Government Tender Awards: Winners, Values & Bids

Every other government tender Actor gives you the contracts you can still bid on. That is the easy half, and it is the half that tells you nothing about whether bidding is worth your time.

This one reads the other end of the process: contracts that have already been awarded. Who won, how much they were paid, how many bidders they beat, and which buyers keep going back to the same supplier — from the US, UK and EU registers, in one schema.

Four of the tender Actors currently on the store were read line by line before this was built. All four list open opportunities; none of them returns an award result. One even ships an awardedTo field with nothing in it.

The finding this was built around

A US contract can be labelled FULL AND OPEN COMPETITION and receive exactly one offer. The label describes the procedure. The offer count describes what happened. When they disagree, that is the most useful line in the dataset: a market where the paperwork says anyone could bid and, in practice, nobody else did.

Measured on 12 US software contracts while this was written — three of them:

Winner	Label on the contract	Offers received
CARAHSOFT TECHNOLOGY CORP	FULL AND OPEN COMPETITION	1
V3GATE, LLC	FULL AND OPEN COMPETITION AFTER EXCLUSION OF SOURCES	1
THUNDERCAT TECHNOLOGY, LLC	FULL AND OPEN COMPETITION AFTER EXCLUSION OF SOURCES	1

Those rows carry uncontestedOpenAward: true and are collected in the UNCONTESTED_AWARDS record. A contract that was openly advertised and drew one bid is either a market nobody knows about or one nobody thinks they can win. Both are worth knowing before you write a proposal.

A sole-source award that received one bid is not flagged. Its label never claimed competition, so there is no contradiction — it is reported as single_bid_received instead. The two are kept apart on purpose.

What it will not do

It never fills a blank with a zero. Award registers are full of holes. A UK buyer often publishes the winner and withholds the price; a US contracting officer often leaves the offer count empty. Writing those as 0 would turn we do not know into nobody bid, and the price is secret into the contract was free.

So every uncertain number ships with the boolean that says whether it was published:

Field	Companion
awardAmount	amountDisclosed
bidsReceived	bidsDisclosed
singleBidder	null when the count is unknown — never false

The disclosure rates are an output in their own right. Over one live run of 177 awards: the value was published on 47 and the bid count on 7. That is not a defect in the data collection; it is what the registers publish, and a tool that hid it would be lying by omission.

It never adds currencies together. GBP, EUR, USD, CZK and PLN awards all appear in the same run. Totals are kept as a map per currency, because converting them would need an exchange rate this Actor has no business inventing.

It never credits a shared contract twice. When three firms win a contract jointly, each is credited with a third of the value in the supplier ranking, not the whole amount.

What comes out

One row per award, plus four aggregate records.

Dataset row

Field	Meaning
source, country, sourceUrl	Which register, and a link straight to the original notice
buyer, buyerUnit	The contracting organisation
suppliers, supplierCount, supplierCountries	Who won, including joint winners
awardAmount, currency, amountDisclosed, amountBasis	What it was worth, in the register's own currency, and what the number actually measures
bidsReceived, bidsDisclosed, singleBidder	How many bids came in
competitionLabel, uncontestedOpenAward	The procedure claimed, and whether it matches the bid count
lowestBidAmount, highestBidAmount	The bid range, where the register publishes it
awardDate, contractStart, contractEnd	Dates, normalised to yyyy-mm-dd across three different formats
awardsOnThisNotice	How many awards the parent notice holds — a framework can hold over a hundred
classification	NAICS/PSC for the US, CPV for the EU
issues	Machine-readable flags: open_competition_with_one_bid, award_value_not_published, …

Key-value records

Record	What it answers
TOP_SUPPLIERS	Who keeps winning, across how many buyers, for how much per currency
BUYER_PROFILES	Per buyer: repeatWinnerShare — how often its awards go to its single most frequent supplier — plus its disclosure rates
UNCONTESTED_AWARDS	The openly-competed contracts that drew one bid, biggest first
SUMMARY	Counts, disclosure rates, the exact query sent to each register, and every caveat that applied to this run

The three registers

	Winner	Value	Bid count	Keyword filter
US — USAspending.gov	always	always	in about half the awards	server-side
UK — Find a Tender	almost always	sometimes	rarely	applied here, after fetching
EU — TED	always	always	never published	server-side, full text

All three are official public APIs and none of them needs a key. Nothing here is scraped: no HTML is parsed, no login is used, no page is crawled.

Three differences worth knowing before you read the numbers:

• US amounts are obligated dollars, not a bid price, and every US row says so in amountBasis. Bid counts live in a separate detail endpoint, which costs one extra request per award — that is what includeUsCompetitionDetail controls.
• UK dates are notice-publication dates, not award dates; where the award itself carries a date it is used, and ukAwardDatePublished says which one you got.
• EU rows are one per notice, not one per award. TED returns lot results as parallel arrays with no reliable pairing between a winner and a lot, so a multi-lot notice lists every winner and states in amountBasis that the amount covers all of them. Splitting it would mean inventing a breakdown the publisher did not give.

Three things real data changed

Unit tests pin the shapes. These only appeared once the code met the registers.

1. The obvious UK filter hides every award that has a price. Find a Tender accepts stages=award, which looks like the correct way to ask for results. Measured over one window it returned 486 awards — all under the old EU directive, none with a published value. The unfiltered feed over the same window returned 435 awards of which 118 had values and 5 releases carried bid lists, because notices published under the UK Procurement Act 2023 are missing from that filter entirely. Using it would have made this Actor report that the UK never discloses what it paid. The filter is applied here instead, on the release tag.

2. One framework notice can swallow the whole run. UK notice 071472-2026 holds 127 lot awards on its own — enough to fill the record budget on the first page and hide the other 1,283 awards in the window. maxAwardsPerNotice bounds any single procurement, whatever was left out is counted in SUMMARY, and every row carries awardsOnThisNotice so the real size stays visible.

3. The same award is republished every time its notice is updated. In one 400-row sample, 32 rows were repeats of an award already collected — enough to inflate every supplier's win count. Awards are now deduplicated per procurement, keeping the most recent copy.

Limits

• Contracts only. Grants and loans are out of scope: they have no bidders.
• No bid counts for the EU. TED's search API exposes 1,830 fields and the received submissions count is not one of them. bidsReceived is null for every EU row rather than guessed from the tender value range.
• The UK has no keyword parameter, and its feed is rate limited. Keywords are matched here against title, description, buyer and winner, so a UK run reads the whole window looking for them — one test run scanned 3,400 releases to find 3 awards mentioning "cybersecurity". Find a Tender allows 12 requests per 90-second window per IP, and an Actor shares its egress with other tenants, so pages are spaced 8 seconds apart and reading stops after 40 pages with the truncation reported. Narrow the date range rather than relying on the keyword when you want UK coverage.
• A refused page never discards the pages before it. If a register starts refusing mid-run, what was already collected is kept, the reason is written into SUMMARY, and the run is marked truncated.
• repeatWinnerShare is a share, not a verdict. A buyer with three awards and one supplier scores 1.0 on very little evidence. Read it next to awards.
• Disclosure rates describe the awards this run collected, not the register as a whole.

Example input

{
  "sources": ["us", "uk", "eu"],
  "keywords": ["cybersecurity"],
  "awardedFrom": "2026-06-01",
  "awardedTo": "2026-07-31",
  "maxRecordsPerSource": 200,
  "includeUsCompetitionDetail": true
}

To go straight to the contradictions, set onlyUncontested to true. The rankings in the key-value records are still built from every award found, not only the flagged ones.