The concepts behind procurement-derived intelligence, defined the way Valan Technologies builds them: point-in-time discipline, deterministic entity resolution, honest NULLs, and the negative space.
Procurement intelligence is the systematic analysis of public government contract data — tenders (solicitations) and awards (signed contracts) — as an alternative data source. Because governments publish what they buy, from whom, and for how much, procurement data reveals corporate revenue events before earnings, state stockpiling programmes, supply-chain relationships, and geopolitical risk signals. Valan Technologies aggregates over 67 million procurement records from 140 countries in 8 languages into a single entity-resolved dataset.
Linking a contract award to the stock ticker the supplier traded under on the award date — not the ticker it trades under today. Companies rename, relist, merge, and delist; using today's ticker for a historical award introduces forward-bias into any backtest. In Valan's feed, ticker_as_of is populated only when a genuine dated listing window covers the award date (flagged pit_confirmed=TRUE); otherwise it is honestly NULL. The rule: never a current value dressed as as-of.
The contamination of historical data with information that only became available later. In procurement-to-equity mapping the classic error is stamping a 2019 award with the supplier's 2026 ticker or current parent ownership — a backtest built on such data trades on information that did not exist at the time. Valan separates the forward-bias-free field (ticker_as_of with pit_confirmed) from current-knowledge fields (ticker_current, ultimate_parent_ticker) so researchers choose explicitly.
An award whose supplier is linked to a tradable security — either its own listing or its ultimate parent's (investable_flag=TRUE). Of Valan's 71.9 million financial-modelled awards, 12.3 million (~17%) are investable. The link is made deterministically via LEI-bridged entity resolution, never via similarity matching.
Mapping the messy supplier names on contract notices — a single company can appear under dozens of spellings across portals and languages — to one resolved legal entity with stable identifiers: LEI, ISIN, ticker, exchange MIC. Valan resolves suppliers against a 72.4-million-row global entity master built from GLEIF, national company registries, exchange data, and curated sources, using deterministic exact and normalized matching. Similarity matching is never used for attribution, because it produces false joins.
A proprietary taxonomy of 666 sectoral codes that harmonises procurement categories across sources. National portals classify contracts in incompatible schemes (CPV in the EU, NAICS/PSC in the US, free text elsewhere); VPV provides one cross-source category per record, with crosswalk mappings to CPV, NAICS, ISIC, and UNSPSC.
The linkage of subcontract awards to the prime contracts they sit under, exposing the supply chain beneath the headline award. Valan's subcontract_graph carries 3.3 million sub→prime edges with tier depth, both entities, and the tradable ticker the prime ultimately rolls up to — so a revenue event can be traced from a tier-2 subcontractor to a listed parent.
IDIQ (indefinite delivery / indefinite quantity) and framework awards publish a ceiling — the maximum that may ever be ordered — rather than money actually obligated. Treating ceilings as spend massively overstates values, because the same ceiling figure is often repeated across many suppliers admitted to the vehicle. Valan flags these rows value_is_ceiling=TRUE (a top-0.1%-of-currency value repeated across ≥10 distinct suppliers) so analysts can exclude them from real obligated-value analysis.
Valan's design rule is honest-NULL over fabricated coverage: when a genuine value cannot be established — a ticker with no dated listing window covering the award date, a country that cannot be deterministically resolved, a stub value like 0 or -1 in a value field — the field is NULL rather than filled with a plausible guess. Fake coverage is worse than no coverage, because it silently corrupts downstream research.
Procurement data is a record of what governments are afraid of running out of. It reveals the demand side of state behaviour — stockpiling, rearmament, infrastructure buildouts, emergency responses — attributed to named, often listed, suppliers, with values and dates, published as a matter of law. Its structural blind spots are equally informative: activity routed through equity stakes or commercial offtake contracts generates no tender footprint, so the negative space in procurement data is itself a risk map. Valan's research series documents both edges.
A free 1,000-row sample with a reference Python loader is at valan.io/sample — no registration. The full feed (71.9M awards, 25M tenders, entity dimension, subcontract graph, daily refresh, S3 parquet with SHA256-checksummed manifests) is available by institutional arrangement: john@valan.io. Full schema: valan.io/data-dictionary.