OpenMacroState v0.1.0a5 — Fed accounting audit

OpenMacroState v0.1.0a5 — Fed accounting audit

This pre-alpha release moves OpenMacroState from collecting official macro evidence to checking a real accounting state.

New

• fed-h41-release now extracts seven synchronized H.4.1 state variables: total assets, total liabilities, total capital, securities held outright, primary credit, the Treasury General Account, and reserve balances.

• New experimental command:

  oms audit accounting CASE_DIR \
    --rule fed-h41-balance-sheet-v1 \
    --observed-at 2023-03-15T00:00:00Z

• The fixed rule checks:

  • total assets = total liabilities + total capital, with a non-configurable 1 USD million rounding tolerance;
  • selected asset components remain within total assets; and
  • selected liability components remain within total liabilities.

• The audit re-reads the preserved local HTML with a 2 MiB bound, verifies its length and SHA-256, runs H.4.1 normalization again, reconstructs the seven core observation records, and requires exact matches before calculating.

• H.4.1 normalization is now ruleset /3, with stronger semantic table lineage and native hidden-state handling.

Why the tolerance is 1

Seven official H.4.1 pages spanning 2021-08-12 through 2026-08-06 were checked. Their accounting residuals are 0, -1, 0, 0, 0, 0, 0 USD million. The single -1 is the concrete source-rounding case; users cannot widen the tolerance.

Integrity boundaries

• Exact local replay proves that normalized values came from the preserved bytes under the named parser ruleset.
• It does not authenticate who acquired unverified recording bytes, when they first became public, or whether today's dated page is an original historical vintage.
• This remains an experimental report, not a stable accounting schema, general state graph, causal model, or forecasting system.
• Semantic HTML extraction is not a browser computed-CSS engine.

Verification

• 172 tests.
• Ruff and formatting checks.
• Python 3.10, 3.11, 3.12, and 3.13 CI.
• Offline SOFR, Treasury Debt to the Penny, H.4.1, synthetic replay, wheel, and source-distribution smoke tests.
• Release artifacts were built by the tag-triggered GitHub Actions workflow. Verify them with the attached SHA256SUMS.

Full change: PR #22 · Scope: Issue #21

OpenMacroState v0.1.0a4 — audited Fed H.4.1 capture

OpenMacroState now has a third audited official-source vertical slice: fed-h41-release.

What is new

• Captures one fixed, dated Federal Reserve Board H.4.1 HTML release.
• Emits five Wednesday balance-sheet observations in USD_million: total assets, securities held outright, primary credit, the Treasury General Account, and reserve balances.
• Binds exact semantic headings, units, table/header relationships, row labels, and the Wednesday stock column; structural drift fails closed.
• Hardens the core live HTTP transport against instance-level response injection.
• Adds an offline test_only_excerpt, source contract, license boundary, CI replay, and wheel/sdist smoke tests.

Evidence boundary

This remains a prospective-capture connector. A dated Fed URL is not proof that today’s bytes were available at the historical date. released_at, vintage_at, and current pre-alpha ingested_at therefore use the core capture/replay time; historical_evidence=false and source_published_at=null. The complete live HTML artifact is conservatively marked redistribution: restricted; the checked-in fixture contains only a reviewed Board-authored excerpt.

Verification

• 140 tests on Python 3.10–3.13
• Ruff check and format check
• Seven official H.4.1 page generations spanning 2021–2026
• Offline replay for SOFR (6 observations), Treasury Debt to the Penny (2), and H.4.1 (5)
• CI-built wheel installed and exercised outside the source checkout

OpenMacroState is still pre-alpha and is not on PyPI. Download the wheel or source archive below and verify it with SHA256SUMS.

Implemented in #20; closes #9.

OpenMacroState 0.1.0a3 — audited official-source captures

OpenMacroState’s first public pre-alpha release turns the repository from a contract skeleton into a small, executable macro time machine.

What works

• deterministic point-in-time case validation, frozen snapshots, physically separate reveal bundles, and Brier/log scoring
• a synthetic 2023-banks teaching fixture that deliberately tests future-information quarantine
• frbny-sofr: a bounded New York Fed SOFR capture with six normalized rate/distribution observations
• treasury-debt-to-penny: a fixed, single-page Treasury Fiscal Data capture for total public debt outstanding
• offline oms connector list [--json] discovery for the two review-trusted built-ins
• recorded replay, exact-byte hashing, content-addressed artifacts, strict output ownership, and source-specific licensing notices

Integrity boundaries

This release does not claim that present-day recordings prove what was available at an earlier historical cutoff. Recorded transport timestamps are unauthenticated claims; eligibility uses the core replay clock. Both official-source slices remain historical_evidence=false. Built-in review trust is not a sandbox for arbitrary third-party Python.

The connectors construct their own exact HTTPS endpoints, accept no arbitrary URL or secrets, reject redirects and malformed/incomplete pages, and fail closed on unsupported dates, schema drift, future values, or source-license ambiguity.

Breaking pre-alpha changes

• the connector contract now separates request planning from normalization and keeps networking, hashing, IDs, and output inside the core
• capture ruleset is now /3; case and collection IDs use 128-bit digest suffixes and therefore differ from earlier development snapshots
• public interfaces may still change before a stable release

Known limitations

• no authenticated historical evidence pack yet; the real 2023 banking-stress design is under separate Draft RFC review
• no malicious third-party plugin sandbox, scheduler, multi-page capture, or revision graph
• Treasury v0.1 normalizes only total debt, not public/intragovernmental components
• FRED/ALFRED data is intentionally excluded under the project’s current software/archival rights decision
• this is research infrastructure, not trading, investment, or policy advice

Community

Thank you to @HeaTTap for the project’s first external contribution: the offline connector discovery command in #16. The Treasury connector and audited integration landed in #17.

Verification

Release checks passed on Python 3.10–3.13 with 107 tests, Ruff, both offline connector replays, source-distribution inspection, and wheel installation outside the checkout. The attached files are the exact CI candidate artifacts.

SHA-256:

• openmacrostate-0.1.0a3-py3-none-any.whl: 7bc02a96149e3493393d775983ea32f6d4e0bfa63778d88074494072f2486eda
• openmacrostate-0.1.0a3.tar.gz: 36ef51c1d0909c7cea0a07383ce254889111fb69237c7425b27e12fc1765142f

No PyPI package is published by this release; install from the attached wheel if you want to reproduce it.