090°Données ouvertes
Jeux de données ouverts, entièrement documentés — interrogeables ici, et lisibles par n’importe quel LLM.
Les titres et les descriptions proviennent des sources de données, en anglais.
18 jeux de données
US State Retail Demand (Census MSRS via FRED, keyless)
Monthly US state retail-demand panel, 2019-01 onward, for 52 geographies (50 states + District of Columbia + the USA national aggregate) x 12 retail measures (Total retail sales excluding nonstore retailers + the 11 NAICS 3-digit retail subsectors): the Census Monthly State Retail Sales year-over-year percent change (not seasonally adjusted), redistributed by FRED as MSRS<STATE><SECTOR> series, with YoY acceleration, 3-month trend, deviation from the national sector, a documented 0-100 demand-pressure score, per-month-x-sector demand ranks, d1..d4 demand tiers, and contraction/soft/steady/strong/booming demand regimes. MSRS is an experimental modeled Census product (Monthly Retail Trade Survey + administrative + third-party data) released with a ~4-month lag; series are complete by construction except one documented upstream gap — 2020-05 for the clothing sector (blank on FRED, the COVID retail shutdown), kept as honest nulls — and small-state niche sectors have genuine interior gaps (e.g. Alaska sporting-goods blank 2025-05..2025-09) and genuine discontinuations (Alabama sporting-goods ends 2026-01), also kept as honest nulls. The fail-loud gates guard the FEED's liveness (global latest month within 200 days) and structure (closed 624-series world, no duplicates, implausible-value bound), not per-series recency or interior completeness. One row per state x sector x month; join keys: state_code, sector_code, year_month, country_code=USA. Who joins this: Shopify shops join demand_pressure_score on month + state + sector to forecast category demand and time promos; subscription businesses use state retail demand as a consumer-strain churn feature; sales teams rank territories by retail vitality.
US-Canada Consumer-Macro Panel (derived)
Monthly US-Canada consumer-macro panel: CPI inflation (level, MoM %, YoY %; US series seasonally adjusted, Canadian series not), unemployment rate (SA %, MoM/YoY point changes, deviation from the trailing 12-month average), consumer sentiment (US index + YoY %; Canada null — no comparable monthly series), and retail sales (SA, millions of local currency, MoM %/YoY %; US series excludes motor vehicle and parts dealers, Canadian total includes them), plus a deterministic demand_pressure_score = mean of trailing-24-month z-scores of retail YoY growth, falling unemployment and CPI YoY. Upstream: FRED keyless fredgraph.csv (CPIAUCSL, UNRATE, UMCSENT, RSXFS) and Statistics Canada full-table CSV downloads (tables 18-10-0004-01 vector 41690973, 14-10-0287-01 vector 2062815, 20-10-0056-01 vector 1446859483). One row per year_month x country_code (USA/CAN); growth columns are the cross-country-comparable fields. Who joins this: a sales team joins monthly macro to pipeline value on year_month; a subscription business joins it to MRR on year_month + country_code. Upstream gaps propagate as nulls (e.g. the US October-2025 unemployment and November-2025 CPI releases were never published during the federal shutdown); the score needs at least two of its three z-components.
U.S. Federal Highway Administration, Traffic Volume Trends
Monthly US vehicle-miles-traveled (VMT) on all roads and streets, 2000-01 to 2026-07 (319 months), from the Federal Highway Administration's keyless Traffic Volume Trends reports (SAVMT sheet: national raw and seasonally adjusted VMT in millions). Each row carries raw and seasonally adjusted VMT, year-over-year change, month-over-month change on the adjusted series, the trailing 12-month moving average and deviation from it, a 12-month rolling z-score, the trailing 12-month moving total, a record-high flag, and a documented 0-100 mobility_score (50 + 2.5*clip(dev_vs_ma12_pct,-10,10) + 5*clip(z12,-2.5,2.5)) with quintile tiers. Method: the latest monthly TVT workbook is fetched, the SAVMT sheet is parsed with fail-loud gates (monthly continuity, positive VMT, SA within 40% of raw, no duplicate months), and features are derived deterministically. Caveats: TVT estimates come from ~5,000 counting locations and are revised annually — recent months are preliminary; the series lags ~2 months; the April 2020 COVID trough is a genuine observation. FHWA data are U.S. federal public domain (17 U.S.C. 105); commercial use allowed with attribution.
US Consumer Credit Stress (FRED, keyless)
Quarterly US consumer credit-stress panel, 2005-Q1 onward: credit-card delinquency rate (SA, all commercial banks), credit-card plan APR, bank prime loan rate (quarterly mean of monthly), and the household debt-service ratio (required debt payments as % of disposable income), plus 4-quarter point changes, the card-APR-over-prime spread, and a documented 0-100 credit_stress_score (mean of full-panel min-max normalizations of delinquency, APR and debt-service; higher = more stressed) with s1..s4 stress tiers. Upstream: FRED keyless fredgraph.csv (DRCCLACBS, TERMCBCCALLNS, MPRIME, TDSP; Board of Governors of the Federal Reserve System). Panel start is set by TDSP's keyless coverage (2005-Q1). One row per quarter x USA; in-window nulls fail loudly, never imputed. Who joins this: subscription businesses join churn on year_quarter; online shops join demand on year_quarter + country_code.
US Consumer Sentiment Intelligence (FRED, keyless)
Monthly US consumer-sentiment intelligence panel, 1978-01 onward: the University of Michigan Consumer Sentiment Index (UMCSENT, 1966:Q1=100) and households' 1-year-ahead inflation expectations (MICH, median %), plus 1- and 12-month point changes, a weak/neutral/strong sentiment_regime on documented cut points, and an expectations_anchored flag (1y expectations < 3.0). Upstream: FRED keyless fredgraph.csv (UMCSENT, MICH; University of Michigan Surveys of Consumers). One row per month x USA; in-window nulls fail loudly, never imputed. Who joins this: online shops join demand on year_month + country_code; subscription businesses join churn on year_month.
US Business-Cycle & Core-Inflation Intelligence (FRED, keyless)
Monthly US business-cycle reference panel, 1854-12 onward: the NBER-based recession indicator (FRED USREC), a 0/1 dummy marking the period from the month following each NBER business-cycle peak through the trough month. Carries recession_start and recession_end flags, a per-episode episode_id, months_into_recession and months_since_last_recession for regime-duration features. Upstream: National Bureau of Economic Research business-cycle reference dates, redistributed keyless via FRED fredgraph.csv (restores the reference series purged from the API-key FRED connector on 2026-09-29). Method caveats: the NBER declares peaks and troughs with a long lag, so recent months can be revised from 0 to 1; the opening 1854-12 episode starts in-recession with no observed start. One row per month x USA; the series is binary by construction and any non-0/1 value fails loudly. Who joins this: all three personas join on year_month + country_code to condition demand, churn and pipeline models on the official recession regime.
US Housing Starts & New-Home Sales Intelligence (FRED, keyless)
Monthly US housing-starts intelligence panel, 1959-01 onward: total new privately-owned housing units started (thousands of units, seasonally adjusted annual rate; FRED HOUST, U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction release, redistributed keyless via FRED fredgraph.csv). Carries 1-month and 12-month changes (levels and percents), the 3-month annualized percent change, and a contraction_flag marking months with fewer starts than a year earlier. Method caveats: Census revises history with each vintage (kept as published, never back-filled); in-window nulls fail loudly, never imputed. Who joins this: home-improvement merchants join demand on year_month + country_code and use starts_yoy_pct / contraction_flag as housing-demand features; sales teams weight real-estate-adjacent pipeline by the contraction regime.
Monthly US new-home-sales intelligence panel, 1963-01 onward: new one-family houses sold in the United States (thousands of units, seasonally adjusted annual rate; FRED HSN1F, U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Sales release, redistributed keyless via FRED fredgraph.csv). Carries 1-month and 12-month changes (levels and percents), the 3-month annualized percent change, and a contraction_flag marking months with fewer sales than a year earlier. Method caveats: a 'sale' is recorded when a deposit is taken or a sales agreement is signed, which can precede the building permit — monthly sales can exceed starts for this reason; Census revises history with each vintage (kept as published, never back-filled); in-window nulls fail loudly, never imputed. Who joins this: home-improvement and building-supply merchants join demand on year_month + country_code and use sales_yoy_pct / contraction_flag as housing-demand features.
US Initial Jobless-Claims Intelligence (FRED, keyless)
Weekly US initial jobless-claims intelligence panel, 1967-01-07 onward: initial claims for unemployment insurance (FRED ICSA, U.S. Employment and Training Administration, seasonally adjusted, weekly ending Saturday, number of claims, redistributed keyless via FRED fredgraph.csv). Carries the trailing 4-week average (the standard ICSA read), 1-week and 52-week changes (levels and percents), the change of the 4-week average, a trailing 52-week z-score (the layoff speedometer), and an elevated_flag marking a labor-softening watch when the 52-week percent change reaches +15%. One row per Saturday x USA; in-window nulls fail loudly, never imputed. Who joins this: employment teams join hiring plans on year_week + country_code and use the 4-week average / elevated_flag as layoff-regime features; consumer-spending models use the z-score as the income-risk gauge.
US Building-Permits Intelligence (FRED, keyless)
Monthly US building-permits intelligence panel, 1960-01 onward: new privately-owned housing units authorized in permit-issuing places (total units, thousands, seasonally adjusted annual rate; FRED PERMIT, U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction release, redistributed keyless via FRED fredgraph.csv). Carries 1-month and 12-month changes (levels and percents), the 3-month annualized percent change, and a contraction_flag marking months with fewer permits authorized than a year earlier. Method caveats: the 2005-02-16 release expanded the permit-issuing universe from 19,000 to 20,000 places (level break kept verbatim); Census revises history with each vintage. One row per month x USA; in-window nulls fail loudly, never imputed. Who joins this: home-improvement merchants join demand on year_month + country_code and use permits_yoy_pct / contraction_flag as housing-demand features; sales teams weight pipeline by the contraction regime.
US Vehicle-Sales Intelligence (FRED, keyless)
Monthly US total-vehicle-sales intelligence panel, 1976-01 onward: total vehicle sales (cars + light trucks, millions of units, seasonally adjusted annual rate; FRED TOTALSA, U.S. Bureau of Economic Analysis, redistributed keyless via FRED fredgraph.csv). Carries 1-month and 12-month changes (levels and percents), the 3-month annualized percent change, and a contraction_flag marking months with fewer vehicles sold than a year earlier. Method caveats: BEA/FRED revise history with each vintage; the April-2020 COVID trough (8.919 million units) and the 2021 chip-shortage dip are genuine upstream prints kept verbatim. One row per month x USA; in-window nulls fail loudly, never imputed. Who joins this: auto dealers and parts/accessories merchants join demand on year_month + country_code and use sales_yoy_pct / contraction_flag as auto-demand features; subscription businesses use the 3-month momentum as big-ticket household demand pressure; sales teams weight auto-adjacent pipeline by the contraction regime.
US Fed Balance-Sheet Intelligence (FRED, keyless)
Weekly US Federal Reserve balance-sheet intelligence panel, 2002-12-18 onward: total assets less eliminations from consolidation (Wednesday level, millions of U.S. dollars, not seasonally adjusted; FRED WALCL, Board of Governors of the Federal Reserve System, H.4.1 Factors Affecting Reserve Balances release, redistributed keyless via FRED fredgraph.csv). Carries 1-week and 52-week changes (levels and percents), the 13-week annualized percent change (the quantitative-tightening pace read), the percent below the trailing 260-week peak (the QT depth gauge), and a contraction_flag marking weeks the balance sheet is smaller than a year earlier. One row per Wednesday x USA; in-window nulls fail loudly, never imputed. Who joins this: rates desks and mortgage lenders join funding costs on year_week + country_code and use the QT pace and contraction_flag as liquidity-regime features; macro models use pct_off_peak as the QT-depth gauge.
Une fois l’allocation gratuite du jour épuisée, les fonctions d’IA peuvent passer par votre propre compte fournisseur.
Conservée uniquement dans cet onglet (effacée à sa fermeture) et envoyée avec chaque requête d’IA. Nos serveurs l’utilisent pour cette requête et ne la stockent ni ne la journalisent jamais.