igot-ai/enterprise-trade-graph

Enterprise Trade Graph

Real trade, macro and hazard data staged from public sources, the ground tables joined from them, and a synthetic enterprise world built on them: firms from SDG-PGMs, a supply graph reconciled to BACI, and a structural causal layer on the real 2017-2024 calendar with exact truth. It serves the RFC "Causal Cascades on the Ontology" (code: igot-ai/flowgpt, research/trade_graph).

Terms of use in brief. The dataset is free to download. Each table keeps the licence or terms of the source it was built from: see the Licence column below and Licences, attribution and terms of use. For tables marked IMF-terms, credit the IMF, keep the data accurate, ask the IMF (copyright@imf.org) before any commercial reuse, and do not use them to train large language models without the IMF's explicit permission. No original UN Comtrade trade records are included. This is an independent compilation, not affiliated with or endorsed by any data provider.

Using the data

Each table is a separate config. Load one at a time with the datasets library:

from datasets import load_dataset

cpi = load_dataset("igot-ai/enterprise-trade-graph", "real_imf_cpi_m", split="train")

# Very large tables (for example synthetic_firm_month_panel, 154M rows) are easier to stream
panel = load_dataset(
    "igot-ai/enterprise-trade-graph", "synthetic_firm_month_panel",
    split="train", streaming=True,
)

The Parquet files sit under data/real/, data/ground/ and data/synthetic/, one folder per table. If this page asks you to accept the terms first, accept them and log in with a Hugging Face token (for example hf auth login) before loading.

Real data, staged from public sources

Config	Rows	What it is	Licence
real_baci_flows	89,207,221	Bilateral merchandise trade flows, year x exporter x importer x HS6 product, value in thousand USD and quantity in metric tons; M49 exporter/importer mapped to ISO3.	Etalab-2.0
real_chokepoint_transits_daily	79,156	IMF PortWatch daily chokepoint transit calls by ship class and estimated transiting trade volume in metric tons (volume_t_*; the provider's 'capacity' fields), 28 chokepoints, UTC dates from 2019-01-01.	IMF-terms
real_chokepoint_transits_monthly	2,604	Chokepoint x calendar month sums of chokepoint_transits_daily with days_observed; the latest month is partial.	IMF-terms
real_chokepoints	28	IMF PortWatch chokepoints (28 straits, canals and capes) with coordinates, yearly-average transit calls by ship class and top industries.	IMF-terms
real_climate_indices_m	3,693	Monthly climate indices, long format.	public-domain
real_company_denylist	7,524	Guard list of about 7.5k of the world's largest companies and brands (name, kind, coarse sector, coarse HQ area); generated names must never hit it.	constructed
real_comtrade_corridors	193	The (exporter, importer, hs2) corridors crawled from Comtrade and why each was chosen (RFC scenario and/or top-120 BACI-2024 corridor, with its 2024 rank and value), plus per-direction support and the months actually obtained.	UN-Comtrade-derived
real_countries	331	Country reference: ISO3 key, M49, ISO2, name, World Bank region/income group, capital coordinates, aggregate flag	Etalab-2.0
real_country_sector_notes	1,130	Main export sectors of each country: the top HS2 chapters of goods exports with shares, reference window mean of BACI trade_hs2 (no HS-to-ISIC mapping).	constructed
real_enso_events	42	El Nino / La Nina events = maximal runs of oni_monthly.phase.	public-domain
real_enso_exposure	253	Per-country ENSO exposure score in [0,1] (half peak cell strength, half breadth over the 8 phase-season cells) with its class, components and derivation text.	constructed
real_enso_sources	100	Reference list (key, citation, url where one is known) for the source keys used in enso_teleconnection.	constructed
real_enso_teleconnection	1,114	ENSO teleconnection by country: iso3 x phase (el_nino/la_nina) x season (DJF/MAM/JJA/SON) x hazard (drought, flood/heavy rain, heat, tropical cyclone activity, cold) with direction (+ more / - less than normal), strength (none/weak/moderate/strong), the sectors affected, the rationale and cited source keys.	constructed
real_hazards	89,895	GDACS disaster events 2017-: earthquakes, tropical cyclones, floods, droughts, volcanoes, wildfires (all alert levels), one row per event with type, name, ISO3 countries (iso3_source says whether GDACS, its free text or the event's coordinates gave them), dates, alert level/score, severity, coordinates and, where GDACS gives them, affected-population figures (persons).	CC-BY-4.0
real_hs2_isic	96	HS chapter (01-97) -> primary ISIC rev.4 division and section, secondary divisions above 10% of chapter trade.	constructed
real_hs2_pcps_index	29	HS chapter -> broad IMF PCPS commodity price index that contains its commodities.	constructed
real_hs4_bec	1,222	HS4 -> intermediate/capital/consumption shares (trade-weighted mean of its lines) and use class.	constructed
real_hs4_isic	1,354	HS4 -> ISIC rev.4 divisions producing it, share by 2024 BACI trade value, primary and material secondary flagged.	constructed
real_hs6_isic	5,383	HS 2017 line -> producing ISIC rev.4 division/class/section (UNSD HS-CPC-ISIC chain, equal splits; primary by largest share).	constructed
real_hs_bec	5,383	HS 2017 line -> BEC rev.5 code, end use and intermediate/capital/consumption shares (UNSD HS-BEC5 correlation).	constructed
real_hs_pcps	114	HS4/HS6 -> IMF PCPS commodity code and World Bank Pink Sheet series, exact/variety/proxy, validated against both catalogues.	constructed
real_ilo_employment_estsize	218,511	ILOSTAT employment by establishment size band x ISIC rev.4.	CC-BY-4.0
real_ilo_employment_isic	175,265	ILOSTAT employment by ISIC rev.4.	CC-BY-4.0
real_ilo_status_isic	69,200	ILOSTAT employment by status (ICSE-93: total, employers, own-account workers) x ISIC rev.4.	CC-BY-4.0
real_imf_cpi_m	35,627	IMF Consumer Price Index, monthly, headline (all items): iso3 x month index level, its base period and year-on-year % change	IMF-terms
real_imf_fx_m	39,678	IMF exchange rates, monthly: iso3 x month national currency per US dollar, period average (euro-area members filled from the euro-area series, flagged)	IMF-terms
real_imf_imts_m	3,725,959	IMF International Trade in Goods by partner: reporter x partner x month exports (FOB) and imports (CIF) in US dollars, one file per year	IMF-terms
real_imf_imts_world_m	28,496	IMF International Trade in Goods: reporter x month total exports (FOB) and imports (CIF) with the World partner (G001), US dollars	IMF-terms
real_imf_pcps_m	42,997	IMF Primary Commodity Prices, monthly: commodity x month price level in the commodity's own unit, plus the IMF price index (2016=100)	IMF-terms
real_isic4_divisions	88	ISIC rev.4 divisions with their section (UNSD structure).	constructed
real_isic_input_rules	51	The cited technology rules behind isic_inputs: which supplier divisions or HS lines an ISIC user really buys (agriculture buys fertiliser, valuables are not intermediates), with rationale and source per rule.	constructed
real_isic_inputs	2,125	ISIC rev.4 division -> top HS4 products bought as intermediates (weights sum to 1): Eurostat 2019 use table (product by product, summed over the member states that publish it), supplier flows narrowed by cited technology rules (isic_input_rules), magnitude checked against OECD TiVA intermediate intensity.	constructed
real_macro_catalogs	13	One row per igot-ai/tmp-trade catalog (13): source, adapter, ingestion status, real row / series / area counts, period_types and chronological period coverage, level-break screening (n_break_flags, break_screening), the Data360 provider metadata, licence family, documented limitations and the STATUS.md resume history.	constructed
real_macro_series	44,264	Long-format macro / commodity / shipping series from igot-ai/tmp-trade's 13 SQLite catalogs (IMF IFS mirrors, World Bank WDI/GEM, UNCTAD LSCI, EIA and IMF via FRED), 2016-09 to 2026: one row per series x area x period.	IMF-terms
real_name_country_map	253	Per-country name profile: languages used (primary first), legal-form family, word order and the basis of the language choice.	constructed
real_name_lexicon	7,881	Building blocks of synthetic company names: generic trade words per ISIC section, place-neutral nature and virtue words, per-country legal forms, and romanised onset/nucleus/coda morphemes per language.	constructed
real_oecd_tiva	25,839,360	OECD TiVA 2025 edition (2015-2022): trade in value added, thousand USD (value_kusd; the OECD publishes USD million), reporter x counterpart x industry x measure x year.	CC-BY-4.0
real_oecd_tiva_exgrva	5,120,000	OECD TiVA 2025 edition (2015-2022): value added in the gross exports of an industry, by the country that created it (source area), thousand USD (value_kusd).	CC-BY-4.0
real_oecd_tiva_shares	1,115,520	OECD TiVA 2025 edition (2015-2022) percentage shares by reporter x industry: EXGR_DVA and EXGR_FVA with unit PT_EXGR (domestic / foreign value-added share of the industry's gross exports), EXGR_INT_DVA, EXGR_SERV_DVA, VALU_PROD (value added per unit of output).	CC-BY-4.0
real_oni_monthly	919	NOAA CPC Oceanic Nino Index: overlapping 3-month Nino 3.4 SST anomaly (oni_c, degC) and running-mean SST (sst_c, degC).	public-domain
real_port_activity_daily	5,833,625	IMF PortWatch daily port activity (UTC dates from 2019-01-01): port calls by ship class and estimated import/export volume in metric tons (AIS draught change x deadweight), one row per port and day, zero-filled by the provider.	IMF-terms
real_port_activity_monthly	192,045	Port x calendar month sums of port_activity_daily (port calls, import/export metric tons by ship class) with days_observed; the latest month is partial (is_complete_month=False).	IMF-terms
real_port_crosswalk	1,531	Crosswalk between IMF PortWatch ports (pw:) and NGA World Port Index ports (wpi:) by name and distance, same country.	IMF-terms
real_port_disruption_links	476	Event x affected port or chokepoint (from the events' affectedports list) with the event-to-port distance in km where the disruptions_with_ports layer lists the pair.	IMF-terms
real_port_disruptions	133	IMF PortWatch disruption events: GDACS-derived natural hazards (EQ/TC/FL/DR/WF/VO, join on event_id to GDACS) plus PortWatch-curated 'OT' events (Red Sea, Panama drought, Baltimore bridge, US ports strike, Hormuz).	IMF-terms
real_ports	2,065	IMF PortWatch port database: 2,065 ports and offshore oil terminals with coordinates, country, yearly-average vessel calls by ship class (2019-2025 AIS), top-3 traded industries (HS-section groups, estimated) and the port's share of its country's maritime imports/exports (%).	IMF-terms
real_ports_wpi	3,669	NGA World Port Index: 3,669 ports with coordinates, ISO3 country, harbor size/type, shelter, maximum vessel class, depth classes for the channel, anchorage, cargo pier and oil terminal (metre bounds of the Pub. 150 depth codes).	CC0-1.0
real_products	5,384	HS6 product reference: HS4/HS2 parents, HS section (I-XXI, constructed from the WCO nomenclature), BACI HS6 description, official HS4/HS2 titles from the UN Comtrade H5 reference, and 2024 world trade value.	Etalab-2.0 + UN-Comtrade-derived
real_trade_hs2	7,037,961	BACI bilateral flows aggregated from HS6 to HS2: year x exporter x importer x hs2.	Etalab-2.0
real_trade_hs4	36,160,567	BACI bilateral flows aggregated from HS6 to HS4: year x exporter x importer x hs4.	Etalab-2.0
real_trade_totals	1,808	Per country-year trade totals from BACI: exports_kusd, imports_kusd, n_partners (distinct countries traded with, export or import side) and n_products (distinct HS6 products exported).	Etalab-2.0
real_unctad_container_teu	2,900	Container port throughput in TEU by economy and year (UNCTAD annual).	CC-BY-3.0-IGO
real_unctad_fdi	35,260	Foreign direct investment inward/outward flows and stocks by economy and year, US$ and shares (UNCTAD, 1990-latest).	CC-BY-3.0-IGO
real_unctad_fleet	58,814	Merchant fleet by flag of registration, ship type and year: ships, dwt, gross tonnage, age, shares (UNCTAD, 1980-latest).	CC-BY-3.0-IGO
real_unctad_gdp	15,279	Total GDP and GDP per capita by economy and year, current and constant 2020 US$, real growth (UNCTAD, 1970-latest).	CC-BY-3.0-IGO
real_unctad_lsbci	655,370	Bilateral liner shipping connectivity index for every ordered economy pair, first quarter of each year 2006-latest (UNCTAD publishes only Q1).	CC-BY-3.0-IGO
real_unctad_lsci_m	20,070	Liner Shipping Connectivity Index by economy and month (UNCTAD): index (Feb 2023 = 100) and growth rates.	CC-BY-3.0-IGO
real_unctad_lsci_q	14,724	Liner Shipping Connectivity Index by economy and quarter (UNCTAD, 2006Q1-latest): index, growth rates, rank.	CC-BY-3.0-IGO
real_unctad_port_calls	1,161	Port-call statistics by economy, year and ship market: time in port, vessel age and size (UNCTAD, AIS-based).	CC-BY-3.0-IGO
real_unctad_port_lsci	76,003	Port-level Liner Shipping Connectivity Index by UN/LOCODE port and quarter (UNCTAD, 2006Q1-latest).	CC-BY-3.0-IGO
real_unctad_trade_matrix	10,488,019	Merchandise trade matrix by SITC section: reporter x partner x flow x product x year, value in US$ (UNCTAD, mirror-estimated).	CC-BY-3.0-IGO
real_wb_edd	3,094	Exporter concentration and dynamics by economy and year: number of exporters and the shares of total export value held by the top 1%, 5% and 25% of exporters.	CC-BY-4.0
real_wb_pinksheet_m	63,183	World Bank Pink Sheet monthly commodity prices (nominal USD) and price indices (2010=100), 1960-latest: one row per month x series	CC-BY-4.0 (unconfirmed)
real_wb_pinksheet_series	87	Catalogue of the Pink Sheet series: label, code, unit, kind, index hierarchy (level, parent_code), coverage, ends_early flag	CC-BY-4.0 (unconfirmed)
real_wdi	249,506	World Bank WDI (plus Doing Business / Gender Statistics for archived codes), long: one row per iso3 x year x indicator with its unit, non-null observations, 2000-2025	CC-BY-4.0
real_wdi_indicators	53	One row per fetched indicator: code actually queried vs code requested, kind (requested/renamed/successor/recovered/extra), API source, definition, source organisation, coverage	CC-BY-4.0
real_wdi_wide	6,870	World Bank WDI wide: iso3 x year, one snake_case column per indicator with its unit in the name (_usd, _pct, _persons, _teu, _mt, _hours, _days, _score, _idx)	CC-BY-4.0

Ground tables joined from the staged sources

Config	Rows	What it is	Licence
ground_corridors	36,160,567	Bilateral HS4 trade flows 2017-2024 with liner connectivity, distance and monthly-data availability (for IMF IMTS and UN Comtrade: month counts and flags only, no monthly values).	CC-BY-3.0-IGO + CC-BY-4.0 + Etalab-2.0 + IMF-terms + UN-Comtrade-derived
ground_destination_mix	6,340,572	Importer share of each exporter's HS4 exports, mean of 2022-2024.	Etalab-2.0
ground_economies_annual	2,390	Country-year profile 2015-2024: income group, region, GDP, population, sector shares, trade openness, LPI, LSCI (annual mean), TEU, FDI, TiVA value-added shares and the ENSO exposure score.	CC-BY-3.0-IGO + CC-BY-4.0 + Etalab-2.0 + constructed
ground_export_mix	226,071	Export value by exporter, ISIC division and HS4 product, mean of 2022-2024.	CC-BY-3.0-IGO + Etalab-2.0
ground_macro_monthly	181,314	Monthly commodity prices, exchange rates and consumer prices with the 3-month log change.	CC-BY-4.0 + IMF-terms
ground_regime_monthly	919	ENSO regime by month from 1950: ONI, phase, event, strength and the DMI and SOI indices.	public-domain
ground_sector_mix	4,137	Share of employment by ISIC section (A-U) for each country, with the basis and the imputations flagged.	CC-BY-4.0 + Etalab-2.0
ground_shocks_monthly	269,700	Monthly shocks per country (hazard intensity), port and chokepoint (disruptions and activity anomaly z-scores).	CC-BY-4.0 + IMF-terms + public-domain

Synthetic enterprise world

Config	Rows	What it is	Licence
synthetic_b2b_corridors	698,373	One row per planned corridor (exporter, importer, hs4) of the synthetic B2B graph: the real BACI flow (or, for domestic corridors, the capacity reference), the edge slots, the edges drawn, the realised synthetic value, the coverage share phi and the reason a corridor carries no edge.	synthetic
synthetic_b2b_edges	4,396,977	Synthetic business-to-business supply edges between the synthetic enterprises: cross-border edges drawn from the real BACI corridors (2022-2024 mean of at least USD 1 million) and domestic multi-tier edges from the ISIC input table, with yearly values reconciled to phi x the real corridor value, the share of the customer's input, the primary-supplier flag, contract type, relationship start year and sea lane.	synthetic
synthetic_enterprises	1,000,000	Synthetic enterprises that take part in cross-border goods supply chains in 2024: location, activity, scale, ownership, trade profile, latent traits, exposures and a template profile text.	synthetic
synthetic_firm_month_panel	153,742,848	Synthetic monthly firm panel on the real 2017-2024 calendar from a structural causal model over the synthetic B2B graph: for every firm that exists and every month, whether it was disrupted and whether it delivered late, in four worlds (declared, no_cascade, hidden_off, bypass) that share their random draws, with the five observable drivers that entered its equation (ENSO, hazard, home port, chokepoint, commodity).	synthetic
synthetic_lanes	409,493	Sea lanes (origin home port, destination home port) used by at least one cross-border edge: route length, chokepoints crossed, typical transit days, the alternative route around Suez, Bab el-Mandeb or Panama, and the bilateral liner connectivity of the two port countries.	synthetic
synthetic_truth_edges	4,396,977	Exact causal effects of the declared world, one row per edge of the synthetic B2B graph in force in 2024: tau (high for the customer's primary supplier), the lag profile and upsilon, the cascade and congestion coefficients, the effect of a supplier's disruption on its customer's lateness at lags 1 to 6 and summed, the upstream channel and the bypass loading.	synthetic
synthetic_truth_hidden_demand	2,016	The latent AR(1) demand driver of each ISIC section and month of the calendar, its loading and its contribution to the probabilities of the worlds that carry it (declared, no_cascade, bypass): the confounder that the panel does not publish.	synthetic
synthetic_truth_worlds	2,123,656	The average exact effects per scenario world (the truth_scenarios of SYNTHESIS.md section 4): one row per world and one per corridor within each world, with the mean and the value-weighted mean of tau, the cascade, congestion and total effects at each lag 1 to 6 and summed.	synthetic

How it was built

• Tiers: raw downloads, then staging (one normalised table per source table), then ground (joined tables), then synthetic tables. Every table carries a manifest with its sources, licence, keys, row count and sha256; they are in manifests/.
• The synthetic tables are deterministic for their seed and scale; every synthetic parameter is named in the manifest notes, and validation reports are in docs/synthetic/.
• Partitioned tables keep their layout (for example year=2017.parquet); files above 500 MB are split into numbered shards.

Real-data scenarios

• rfc/insights/: four hypothesis cards frozen before estimation, and their results (estimate, falsification tests, verdict, deviations).
• rfc/scenarios/: the 144 causal-cascade scenarios regrounded on real corridors.

Not included in this release

• raw/: third-party downloads exactly as fetched. They can be re-downloaded from their sources; some may not be redistributed as-is (UN Comtrade), and others are another publisher's dataset.
• staging/comtrade_monthly: original UN Comtrade monthly records, which may not be redistributed as-is without UNSD's permission. Only the derived content listed under UN Comtrade is published.
• eval/: evaluation runs still being produced.
• ground/_smoke/ and synth/_smoke/: smoke-test outputs at 1% scale, not the release.

Licences, attribution and terms of use

The dataset is free to download and is a compilation: each table keeps the licence or terms of the source it was built from, as shown in the Licence column above and in the licence field of its manifest in manifests/. A table built from several sources carries all of their terms. If you use or redistribute a table, follow the terms of every source behind it.

Licence tags

Tag	Source	Terms	Credit as
Etalab-2.0	CEPII, BACI	Licence Ouverte / Open Licence 2.0 (Etalab): reuse, including commercial, with attribution	CEPII, BACI (Gaulier and Zignago, 2010)
CC-BY-4.0	World Bank (WDI, Exporter Dynamics Database, Pink Sheet), ILO (ILOSTAT), OECD (TiVA 2025 edition), GDACS	CC BY 4.0: reuse, including commercial, with attribution	Publisher and database, for example "World Bank, World Development Indicators"
CC-BY-3.0-IGO	UNCTAD (UNCTADstat)	CC BY 3.0 IGO: reuse, including commercial, with attribution	UNCTAD, UNCTADstat
CC0-1.0	NGA, World Port Index	CC0 1.0: no restrictions	NGA, World Port Index (Pub. 150)
public-domain	NOAA (Oceanic Nino Index, climate indices)	Public domain	NOAA
IMF-terms	IMF data portal and IMF PortWatch	IMF Copyright and Usage, data section: see IMF data	See IMF data
UN-Comtrade-derived	UN Comtrade (UNSD)	Transformed data only: see UN Comtrade	UN Comtrade
constructed	Built by this project from the sources named in its row	CC BY 4.0 for this project's own contribution; the terms of the sources it draws on still apply	iGOT AI, Enterprise Trade Graph
synthetic	Generated by this project	CC BY 4.0 for the synthetic part; real values carried over keep their source's terms	iGOT AI, Enterprise Trade Graph

Constructed tables also draw on the Eurostat 2019 use tables (real_isic_inputs), UNSD classification correspondences (HS-CPC-ISIC, HS-BEC5) and the ISIC structure, the WCO Harmonized System nomenclature, and the ENSO literature listed in real_enso_sources.

IMF data

Tables: real_imf_cpi_m, real_imf_fx_m, real_imf_imts_m, real_imf_imts_world_m, real_imf_pcps_m, real_macro_series, the IMF PortWatch tables (real_ports, real_chokepoints, real_port_activity_daily, real_port_activity_monthly, real_chokepoint_transits_daily, real_chokepoint_transits_monthly, real_port_disruptions, real_port_disruption_links, real_port_crosswalk), and the ground tables marked IMF-terms.

These tables reuse published IMF statistics under the data section of the IMF Copyright and Usage terms. Those terms allow you to download, copy, transform, publish and redistribute IMF data, provided that you:

1. credit the IMF as the source, naming the database and linking to it (credit lines below);
2. keep the data accurate, and state any material transformation you make (ours are listed below);
3. take reasonable steps to pass these terms on to anyone you redistribute the data to;
4. tell buyers the data is available free from the IMF, if you ever sell it as a standalone product.

Two further rules from the same terms apply to every user of these tables:

• Commercial reuse: the IMF asks anyone considering commercial reuse of its data to request permission at copyright@imf.org.
• AI training: the IMF does not allow its content to be used to train large language models without its explicit permission.

Credit lines

Tables	Credit line
real_imf_cpi_m	Source: International Monetary Fund, Consumer Price Index (CPI), https://data.imf.org/Datasets/CPI
real_imf_fx_m	Source: International Monetary Fund, Exchange Rates (ER), https://data.imf.org/Datasets/ER
real_imf_imts_m, real_imf_imts_world_m	Source: International Monetary Fund, International Trade in Goods (by partner country) (IMTS), https://data.imf.org/en/datasets/IMF.STA:IMTS
real_imf_pcps_m	Source: International Monetary Fund, Primary Commodity Price System (PCPS), https://data.imf.org/Datasets/PCPS
IMF PortWatch tables	Source: International Monetary Fund, IMF PortWatch, https://portwatch.imf.org (developed by the IMF and the University of Oxford; vessel data from the UN Global Platform)
real_macro_series	Source: the publisher and database named for each series in real_macro_catalogs (International Monetary Fund, World Bank, UNCTAD, U.S. EIA; some series obtained via FRED, Federal Reserve Bank of St. Louis)

Transformations of IMF data

• real_imf_cpi_m: the year-on-year % change is computed from the published index.
• real_imf_fx_m: euro-area members are filled from the euro-area series and flagged.
• real_imf_imts_m, real_imf_imts_world_m, real_imf_pcps_m: reshaped to long format.
• PortWatch daily tables: the provider's 'capacity' fields are renamed volume_t_* (metric tons); the zero-filling of days without activity is PortWatch's own.
• real_port_activity_monthly, real_chokepoint_transits_monthly: calendar-month sums of the daily tables, with days_observed; the latest month is partial.
• real_port_crosswalk: our matching of PortWatch ports to NGA World Port Index ports, by name and distance within the same country.
• real_port_disruption_links: event-to-port pairs from each event's affected-ports list, with the distance where PortWatch lists the pair.
• ground_macro_monthly: 3-month log changes computed from the monthly series.
• ground_shocks_monthly: activity anomaly z-scores computed from PortWatch activity.
• ground_corridors: IMF IMTS enters only as month counts and availability flags.

UN Comtrade

UN Comtrade data are copyrighted by the United Nations. UNSD's policy on use and re-dissemination restricts redistributing original Comtrade records, while data that has been substantially transformed (new indicators, aggregations, derived metadata) is not subject to that restriction. This release therefore contains no original UN Comtrade trade records. Comtrade-derived content (tag UN-Comtrade-derived) is limited to:

• real_comtrade_corridors (193 rows): which exporter-importer-HS2 corridors were queried and why, per-direction support and the months obtained (derived ratios and counts);
• two columns of ground_corridors, comtrade_monthly_months and comtrade_monthly_available: month counts and availability flags, no values;
• the official HS4/HS2 titles in real_products, taken from the UN Comtrade H5 reference list.

Cite the source as "UN Comtrade" (https://comtrade.un.org). The monthly records these were derived from came through the Comtrade preview API and are not redistributed.

World Bank Pink Sheet

real_wb_pinksheet_m and real_wb_pinksheet_series are listed under CC BY 4.0, the World Bank's default licence for the open data it produces. Redistribution of the Pink Sheet has not yet been confirmed with the World Bank; the same caveat applies to any ground table built from these two tables.

Synthetic and constructed tables

Synthetic tables are generated by this project, and their synthetic part is released under CC BY 4.0. They were computed from the real tables above, so real values they carry over, such as the BACI corridor values in synthetic_b2b_corridors, keep their source's terms. Constructed tables (mappings, rules, reference lists and scores built by this project) follow the same rule.

No affiliation, no warranty

This dataset is an independent compilation by iGOT AI. It is not produced, reviewed or endorsed by the IMF, the United Nations, UNCTAD, the World Bank, the ILO, the OECD, CEPII, NOAA, NGA, GDACS or any other source, and nothing in it implies affiliation with them. All data are provided as is, without warranty of any kind; check the original source before relying on a figure.

Questions and takedown requests

Data providers and users can raise licensing questions in the Community tab. We will review any concern about a source's terms promptly and update or remove the affected tables where needed.

Limitations

• Synthetic tables are not measurements: their coefficients are project choices, and only a few borrow a published shape.
• Some staged sources are partial (the UN Comtrade preview API); each manifest says so.

Citation

If you use this dataset, please cite it together with the original sources of the tables you use (credit lines above).

@misc{igotai_enterprise_trade_graph,
  title        = {Enterprise Trade Graph},
  author       = {{iGOT AI}},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/igot-ai/enterprise-trade-graph}}
}

BACI: Gaulier, G. and Zignago, S. (2010). BACI: International Trade Database at the Product-Level. The 1994-2007 Version. CEPII Working Paper 2010-23.