My PhD research asks whether content-oriented Very Large Online Platforms and Search Engines, the kind the EU Digital Services Act now classifies as systemic risks to public discourse, are quietly hollowing out democratic functions in the Nordic states. The question is empirically framed to a problem with high socio-political stakes.

To answer it I need a framework that does three things at once: a) tells me what democracy is supposed to do, b) locates where platform power gets in, and c) explains why those interventions tip from local nuisance into systemic damage. For a long time I had two of those three elements. Last month, through a very strange route, I found the third element, the missing piece of my puzzle. This post is about that route, and about the framework now sitting on my desk, demanding attention.

How I fell for Warren’s framework

Warren’s democratic framework is a familiar evaluation tool by now. I found it through Digital Threats to Democratic Elections, a report Chris Tenove led at UBC’s Centre for the Study of Democratic Institutions, and the subsequent McKay & Tenove piece on epistemic harms. Tenove et. al applied Warren’s problem-based approach to organize the question of how digital interference damages elections, and the framework did real analytic work where most of the field was still gesturing at “harms to democracy” without specifying which democracy, which harm, or to which function. That clarity became the spine of my MA thesis, Cyberthreats to Democracies: Constructed Threats to Democratic Functions where I examined how Russian cyberattacks impacted UK and US 2016 elections. My MA thesis used Warren’s three functions to discipline the threat-construction analysis, which is to say I have been working inside this framework for years.

The PhD scales it up: from threat construction within platform-systems to platform-mediated structural distortion across all five Nordic democracies.

The Warren matrix, as I have built it out

Mark Warren’s problem-based approach to democratic theory walks away from the old Schumpeter-Dahl move of defining democracy as a particular procedure (elections, deliberation, take your pick) and instead asks what problems a political system has to solve in order to count as democratic at all.

Warren arrives at three functions: empowered inclusion of all affected interests (w-1), collective agenda and will formation (w-2), and collective decision-making with state capacity to back it up (w-3). Across these three functions he then maps seven democratic practices: recognition, resistance, deliberation, representation, voting, joining, exiting. Three by seven. Twenty-one cells. Each cell a democratic norm precise enough to operationalize. As an old data-nerd and software developer I loved the organization and immediately saw both how this could be coded into an interactive tool and how data-science and AI could be used to empirically back up democratic developments.

I have spent more time with that matrix than I would care to admit. All twenty-one cells are now populated with the constitutional and statutory norms of the five Nordics, with Grágás supplying the medieval Icelandic anchor for the rule-of-law cells (because if you are going to write about Nordic democracy from Reykjavík the Viking-justice reference is mandatory). For each cell I am asking the same question: under platform pressure, is this norm sustained, distorted, co-opted, or contested? Recently Warren theorized a brilliant article with Roskilde’s Eva Sørensen, formulating a robustness framework which gives me five institutional conditions (scalability, modularity, experimentalism, decentered autonomy, polyvalence) that specify what it actually means for a democratic practice to hold up under stress. V-Dem and established public indexes give me the population-level indicators against which Nordic trajectories can be measured.

Parsons, and what P-4 actually means

Sitting on top of all this is a deductive overlay drawn from Craig Parsons’s How to Map Arguments in Political Science. Parsons argues that any causal claim in political science can be sorted into one of four logics depending on what kind of thing is doing the explanatory work. Structural explanations (P-1) point to material and positional constraints: geography, resources, the distribution of power. Institutional explanations (P-2) point to rules, procedures, and the path-dependent settlements that channel behavior. Ideational explanations (P-3) point to beliefs, ideologies, and the cultural scripts actors carry around in their heads. Psychological explanations (P-4) point to the cognitive and emotional architecture of human minds: attention, reciprocity instincts, status-seeking, the heuristics we run on when we cannot run on deliberation.

This typology matters because platform distortion is not one kind of cause. A platform’s market dominance is a structural fact (P-1). The DSA’s Article 34 risk assessment regime is institutional (P-2). The narratives platforms amplify and the ideological feedback loops they enable are ideational (P-3). And the way a recommender system directs attention, flattens reputation signals, and rewires reciprocity at scale, that is psychological (P-4).
Warren’s 3×7 matrix tells me where distortion happens. Parsons tells me what type of cause is doing the work.

Most of the cells in my matrix are mixed, but a particular row, the P-4 (psychological factors) across Warren’s twenty-one cells, is where platform power touches the user’s mind directly. This is the layer Zuboff names as instrumentarian power but leaves unformalized: her 2022 article Surveillance Capitalism or Democracy? develops the P-1 structural and P-2 institutional conditions of surveillance capitalism in depth, with the P-3 ideational scaffolding sketched and the P-4 mechanism asserted rather than given mathematical form. That is the row where I felt the gap most acutely. Description wasn’t enough, I needed a mechanism to falsify or prove it.

How I encountered Nowak

Here is the strange route. I have been in conversation with a couple of AI developers (creators of StoryPrism) who have been working through the Epstein files released under the 2025 Transparency Act, specifically the trail of scientific patronage. Which researchers did Epstein fund? Which institutions took the money? Which intellectual projects he chose to underwrite – and why. The exercise is interest-based forensic analysis of power-relations, classic political science nerd-ism. It is an attempt to map the intellectual ecosystem a sex offender cultivated around himself at the world’s most prestigious universities.

One name kept surfacing: Martin Nowak, the Austrian-born Harvard mathematical biologist who ran the Program for Evolutionary Dynamics on roughly $6.5 million of Epstein money between 1998 and 2008, who was sanctioned by Harvard in 2021, reinstated in 2023, and as of this February placed on administrative leave a second time after his name turned up more than 8,000 times in the new document release. Uncomfortable provenance. I will not pretend otherwise. But the science is separable from the circumstances of its funding and usable for this lone democratic academic just like the power-brokers that backed Epstein. I’m co-opting their funding and findings for my work for the benefit of the greater good.

I dove into the book SuperCooperators, Nowak’s 2011 trade synthesis with Roger Highfield, mostly out of curiosity. What I found was the missing layer of my methodology, sitting there in plain language.

What Nowak adds

Nowak’s evolutionary cooperation theory specifies the mathematical conditions under which cooperation in a population either stabilizes or collapses. Five mechanisms: kin selection, direct reciprocity, indirect reciprocity, network reciprocity, group selection. Each carries a threshold. Cross it and defection sweeps the population. Cooperation does not erode politely. It tips. The theory is content-neutral, which is the point. It applies wherever agents face mixed-motive interactions at scale, and that is precisely what billions of users do every day on a content-oriented VLOP.

The breakthrough, for my purposes, is locating Nowak’s mechanisms inside the P-4 psychological row of Warren’s matrix at the cells where platform-mediated interaction reaches population scale. Three cells carry most of the load: deliberation, recognition, and voting, all under P-4. In each, the platform’s recommender system, engagement-optimized ranking, and reciprocal-visibility architecture function as parameter modifiers on Nowak’s reciprocity mechanisms.

Indirect reciprocity, the mechanism Nowak identifies as foundational for reputation-based cooperation in large groups, is the most exposed. When platforms substitute algorithmic visibility for organic reputation transmission, the indirect reciprocity threshold shifts. Cooperation does not erode gradually. It collapses at the threshold. This is why platform-induced democratic distortion presents as sudden tipping rather than linear decline, exactly the pattern Lorenz-Spreen and colleagues document across democratic indicators.

The methodological yield is precise enough that I can finally explain it without arm-waving. Nowak’s threshold conditions function as mechanism-level proxies for the psychological micro-foundations of the six matrix cells where platform distortion is most likely to scale. Sørensen and Warren’s robustness conditions specify what the institutional environment has to do to keep those thresholds from being crossed. V-Dem indicators register whether they have been crossed at the population level. Parsons P-4 names the type of cause. Warren names the function and practice at risk. Norm to mechanism to indicator, no remainder.

Why this matters for the Nordic case, and why I am shouting about it

The Nordic democracies are unusually well-instrumented for this kind of test. V-Dem scores still high, constitutional protections explicit and litigated, and the DSA implementation environment converging through EEA transposition for Iceland and Norway and direct application for Denmark, Finland, Sweden. Falsification conditions are favorable. If platform pressure crosses Nowak thresholds in the cells the matrix flags, V-Dem trajectories will register the shift and DSA Article 40 data access will let me verify mechanism. If they do not, the framework returns a null finding for the Nordic case and the question becomes which institutional features held the line.

Inserting Nowak into Warren’s matrix at the P-4 row is, as far as I can tell, novel. It supplies the missing causal layer between democratic theory and platform empirics without importing assumptions foreign to either tradition. The matrix stays Warren’s. The thresholds stay Nowak’s. The bridge is structural.

That this bridge arrived through a forensic walk through a predator’s patronage network is, I think, worth saying out loud. Intellectual genealogies are rarely clean. Doing the work of separating the science from the circumstances of its funding, neither sanitizing the funding nor discarding the science, is itself a small democratic practice. It is the kind of thing this dissertation is, in the end, about.

What comes next: from indicators to raw platform data

The PhD itself will run on V-Dem and the established democratic indices. Those datasets are robust, peer-reviewed, comparable across the Nordic five, and sufficient to test whether the matrix’s predictions hold at population level. They tell me whether thresholds have been crossed. What they do not tell me, and cannot tell me, is what the platforms themselves were doing in the months and weeks before the indicators moved. For that you need the platforms’ own data.

This is where Article 40 of the DSA becomes interesting. Article 40 obliges Very Large Online Platforms and Very Large Online Search Engines to grant vetted researchers access to their internal data for the purpose of studying systemic risks to public discourse, including the exact category of risk my framework is built to detect. Recommender system parameters. Ranking signals. Engagement-weighted distribution data. The actual machinery sitting underneath the indirect-reciprocity collapse my matrix predicts. Article 40 was written precisely so that researchers like me would stop having to infer mechanism from outcome and could instead observe mechanism directly.

The post-doc project I have my eye on does exactly that. Take the Nowak-inside-Warren framework the PhD will validate against population indicators, and run it back through raw platform data obtained under Article 40. Test the threshold conditions on the actual reciprocity signals the platforms record and modify. Move from “the indicators shifted, the framework predicted it” to “the recommender parameters changed on this date, the indirect-reciprocity threshold was crossed in this cohort, the V-Dem indicator registered the shift four months later.” That is the level of mechanism-to-outcome traceability democratic theory has not had access to before, and it is the level the DSA was designed to make possible.

My PhD underpins the instrument, the digital-dashboard software that I’m coding and populating. The post-doc work takes it to the source. The framework is near-ready in my head. Now I need to wrap up my thesis as fast as I can, create that interactive digital-dashboard design and get to the raw data that allows us to analyze what is happening to our public discourse and democracy in real-time.

Warren’s functional democratic matrix

---

---

Practice	Empowered Inclusion (w-1)	Collective Agenda and Will Formation (w-2)	Collective Decision-Making (w-3)
Recognizing (w-x-1)	Equal moral standing; membership in the demos; non-discrimination; AAIP, those affected have a right to be acknowledged as participants. (w-1-1)	Mutual recognition as the functional precondition of deliberation; recognition of plural perspectives as legitimate inputs to collective agenda formation. (w-2-1)	Acceptance of collective decisions as binding, grounded in recognition that the process fairly considered the interests and perspectives of those affected and aimed at the common good. (w-3-1)
Resisting (w-x-2)	Right to dissent, organize opposition, and contest exclusion; protection of minority rights against majority override; freedom of assembly. (w-1-2)	Responsiveness to reason through organized pressure; right to contest agenda-setting power and challenge dominant narratives. (w-2-2)	Accountability of collective decisions to organized dissent; protection against capture of decision processes by factional interests. (w-3-2)
Deliberating (w-x-3)	Empowerment through education and voice; freedom of expression and freedom of thought as preconditions of meaningful participation; epistemic plurality. (w-1-3)	Formation of collective agendas through reasoned public justification; freedom of thought, expression, and access to information as deliberative infrastructure; Habermas’s public and private autonomy as co-conditions. (w-2-3)	Deliberatively achieved agreements as sources of decision legitimacy; throughput legitimacy, efficacy, accountability and transparency as conditions of binding decisions. (w-3-3)
Representing (w-x-4)	Political equality and inclusive suffrage; demos membership of those affected by collective decisions; the right to represent one’s own perspective and be counted; authorization: constituencies empowering representatives to act on their behalf, including marginalized groups and perspectives. (w-1-4)	Perspective-taking and agenda formation across constituencies; free media and epistemic plurality as conditions of informed representation. Undistorted information ecosystem for meaningful evaluation of reputation. (w-2-4)	Legitimate representatives chosen by the demos (input); accountable, transparent, and non-corrupted governance processes (throughput); decisions enacted through public authority and delivered as collectively willed (output). (w-3-4)
Voting (w-x-5)	Equal and universal right to vote; inclusive citizenship; all affected have a place at the table; electoral system free from structural distortions; votes carry equal weight. (w-1-5)	Voting as preference expression depends on deliberative conditions so citizens can know their own interests and rank choices meaningfully; undistorted perception of interests needed for votes to express genuine collective will. (w-2-5)	Majority rule as a decision rule; due process and rule of law as conditions of enforceable collective decisions; protection of minority rights against majority override. (w-3-5)
Joining (w-x-6)	Freedom of association and assembly; AAIP operationalized through civil society; self-sovereignty as the basis for voluntary collective action. (w-1-6)	Formation of public agendas through organized civil society; pluralism and separation of power as structural conditions for associational agenda-setting. (w-2-6)	Good governance and accountability as conditions under which associational governance partnerships operate democratically; prevention of capture by private or factional interests. (w-3-6)
Exiting (w-x-7)	Empowered choice as an individual empowerment mechanism; privacy and freedom of thought as conditions of autonomous exit decisions. (w-1-7)	Exit as a signaling mechanism that feeds back into collective agenda formation; freedom of expression as the public counterpart to exit. (w-2-7)	Accountability through competition and choice; rule of law as the framework within which exit options are legally protected and enforceable. (w-3-7)