Sample
PyArrow CSV ingestion retains null tokens in string columns as literal strings unless strings_can_be_null is enabled in ConvertOptions, and raises KeyError for unlisted columns unless include_missing_columns is set
sha256:96f7e54f838dd5fd2cca97dba04561e3bf2ad30732280f3e451fb70e770b6729
PUBLISHED
L3_CONTRACT_PASS
MIT-0
Execution evidence
Declared environment and signed verification runs are separated so you can see exactly what this sample proves.
Evidence basisSigned contract pass
Verification receipts1
Verification levelL3_CONTRACT_PASS
Declared environment
- Execution context
- python
- Operating system
- linux
- Architecture
- x64
- Runtime
- python
- Language
- python
- Package manager
- pip
Verification-run environments
- Execution context
- python 3.12
- Operating system
- linux alpine · musl
- Architecture
- x64
- Runtime
- python 3.12
- Language
- python
- Package manager
- pip
- Execution
- container · docker
CONTAINER_RUN · compile:SKIPPED · contract:PASS · load:PASS · resolve:PASS · python@1 · 2026-08-17
Case
- Goal
- PyArrow CSV ingestion retains null tokens in string columns as literal strings unless strings_can_be_null is enabled in ConvertOptions, and raises KeyError for unlisted columns unless include_missing_columns is set HOW
- Packages
-
pyarrow 25.0.1
- Environment
- python
- Created
- 2026-08-17T10:42:07Z
Commonly assumed
pyarrow.csv.read_csv parses tokens defined in null_values as null values across all columns regardless of inferred column data type.
The sample's author recorded this as what a developer or model would expect here. The contract below is what actually ran.
Contract
- CSV columns inferred as strings preserve tokens matching null_values as literal strings unless strings_can_be_null is explicitly set to True, whereas numeric columns convert them to nulls by default.
- assert tbl_default['num'].null_count == 2 and tbl_default['txt'].null_count == 0
- assert tbl_nulls['txt'].to_pylist() == ['hello', None, 'world', None]
- assert tbl_no_quoted['txt'].to_pylist() == ['NA', 'hello', None]
- assert reading unlisted include_columns without include_missing_columns raises KeyError
- assert tbl_missing['extra_col'].type == pa.null() and tbl_typed_missing['extra_col'].type == pa.int32()
- assert isinstance(tbl_dict['category'].type, pa.DictionaryType)
- assert csv.read_csv raises ArrowInvalid on malformed rows unless ParseOptions(invalid_row_handler=...) returns 'skip'
- assert tbl['flag'].type == pa.bool_() and tbl['status'].to_pylist() == [None, 'PENDING', None]
Files
- NOTES.md
- csx.json
- requirements.lock
- requirements.txt
- test/contract.py
Download the source artifact (tar.gz)
Origin Seeder
csx-seed
Verification receipts
- python 3.12 · linux alpine/x64 · docker · CONTAINER_RUN · compile:SKIPPED · contract:PASS · load:PASS · resolve:PASS · python@1 · 2026-08-17 · ed25519:d91480838ac982c9