FinPanel v0.1.0-alpha.2

FinPanel v0.1.0-alpha.2

Python package version: 0.1.0a2. Python 3.12 is required. This is an experimental pre-release, not stable software.

Install from production PyPI in a clean Python 3.12 environment:

python -m pip install finpanel==0.1.0a2
python -m finpanel.example output/quickstart

Alpha.2 aligns public documentation and package metadata, and adds protected
TestPyPI rehearsal and production PyPI release infrastructure using OIDC.
Financial/accounting semantics, metric contracts, historical cutoffs, revision
selection and provenance behavior are unchanged from the validated A2 software.
The point-in-time panels, pinned evidence snapshots, explicit unresolved states,
CSV/Parquet/DuckDB exports and reproducibility receipts remain available.
Exact replay requires the matching software version and captured evidence.

Validation covers a specified 100-issuer US-GAAP population: 24,000 financial
cells, 255/255 independent golden expectations, and zero unexpected mismatches
within that population. Scalar coverage was 69.61%; this measures resolution
coverage, not accuracy. Exact-source CI passed 814 tests, with zero skips, on
macOS, Linux and Windows, including build and clean-install checks.

Authorized source: 28dcecce208a578b4a7b507bbf6d93018b4ab1d6.
Exact-source CI.
Production publishing workflow.
Production PyPI package.

Production PyPI publication used protected GitHub Actions Trusted Publishing/OIDC.
The wheel and sdist were independently downloaded from PyPI and matched this
workflow's recorded SHA-256 hashes. A fresh repository-external Python 3.12
installation from production PyPI passed CLI, public examples, small-panel,
CSV/Parquet/DuckDB, provenance and receipt/replay checks. The 69 runtime files
are byte-identical to the TestPyPI-validated A2; the complete archives differ
because documentation and associated metadata changed.

Attached distributions, checksums and build provenance are the exact production
workflow outputs. They were not rebuilt locally for this GitHub Release.

This is experimental alpha software, not a claim of universal correctness.
Banks, insurers, IFRS normalization, segment aggregation, FX normalization,
prices and returns remain outside scope. Original-XBRL, extension-heavy and
dimension-heavy evidence coverage remains incomplete.
See validation evidence.

The packaged README preserves the preparation-time sentence "Production PyPI:
not published"; this is now stale. The production PyPI version page and this
release record confirm actual publication. No published distribution was
replaced or modified to change that historical text.

The existing a1 tag, GitHub release and assets remain unchanged. A1 was never
published to production PyPI; this is the first production PyPI version.

FinPanel v0.1.0-alpha

FinPanel v0.1.0-alpha

Research infrastructure for reproducible point-in-time SEC fundamentals.

This first public alpha targets reproducible research on ordinary US-GAAP operating companies. Package version: 0.1.0a1; Python 3.12. Release commit: 30ffbbe658fbc517ec838ba499859ec05093283b.

Installation

No repository clone is needed. Use Python 3.12 and download the attached
finpanel-0.1.0a1-py3-none-any.whl; also download SHA256SUMS.txt for the
recommended checksum check. In the download folder, print the wheel hash with
shasum -a 256 finpanel-0.1.0a1-py3-none-any.whl (macOS),
sha256sum finpanel-0.1.0a1-py3-none-any.whl (Linux), or
Get-FileHash finpanel-0.1.0a1-py3-none-any.whl -Algorithm SHA256 (PowerShell).
Manually compare it with the wheel's row in SHA256SUMS.txt; stop on a mismatch.
Hash comparison is not automatic.

Create and activate a fresh Python 3.12 virtual environment using the
platform-specific instructions,
then run:

python -m pip install ./finpanel-0.1.0a1-py3-none-any.whl
python -c "import finpanel; print(finpanel.__version__)"
python -m finpanel.example output/quickstart

Expected version: 0.1.0a1. Choose a new output directory. The bundled example
needs no SEC credentials, live SEC access or private evidence corpus. Dependencies
may be downloaded during installation. FinPanel has not been published to PyPI.

Capabilities

• Historical SEC filing timelines and explicit as-of information boundaries.
• Revision-aware fundamentals with first-reported, latest-available and all-available contracts.
• Six canonical metrics: revenue, net income, assets, liabilities, cash and cash equivalents, and operating cash flow.
• Reported fundamentals and conservative derived quarterly fundamentals.
• Original XBRL provenance and context verification.
• Immutable evidence snapshots, deterministic receipts and replay.
• Multi-company research panels with CSV, Parquet and DuckDB exports.
• Resumable broad validation with explicit unresolved and conflict states.

Validation and limits of the evidence

The frozen, purposively selected validation population contains 100 audited ordinary US-GAAP operating companies and 24,000 financial cells:

• 255/255 independently audited golden expectations passed.
• 2,435,238 invariant checks and 500 future-evidence mutation checks passed.
• Zero unexpected mismatches within this specified audited validation population.
• Scalar coverage: 16,706 / 24,000 = 69.61%. Coverage is NOT accuracy: it measures the proportion of requested cells that resolved to scalar values. Unavailable, insufficient-evidence, ambiguous, conflicted and unsupported results remain explicit.

These results do not establish universal accuracy or support for all U.S. companies. Golden expectations count distinct cutoff/policy contracts, not necessarily unique economic cells.

Release CI for this exact commit passed on macOS (Python 3.12.10), Linux (3.12.14) and Windows (3.12.10): 814 tests passed, 0 skipped on each platform, with Ruff, formatting, wheel/sdist builds and clean installations. Final artifact checks also passed for offline public APIs, examples, panels, three export formats, provenance, receipt/replay and semantic reproducibility.

The broader audit methodology and coverage breakdown are in the validation report; its earlier WP7 test totals are historical. The approximately 506 MiB Tier B evidence corpus remains local and is not distributed.

Unsupported or partial scope

• Banks and insurers are unsupported.
• IFRS normalization, segment aggregation and FX normalization are unsupported.
• Historical and original-XBRL evidence is incomplete for some issuers.
• Dimension-heavy or extension-heavy disclosures may remain conservatively unresolved.
• SEC aggregate captures do not establish historical API delivery time.

This is an experimental pre-release, not a stable production release. See the alpha API stability policy.

Artifacts

Attached: the 0.1.0a1 wheel, source distribution and SHA256SUMS.txt. Verify checksums before installation. These assets were freshly built from the release commit and checked in separate clean environments. No package was published to PyPI.