The club that pays in code: pooling European demand for open source¶

A plain-language companion to the draft working paper "Buyers' Clubs for the Open-Source Flow: Collective Provision in the FLOSS Value Network" (v0.4, July 2026). The paper carries the model, eight proved propositions, the verification suite, and the survey design; this text carries the ideas.

Read the full draft working paper (PDF)

The short version. European supervisors now ask firms and public buyers to pool their demand and co-fund open-source alternatives to the dominant providers: five Dutch authorities put the proposal on record in 2026. The paper models the institution that proposal implies, a buyers' club for open source, and finds that open source changes the club's economics in ways that mostly help. Members can pay their dues in code as well as cash, which turns free-riding into a problem only small clubs have. A public anchor buyer of computable size solves the everyone-waits problem. Copyleft licensing quietly co-finances the club. Real clubs run on exactly these mechanics in Denmark and Wallonia, and the ones that dissolved failed the way the model says they should. The paper ships with the survey design that will test who joins and what stops the rest; the survey has not run yet, and no result is claimed.

Free code, unpaid bills¶

Open-source code is free, and that is precisely the problem the club solves. The code sitting in the repository (the stock) costs nothing to copy; what costs money is the flow: maintenance, security work, evolution, and, increasingly, the compliance evidence European regulation demands. The flow is chronically underfunded because everyone can use the stock without paying for the flow. A buyers' club funds the flow and contracts the services around it (assurance, adaptation, assistance, the "3A" services of Jullien and Zimmermann), which are excludable: you can be denied the guaranteed edition, the support contract, and the seat at the requirements table, even though nobody can deny you the code. That excludable service layer is the membrane that holds the club together; the co-funded code itself stays open to all, which also removes a fear that haunts proprietary co-funding, since no partner can lock the others out of the asset they paid for.

Two ways to pay, and who still rides¶

The paper's first departure from a generic pooling model: capable firms can pay membership in engineering work rather than cash. Real consortia look exactly like this, vendors and skilled users contributing code while the rest pay fees. Whether a capable firm joins as a code-paying member or free-rides (using the open flow without joining) comes down to a conversion friction: how much it costs to turn the firm's engineering into work the club actually needs.

The free-riding result follows. Riders exist only in a band where that friction is high relative to the fee, and the fee falls as the club grows, because the fixed flow bill is split among more members. So the rider band shrinks as the club scales and vanishes at a finite size: free-riding is a small-club, high-friction phenomenon, and scale converts riders into contributing members. The model also predicts the composition real clubs display, capable members paying in code and less capable, high-need members paying cash. A regulatory compliance burden sharpens the effect from one side: it makes going it alone more expensive, so it recruits capable firms into membership, while the same burden raises the club's bill and prices out members whose need was marginal.

The anchor, and the everyone-waits problem¶

Pooling has the classic chicken-and-egg structure: joining is worth more the more members there are, so "nobody joins" is always a self-fulfilling equilibrium, and a critical mass separates it from the self-sustaining club. The remedy has a computable size: a public anchor buyer big enough makes the club worth joining even when empty, which deletes the everyone-waits equilibrium outright, and the paper derives the threshold from the club's cost structure. The in-kind margin softens the problem from below (when compliance burdens exist, even tiny clubs are weakly viable, because near-costless code dues beat duplicating compliance alone), so the anchor's real job is jumping the critical mass for the cash-paying population.

Copyleft as club finance¶

The license regime enters the model as a policy lever. Under effective reciprocity (copyleft that actually reaches the deployment mode in use), the adaptations that self-serving riders build must return upstream, where they defray the very flow the club funds. Free-riding then becomes partially self-financing: the club's fee falls, and so does the gap between the membership the club reaches and the membership that would be socially best. That gap, the model's persistent inefficiency, never closes on its own; reciprocity shrinks it, and under a fully permissive regime it stands at full width.

The club in the wild¶

The institution is not hypothetical. OS2 in Denmark counts roughly 80 of the country's 98 municipalities and procures development and maintenance from an ecosystem of more than 60 vendors under standardized memoranda of understanding; its members that initiate and collaborate on projects are the in-kind margin in operation. IMIO in Wallonia is the club as its own supplier: more than 140 municipalities co-own the provider that builds and maintains their common components. ADULLACT in France and VNG in the Netherlands pool the support layer in association form. And the dissolved cases fit the model's failure mode: Malta and Iceland launched public adoption programs in the early 2010s that dissipated once support lapsed, which is what the model predicts when the anchor withdraws before a self-sustaining membership mass has formed. Six cases fitting the predicted partition is consistency, and the paper says so; the test is the survey.

The small print¶

The eight results are theorems about a deliberately small model, each with a written proof, a numeric cross-check against brute force, and a Monte-Carlo robustness run; club governance and the valuation of contributed code are assumed, and a club that cannot review and integrate contributions has, in the model's terms, a high conversion friction. The survey design elicits commitment under stated conditions, a conditional-cooperation ladder, and the constraint that binds for non-joiners, with the mapping from answers to policy instruments fixed before fieldwork; it has not been fielded, all its measures are stated preferences from a specialty sample, and no population share will ever be reported. One boundary holds throughout: the paper models who would join and why, and claims nothing about whether pooled demand actually scales European suppliers or delivers autonomy.