = DataFrameSchema({"a": Column(int, Check.gt(0))})   # same check, column level
ns2 = {}; exec(s2.to_script(minimal=False), ns2)
print(ns2["schema"].columns["a"].checks)   # [<Check greater_than: greater_than(0)>]
print(s2 == ns2["schema"])                 # True

The failure is quiet at the point of corruption: the generated script is valid Python and the DataFrameSchema(...) call succeeds. It only surfaces later, as a confusing TypeError from inside validation that gives no hint that to_script was the culprit.

Cause

In to_script the column path formats its checks, but the dataframe-level path passes the raw statistics list straight into the template — 19 lines apart in the same function:

column_code = COLUMN_TEMPLATE.format(
    checks=_format_checks(properties["checks"]),   # line 820 — formatted
    ...
)
...
script = SCRIPT_TEMPLATE.format(
    checks=statistics["checks"],                   # line 841 — raw dicts
    ...
)

Every other caller of the template data (_format_index, _format_column_minimal, the column path) goes through _format_checks; this one line is the exception.

Fix

Format dataframe-level checks the same way the column path does. _format_checks(None) already returns "None", so schemas without dataframe-level checks are unaffected.

After the fix, rt.checks is [<Check greater_than: greater_than(0)>], schema == rt is True, and the round-tripped schema accepts and rejects exactly what the original does.

Testing

• Added test_to_script_dataframe_level_checks to tests/io/test_pandas_io.py. The existing test_to_script asserts the right invariant (schema == schema_to_write) but its _create_schema() fixture defines only column checks, so this path had no coverage. Verified the new test fails before the fix and passes after.
• tests/io/ → 73 passed. The 5 test_frictionless_* failures are pre-existing: they fail identically on an unmodified main in my environment because frictionless isn't installed, and are unrelated to this change.
• tests/pandas/test_schema_statistics.py + tests/pandas/test_schema_inference.py → 80 passed, 1 skipped (to_script goes through get_dataframe_schema_statistics).
• ruff check (the linter CI runs) and ruff format --check clean on both changed files.

Note on #2402

Heads-up that #2402 moves to_script/SCRIPT_TEMPLATE/_format_checks into pandera/io/common_io.py and carries checks=statistics["checks"] across unchanged, so the bug would survive that refactor. Happy to rebase onto it, or to re-send against common_io.py instead if you'd rather land #2402 first — whichever is less work for you.

AI disclosure

Prepared with AI assistance (Claude Code). I reproduced the bug against the unmodified library, verified the fix and the test locally, and take responsibility for the change.