CreditVintage v0.3.1

CreditVintage 0.3.1

This patch delivers the CSV-header fix and explicit code licensing in an
installable release. Existing 0.3.0 tags and assets remain unchanged.

Changes from 0.3.0

• check-inputs now catches an unterminated quoted CSV header in any of the
  three input tables. It returns invalid_inputs JSON on stdout, a
  csv_quoting issue at the physical end line of the malformed header and exit
  code 2, and continues checking the other input files. A single-line broken
  header points to line 1; a multi-line header points to its last physical line.
  Version 0.3.0 instead raises a CSV parsing error for
  this case. The checker still does not fit a model or modify inputs.
• Project code is explicitly MIT licensed. Both original UCI dataset source
  notices and their CC BY 4.0 terms are retained. The wheel and source archive
  include the MIT license and both notices; their combined distribution
  metadata reads MIT AND CC-BY-4.0.
• Building from source now requires setuptools 77 or later to support that
  license metadata. This is a build requirement; numerical runtime dependency
  versions have not changed.

No commands or input fields have been renamed. There is no input-schema or
model migration from 0.3.0. Evaluation rules, frozen data, numerical dependency
pins and report verification tolerances are unchanged. The existing research
limitations still apply: synthetic event data and historical random-split
public benchmarks do not establish a deployable credit policy.

Install or upgrade

In a Python 3.12+ virtual environment, install the exact release wheel:

python -m pip install --upgrade https://github.com/dev-belly/CreditVintage/releases/download/v0.3.1/creditvintage_research-0.3.1-py3-none-any.whl
python -c "from importlib.metadata import version; print(version('creditvintage-research'))"
python -m creditvintage init-inputs --out inputs/check-0.3.1
python -m creditvintage check-inputs inputs/check-0.3.1

On Windows, replace python with .\.venv\Scripts\python.exe when using an
environment named .venv. Create it first with python -m venv .venv if needed.
On Linux/macOS use .venv/bin/python for the same environment.

The version check prints 0.3.1; the checker returns
status: ready_for_evaluation and model_fitted: false. The starter contains
480 explicitly synthetic applications. Choose a new output directory on each
run because the starter never overwrites an existing path.

Release assets come from the successful main CI run, after clean wheel
installation, complete offline replay and cross-runtime card-data refitting.
release-manifest.json records the exact commit, version and file digests;
CHECKSUMS.sha256 covers the wheel, source archive and accompanying evidence.

CreditVintage v0.3.0

Adds init-inputs and check-inputs for editable CSV workflows. The starter writes 480 explicitly synthetic applications, complete feature/performance histories and calendar configuration. Diagnostics locate row errors, duplicate records, unknown references, missing decision-time features and incomplete mature outcomes by file and line, without fitting or modifying the inputs. Invalid data returns structured JSON and exit code 2.

Clean installed-package checks generate the CSVs, diagnose them, run the actual supplied-file evaluation, verify its 120 held-out applications and confirm that an intentionally invalid principal is reported at the correct file and line. Readiness validates the documented event/evaluation contract, not source authenticity or model quality.

The existing offline benchmark on 30,000 attributed UCI credit-card records retains five frozen splits, four methods and every model score and equal-budget result.

Numerical dependency versions are pinned for replay. CI also transfers the same Linux Python 3.12 evidence and wheel to Python 3.13 and Windows for independent refitting. A mismatched numerical runtime reports the recorded versions before fitting; score and metric tolerances remain unchanged.

The v2 card protocol uses Newton-Cholesky logistic fitting with a fixed 1e-10 convergence tolerance and a six-decimal input grid for isotonic calibration to prevent amplification of platform-dependent logistic tail differences; raw logistic scores are retained. Older v1 evidence keeps its original calibration protocol.

These are random partitions of one historical cohort, not future-period or point-in-time validation. The native subsequent-default target remains distinct from the synthetic 180-day DPD target.

Assets are the exact wheel and source archive that passed clean installed-package replay in the linked successful main CI run. release-manifest.json records the commit, version and digests; CHECKSUMS.sha256 covers all other assets. Dataset attribution and CC BY 4.0 terms are bundled with the data.

CreditVintage v0.2.1

Adds an offline benchmark on 30,000 attributed UCI credit-card records, with five frozen train/calibration/test splits, four fixed methods and three equal review budgets. Every score, split, model result, label-free queue and source notice can be downloaded and refitted.

Numerical dependency versions are pinned for replay. CI also transfers the same Linux Python 3.12 evidence and wheel to Python 3.13 and Windows for independent refitting. A mismatched numerical runtime reports the recorded versions before fitting; score and metric tolerances remain unchanged.

The v2 card protocol uses Newton-Cholesky logistic fitting with a fixed 1e-10 convergence tolerance and a six-decimal input grid for isotonic calibration to prevent amplification of platform-dependent logistic tail differences; raw logistic scores are retained. Older v1 evidence keeps its original calibration protocol.

These are random partitions of one historical cohort, not future-period or point-in-time validation. The native subsequent-default target remains distinct from the synthetic 180-day DPD target.

Assets are the exact wheel and source archive that passed clean installed-package replay in the linked successful main CI run. release-manifest.json records the commit, version and digests; CHECKSUMS.sha256 covers all other assets. Dataset attribution and CC BY 4.0 terms are bundled with the data.

CreditVintage v0.2.0

Adds an offline benchmark on 30,000 attributed UCI credit-card records, with five frozen train/calibration/test splits, four fixed methods and three equal review budgets. Every score, split, model result, label-free queue and source notice can be downloaded and refitted.

These are random partitions of one historical cohort, not future-period or point-in-time validation. The native subsequent-default target remains distinct from the synthetic 180-day DPD target.

Assets are the exact wheel and source archive that passed clean installed-package replay in the linked successful main CI run. release-manifest.json records the commit, version and digests; CHECKSUMS.sha256 covers all other assets. Dataset attribution and CC BY 4.0 terms are bundled with the data.