The framework that runs through everything, the doctoral thesis it came from, the papers in progress, the three lines I am opening now, and the instruments I built inside government.
Two years inside the state produced the reforms, the instruments and the data infrastructure. These are the three lines they make possible — one engine, three risks.
Spain’s 2026 extraordinary regularisation is a once-in-a-generation reform, and I helped design the analytical baseline for it. Legal status is a mark administered by the state; removing it should reduce employer uncertainty — a treatment on the second moment of statistical discrimination, not only on its mean.
One in two migrant workers in Spain is overqualified for their job. The fast-track recognition of foreign qualifications rolls out staggered across sectors and regions — an event study waiting to happen, and a rare chance to referee three theories that usually only argue.
Exposure indices ask what AI can do. EPOCH asks what will still be asked of humans, scoring occupations on empathy, presence, judgement, creativity and vision. I built its Spanish application inside government; the academic step is to formalise it and take it to trajectories.
Open a card to read what the paper does.
The theoretical spine of everything else: separating whether a disadvantage is unequally distributed from whether its penalty is unequally absorbed, and evaluating both through implication analysis rather than a single coefficient.
MapIneq Working Paper D3.3 · Horizon Europe
Harmonised European microdata for eight countries. Among those who do not make the transition to tertiary education, parental education cushions the risk of ending in a low-status occupation — moderation, now on labour-market ground. Where transitions are more selective, the cushion matters more.
MapIneq Working Paper D3.2 · Horizon Europe
Pooled European Social Survey, waves 1–10, at regional detail. Within-country variation rivals variation between countries: Spain is among the most internally heterogeneous, and among the steepest decliners over time. Territory is where institutional settings vary inside a single country.
Spain runs one of the OECD's most retention-intensive systems. Gender and birth month — traits exogenous to family background — identify who is exposed; an instrument built from regional retention rules identifies the penalty. Low-SES students face both a higher risk of repeating and a larger penalty when they do: exposed below, sheltered above.
UK cohort data through the sixth-form transition. Sex at birth is not sorted by family background, so exposure is clean by design. High-SES families moderate the male penalty at the transition — net of prior cognitive and non-cognitive skills. The male disadvantage in education is a class phenomenon.
German cohort data, entry age made quasi-random by birth-month rules, two independent identification strategies that converge. There is a clear early-entry penalty and no compensation by family background — the honest null that gives the framework its scope conditions.
What I built at the Presidency of the Government — the instruments, not just the reports.
Spain's first birth-cohort longitudinal study
Register-based sampling, linkage between administrative records and surveys, intergenerational by design. I conceived the idea, designed the linkage across existing surveys and registers, led the international expert meetings and secured the tri-party State agreement between the Foresight Office, INE and CSIC. Governance modelled on the Millennium Cohort Study, NEPS and the PSID.
Spain's national AI foresight strategy
Lead author of the chapter on labour, education and welfare. With the MIT authors of EPOCH we mapped AI complementarity onto the Spanish occupational structure: roughly 65% of workers are in occupations where AI plausibly complements rather than substitutes. That instrument is now the foundation of my research on AI and stratification.
Directorate for Public Policy, Presidency of the Government
The extraordinary regularisation (RD 316/2026): I led the technical, ex-ante analytical work, including the child-protection provisions and the application form that captured applicants' socioeconomic baseline. The Plan de Integración y Ciudadanía: four pillars, sixteen measures, ten quantified targets for 2030. The organic law on minors in digital environments: evidence base and coordination on age verification, algorithmic harms and digital literacy.
Infrastructure I want to build next, and the questions each of them opens.
Formalise the crosswalk from O*NET occupations to the Spanish classification, validate the scores against existing exposure indices, and publish them with documentation so other researchers can use them. The measurement piece that everything else in the AI line rests on.
Administrative processes generate enormous amounts of conversational data — helplines, casework, chat — held by the organisations that accompany people through them. With consent designed at the intermediary and LLM-assisted coding validated by hand, that becomes a way to hear why people who are entitled to something never claim it. Reusable, and a research infrastructure a department keeps.
Two routes into digital-trace data: consented downloads under the right to portability, and vetted-researcher access to platform data under Article 40 of the Digital Services Act — a regime I know from the inside after writing the evidence base of Spain's law on minors online. A first pilot on youth digital environments.