Skip to content

tossd-reader

Total Official Support for Sustainable Development (TOSSD) is an international standard tracked by the International Forum on TOSSD at tossd.online. It measures cross-border development finance under Pillar I and expenditures for global public goods under Pillar II. The published dataset covers six years (2019 to 2024) across 2.4 million activity-level records.

tossd-reader provides Python analysts with typed, cached pandas DataFrames loaded directly from these official records. Annual files download once and cache locally on disk, so repeat queries run from local storage.

Quick verification

A single query reproduces the 2024 published headline figures of USD 364.1 billion for Pillar I and USD 133.6 billion for Pillar II directly from the raw activity records.

import tossd_reader as tossd

df = tossd.get_tossd(years=2024, columns="minimal", units="usd_million")
df.groupby("tossd_pillar")["usd_disbursement"].sum().round(1)

tossd_pillar
1    364114.1
2    133561.8
Name: usd_disbursement, dtype: float64

Install

Python 3.12 or newer.

uv add git+https://github.com/ONEcampaign/tossd-reader.git

pip install git+https://github.com/ONEcampaign/tossd-reader.git

What the package handles

  • Typed DataFrames with snake_case column names, standard numeric types, nullable integer codes, and string categoricals.
  • Provider and recipient filtering by name or official code, with fuzzy suggestions for misspelled names.
  • Double-counting protection through the is_aggregate flag to separate activity-level transactions from summary records.
  • Domain helpers for multi-goal SDG weighting and thematic keyword tags for climate and gender, plus typed constant-price columns ready for year-over-year comparison.
  • Automatic disk caching to accelerate repeat queries.

Data size and workflow

  • Initial queries for a year download the full published annual dataset (55 to 91 MB per year).
  • The columns="minimal" preset loads the core financial and classification fields, keeping memory usage low during interactive analysis.
  • Query parameters filter on years, providers, recipients, and pillars at load time. The filters= argument extends load-time filtering to sector, purpose, channel, modality, and three more dimensions. See How to filter by sector.

Where to start