esgcalc — Granular offline Python library

Standalone Python library 0.2.0, Python >=3.10, no runtime dependencies. Calculation algorithms and resources derive unchanged from the verified engine; Golden Source provenance pins the implementation and resources. The calculation core implements 47 independently callable capabilities across nine domains. The native R implementation is available as esgcalc-r.

from esgcalc import social, data, rating, Context

assert social.pay_gap(20, 18).value == 10
company = data.load_reference_data("synthetic_companies.json").value["companies"][1]
result = rating.rate_company(company["observations"], company["findings"],
                             context=Context(as_of="2026-06-30"))
print(result.value, result.details["coverage"])

The declared scope includes 23 edition-pinned quantitative taxonomy clauses, six-goal assessment combination, nonfinancial turnover/CapEx/OpEx calculations including plan restatement, all 18 Table1 PAI numerical contracts, selected ENV/SOC/GOV indicators, portfolio metrics, listed-company financed emissions, a configurable transparent ESG performance rating, and local data interfaces. All runtime work is offline on supplied structured inputs. There is no network, AI extraction, service, database or mandatory whole-company data graph.

The enumerated contract defines completeness. This is not every ESG law, every taxonomy activity, complete DNSH investigation or statutory reporting. External assessments carry source/date/method/scope. The reference rating uses explicit model choices, not an alleged statutory universal ESG grade. Coverage, unknown findings and missing observations stay visible. PAI18's source ambiguity requires an explicit interpretation; it is never silently reversed.

21 original source PDFs underpin the local selected-edition review. Public resources retain provenance and hashes, not the private regulations archive. 120 reconciled annual cases for 60 fictional companies in six archetypes have independent numerical expectations; an executable mixed portfolio links the upstream metrics, PAI, ratings and financed emissions. Synthetic data are not empirical calibration, and future-year observations are labelled scenarios.

Start with contracts, API, taxonomy catalog, rating, casebook, method issues and verification.

Install from PyPI: python -m pip install esgcalc.

Local install: python -m pip install --no-index --no-deps <local wheel>. Offline example: python examples/realistic_pipeline.py output.json. Build with local tools: python -m build --no-isolation. GPL-3.0-or-later; copyright RiskDataScience GmbH. Maintainer: Dr. Dimitrios Geromichalos riskdatascience@web.de.

Developer entry points: ./scripts/check.sh and ./scripts/build.sh. See CONTRIBUTING.md, CHANGELOG.md and licence status. GitHub CI verifies Python 3.10, 3.12 and 3.13. The manually triggered release workflow publishes verified distributions through PyPI Trusted Publishing.