v0.1.0 · Financial features with source lineage

A small Python/SQLite reference implementation for rebuilding financial features from data available at a historical decision.

Try it

• Linux/macOS and Windows quick starts: install in a virtual environment, generate an offline report and verify it.
• Runnable Python example: one synthetic January invoice is missing before arrival, selects 100,000 CNY in February, then 135,000 CNY after its April correction becomes available.
• Online snapshot report and rolling report.

Included

• Event, publication and ingestion cutoffs; highest known revision; scoped tombstones and freshness limits.
• SQLite selection checked against an independent Python enumerator.
• Rolling sum, observation mean and count with contributing record membership.
• Deterministic CSV/HTML evidence and hash plus semantic replay verification.
• Contribution guide and issue/PR templates, with one scoped optional pandas example task.

Verified

The release commit CI passed Linux Python 3.11/3.12 checks and Windows Python 3.12 installation and example replay. All 79 tests pass; a fresh local environment reproduces the three documented API rows and both report workflows. Python 3.11+ is required, with no third-party runtime dependencies.

Research scope

This is an experimental reference, not a production feature-store service. Demo records are synthetic; upstream timestamps and version identities must be trustworthy. No distributed-scale benchmark or realized lending/investment outcome is claimed. See the data contract and MIT license.

Bug reports and small counterexamples are welcome through the issue templates.