About Awantia

What Awantia is

Awantia makes public procurement traceable: who won which public contract, at what value, and when it expires. We ingest the official notices, resolve the endlessly varying spellings of a company name into one entity, and compute what the notices only imply — the contract end date, for instance, or how many bidders a contract attracted.

Germany is covered today, and the corpus grows every night with the new notices. The processing is not tailored to one country: further EU countries are a setting, not a rewrite.

Where the data comes from

The source is Tenders Electronic Daily (TED), the notice database of the Publications Office of the European Union. We process the machine-readable eForms documents, not their summaries. Reuse of those documents, including for commercial purposes, is expressly permitted by Commission Decision 2011/833/EU.

Every published figure points back at the document it came from, and every award page links the original notice. That original is stored unchanged and never overwritten. When our reading of it changes, the correction is carried forward — traceably, not silently.

How we handle bad data

Official data is not the same thing as clean data. The notices are free for anyone to download; what is not free is reading them correctly. What we ran into in the first months — and what we now catch:

• A hospital catering contract was filed at over four trillion euro — more than 90% of German economic output. That single row accounted for 99.7% of our total spend figure at the time. Since then we treat values below one euro and beyond ten billion euro as placeholders or filing errors, not prices.
• Some senders restate the framework value on every single lot; one notice was counted six times over. Such values are labelled on the page and enter no total.
• Zeroes and amounts like €0.01 are placeholders, not prices. They drag every average down, so they stay out.
• Joint awards are apportioned across the winning companies. One supplier on a four-way framework has not won the framework.
• Every notice mints fresh identifiers for the same companies. Without entity resolution a single supplier would shatter into hundreds of records. We merge — but only on corroboration beyond name similarity: “SG 2” and “SG 3” are two departments, not a typo.
• Market-size totals count final contract values only.

The rule behind all of it is precision over recall. A wrong merge permanently corrupts the figures of two real companies; a missed one only thins a page. Doubtful cases therefore go to a review queue rather than straight into the corpus.

What we do not know

We write down what we do not know. These limits exist today and no amount of better code removes them:

• Not every notice states a contract value. Many buyers publish none, or redact it — large institutions especially. That is the behaviour of the source, not a gap in our parsing. Where the value is missing, no value is shown; we do not estimate it.
• Contract end dates are frequently absent. Where we derive one from the median duration of comparable contracts in the same sector, the figure is labelled “estimated”. An estimate is never presented as a stated fact.
• Corporate group membership is currently detected from name patterns. Real ownership data from registers is missing, so statements about company groups should be read with that in mind.
• Contracts below the EU thresholds are not published EU-wide and therefore do not appear here either.

Our rules

Four rules every page of this service has to hold up against:

• No number we cannot defend.
• Provenance and confidence travel with every derived fact. “Stated” and “estimated” are different claims and are never rendered the same way.
• Source documents are never overwritten. Corrections are carried forward.
• No language model originates a number, a date, a winner or an ownership link on these pages. Facts come only from verifiable, deterministic code.

Corrections and contact

Found an error? Write to kontakt@awantia.com with the address of the page. We check the entry against the original notice and correct it if we are wrong. What we cannot do is change the notice on TED — it stays public there unchanged.

Provider details are in the imprint, data processing in the privacy notice. To get into the data, start with search.