Game theory and mechanism design¶
This thread models the instruments themselves: what procurement screens can screen, how the layered structure of the cloud market defeats the obvious levers, and why the same Union wrote opposite rules for Chinese telecom vendors and US cloud providers. Every proposition carries a numerical check in the project's verification suite, and the interactive models recompute the verified code live.
Screening for Sovereignty¶
The core mechanism-design paper: procurement when compliance is cheap to fake and costly to verify. Five theorems, among them: a requirement screens only if it targets an attribute the genuine and the pretender satisfy at different cost and the buyer can verify (residency and operational audits fail outright, at any price and any audit intensity); the cost-minimizing audit is a cascade over attributes; and the accuracy a screen needs is set by the richest would-be pretender. Fifteen-plus numerical checks, all passing. Companion explainer: The subsidiary passes every audit.
Cheaper and Losing: A Layered Model of Cloud Dependence and the Limits of Subsidy¶
Why several-times-cheaper European cloud keeps losing share: buyers sort by how much of their needs are differentiated, the differentiated layer runs on network effects, and firms drift up-stack over their lifecycle, so the US share climbs on its own. A commodity subsidy saturates on the measured, bimodal firm population; the levers that work are un-bundling and cutting cross-provider coupling costs, which the Data Act's egress rules do. Seven propositions, proved, cross-checked, and Monte-Carlo stress-tested. Play it live. Companion explainer: Cheaper, and losing. Résumé en français: Moins cher et pourtant perdant (PDF).
Exit or Own? Portability Mandates and Ownership Screens¶
Two policies usually argued as rivals are complements with an exact band: an exit mandate cuts the lock-in rents a fake champion would capture, so the ownership screen needs less audit accuracy. For a whole range of realistic verification capability the screen works only with the mandate beside it. Below the band, neither reaches the exposure: portability there is a consolation, and calling the workload protected is a category error. Companion explainer: Portability or ownership screens? Wrong question.
Two Screenings: Why Europe Screened Huawei but not the Hyperscalers¶
The comparative case: against Chinese vendors, structural criteria passed at design and eroded in national enforcement; against US cloud, structural criteria were deleted in a low-salience drafting room and behavioural non-screens remained. The explanation runs on two variables, verifiability of the decisive attribute and where the losing coalition can spend its influence, and it makes three predictions, each with a live falsifier, confronted with the post-2024 record. Companion explainer: One Union, two opposite screens.
The Unpriced Bundle: Decision Suppression in Software Procurement¶
Why an organisation keeps a bundled component whatever the merits of the alternatives, before any question of switching cost arises: a component that arrives inside a licence at zero incremental price generates no budget line, hence no renewal event, hence no comparison. The model prices the internal champion who would have to convene one, and three results follow. A challenger both better and cheaper than the component's stand-alone value is suppressed whenever the champion's stake falls short of the initiation cost; conditional on non-initiation, adoption is invariant to the challenger's price, so subsidy is inert until it covers a buyer's whole wedge gap, which distinguishes suppression from switching costs empirically; and instruments that re-create the decision event (line-itemised pricing, a periodic evaluation duty, disclosure of the imputed price) restore the price margin for every buyer at once, at near-zero cost. The 2025 Microsoft Teams commitments are those instruments entering the record. Every proposition carries a proof, a numeric cross-check, and a Monte-Carlo report; no identification is claimed. Companion explainer: The product with no budget line.
Leverage Without Additionality¶
What the "leverage factor" of a public technology programme identifies. Under a published co-funding schedule the ratio is a deterministic function of that schedule, so a programme funding half of each project reports a leverage of exactly one however its beneficiaries behave. It bounds the crowding-in ratio from above, with equality only if no beneficiary would have invested anything absent the grant, which reverses the direction of the caveat funders attach to it. Compliance cannot push it past a computable ceiling, so only observations above that ceiling carry behavioural information. And because declared co-funding pools public with private sources, a rationale resting on private capital-market frictions is addressed only by the private component. Applied to the Digital Europe Programme's published tables: the headline 0.79 identifies the share of the envelope disbursed at full funding rates and nothing about behaviour, one technology area of thirteen clears the ceiling, and the private component is a minority of reported leverage in nine of twelve areas. Ordering programmes by the reported ratio inverts the ordering by true additionality on a quarter of sampled pairs. Companion explainer: The leverage that was already in the rules.