Not an unlucky one.

Every year, governments spend trillions buying everything from pencils to highways. The system is elegant: companies compete, lowest price wins, taxpayers get a good deal. That's the theory.

Enter the cartel. Companies secretly agree who wins each contract. They rotate. They inflate. And the public — foots the bill without a clue. Half a century after Posner documented it, most cartels remain undetected.

The best detection tools — Imhof screens, variance tests, kurtosis — require granular bid amounts. But most procurement systems, especially in developing countries, only record who showed up and who won. That's it. Dead end? Not quite.

São Paulo, Brazil. 41,000 firms bid on government contracts between 2009 and 2019. Among them, 16,843 never won. Not once. And within those eternal losers, 2,735 bid with an intensity that defies economic rationality. One company entered 97 tenders over a decade. Won zero. Every bid costs money. Zero wins means negative expected profit. So why do they keep coming back?

Brazilian law requires at least 3 bidders for sealed-bid contracts. Cartels need bodies to fill the room. Cover bidders show up, submit comically high bids, lose by design, and collect a side payment. They're not competitors. They're extras in a rigged play. And here's the key: they must participate. That footprint is all we need.

Win rate = zero. Participation count above the IQR outlier threshold. That's it. Needs only who bid and who won. Any procurement agency can run it with a spreadsheet.

Same product. Same year. Same purchasing unit. When a frequent loser appears, prices run 3.6 to 7.7 percent higher. Cross-fit, OLS, matching, IV — four different estimation approaches, same story. That's not noise. That's a premium somebody is paying.

Brazil’s competition authority, CADE, has convicted 65 procurement cartels. Against that ground truth: AUC = 0.94 — though volume contributes substantially (within participation-count bands, AUC drops to 0.54). The screen’s detection power comes from identifying the high-participation tail of always-losers. A count-matched permutation test (2.2×) confirms the FL classification adds signal beyond volume alone.

Here's the twist that makes the paper click. FL bids are less dispersed than genuine bids. They cluster tightly above the winner. This is coordinated cover bidding — and it means variance-based screens lose power exactly when coordination is strongest. A participation-based screen sidesteps this completely. It doesn't care how they bid. It cares that they showed up at all.

The premium is almost three times larger where cover bidding is purely strategic. When no law forces you to show up and you show up anyway — and lose, every time — that's not compliance. That's strategy. And when we interact FL with a constraint-binding indicator, the entire effect comes from voluntary participation: a 7.6% premium.

FL firms don’t displace genuine bidders — they add to participation. Winners bid closer to reference prices. FL–winner pairs recur far beyond chance. And firms in FL-exposed markets exit less, not more. Each fact alone has benign readings. Together? Hard to square with anything but coordinated cover bidding.

The FL screen and bid-level tools capture almost entirely different information. Correlation of just 0.06. In a horse-race regression, the FL coefficient actually rises when you add the Imhof flag. They're complementary. The paper proposes using them sequentially: FL first to triage, then bid-level forensics on the flagged subset.

A 12.5× gradient across purchasing-unit size. Where oversight is weakest, the screen signal is strongest. This matches the framework's prediction: cover bidding is most profitable where detection probability is lowest. The screen bites hardest where it's needed most.

Not a conviction machine — a flashlight in a dark room. From 41,000 firms to 2,735 flags. From millions of tenders to the ones that matter. It tells investigators exactly where to start looking.

The diagnostics don't prove the mechanism.
They make the case for deploying the screen.