The Arc: From Open Infrastructure to Regulated Scarcity
The 49 papers filed here trace a single overarching narrative: the collapse of an era of relatively open platform data access and its slow, contested replacement by regulated, technically constrained, and unevenly distributed alternatives. Freelon2024-sc provides the master periodization — API prehistory, laissez-faire, authentication, limited options, academic cooperation — while Giglietto2026-855a54cb compresses this into a memorable triad: “Wild West,” “Golden Age,” and “Walled Garden.” Bastos2025-ya and Murtfeldt2025-wu eulogize the Twitter API specifically, with the latter offering hard bibliometric evidence (a 13% publication decline in 2024) that this is not merely a felt loss but a measurable one. Yang2026-tq corroborates this at the level of the whole social-science literature: empirical social media research tripled from 2010 but plateaued and declined in 2022-2024, precisely as API restrictions bit. Bruns2026-yv’s essay on the “death” of Twitter as public-debate infrastructure gives this decline a substantive, not just methodological, register — the platform’s openness enabled a form of public communication that its enshittification under Musk destroyed, with successor platforms (Mastodon, Threads, Bluesky) failing to replicate the underlying moderation and governance commitments.
The DSA as Contested Remedy
Against this backdrop, the EU’s Digital Services Act — especially Article 40 — emerges as the central regulatory response, and a cluster of papers debates whether it is working. Ohme2026-nv, Pierri2025-hm, and de-Vreese2026-zx form a running conversation (explicitly staged as companion Forum pieces) about whether the DSA’s turbulence signals failure or the beginning of genuine constraint on platform power; de Vreese and Tromble argue for a “marathon” posture of sustained implementation rather than retreat. Philipp2026-tl and Lukito2026-nb apply this lens specifically to election research, with Lukito’s “price, proficiency, or permission” framework showing how access barriers reproduce WEIRD-centric inequities even under supposedly universal legal mandates. Peters2026-mo adds a crucial wrinkle: data quality, not just access, is politically contested terrain, with Meta strategically inverting researchers’ own arguments to claim its data is too poor to be worth granting. Crosset2026-mq situates the DSA within a broader comparative frame of security governance, showing how EU “flow optimisation,” German “patrolling,” and US “free circulation” regimes each bureaucratize platform oversight differently, while Bechmann2026-dr argues the DSA moment demands abandoning narrow causal-effects paradigms in favor of treating platform collective behavior as democratic infrastructure itself.
Auditing the New Infrastructure — and Finding It Wanting
A dense empirical cluster tests whether DSA-mandated tools actually deliver on their promise. The verdict is consistently skeptical. Entrena-Serrano2025-gw finds TikTok’s expanded Research API riddled with unexplained limitations; Rieder2025-ju shows YouTube’s search API is “forgetful by design,” with findable videos decaying within 20-60 days in ways that undermine both replicability and DSA systemic-risk research; Tonneau2025-bv documents stark linguistic inequities in moderator allocation across six platforms, with Global South languages receiving a fraction of English’s coverage; Jurg2025-ur audits YouTube’s election-period source-ranking and labeling practices, finding inconsistent application of transparency labels across languages and vague removal communications. Zheng2026-bi responds constructively by building TubeStats and TokStats, random-sampling tools that supply the representative denominators no official API provides. Bruns2026-pn and Cullen2026-cb extend this critique to Meta specifically: clean-room environments like the Meta Content Library, while addressing genuine privacy concerns, structurally privilege quantitative, code-fluent, well-resourced researchers, while CrowdTangle’s now-defunct API had already shaped what could be “seen” through its own technical and governance design. Schulte2026-df and Giglietto2025-ed60bc90 frame these accumulated bottlenecks as an infrastructure problem requiring standardization rather than one-off methodological fixes.
Meta’s Policy Shifts as Object of Study
A distinct but related thread uses newly available Meta datasets to interrogate Meta’s own governance decisions, turning the platform into both data source and subject. Giglietto2022-b30e8b4e and Rossi2023-847d5a9f work with the URL Shares Dataset, with Giglietto and Puschmann identifying an artifact in the 100-share anonymization threshold that produces spurious cross-country convergence — a methodological warning later echoed in critiques of clean-room data quality. Giglietto2026-632ef967 uses the successor Privacy-Protected Full URLs Dataset to show Facebook is an active curator whose partisan-penalty and quality-reward effects fluctuate with known governance interventions like the 2020 “break the glass” measures. Giglietto2025-1765bb4f applies breakpoint detection to the Political Content Reduction Policy itself, finding its Italian rollout preceded Meta’s announced global timeline by ten months and produced a 72% reach reduction for mainstream politicians while extremist accounts compensated via posting volume — a transparency failure documented using the very Meta Content Library whose limitations Bruns2026-pn catalogs. Hurcombe2025-cs and De2026-ld examine Meta’s (and other platforms’) discursive self-presentation, showing how Newsroom framing and “changecraft” narratives manage political and infrastructural change while obscuring power consolidation. Cazzamatta2026-lo extends this into the 2025 fact-checking rupture, empirically debunking Zuckerberg’s censorship framing while documenting fact-checkers’ anxieties about Community Notes — anxieties given empirical teeth by Bouchaud2026-np, which shows Community Notes’ bridging algorithm structurally undermoderates polarizing content by design.
Industry Entanglement and the Crisis of Independent Knowledge
A final cluster interrogates whether platform-dependent research can ever be trusted, given structural asymmetries of funding, data, and access. Bak-Coleman2025-pm and Bak-Coleman2026-mk draw the tobacco/pharma/fossil-fuel analogy explicitly, with the latter’s quantitative audit finding roughly 80% “industrial saturation” once undisclosed author, editor, and reviewer ties are combined. Heiss2026-qv frames this as a uniquely acute problem for platform research, since — unlike food or pharma — social media data can only come from the industry being studied. Munger2025-cz delivers the sharpest verdict on the paradigmatic Meta2020 collaboration itself: even at its most rigorous, platform-permissioned research suffers low “temporal validity” because platforms change faster than science can study them, implying that empirical social science is structurally outpaced and that proactive regulatory mandates (rather than one-off partnerships) are the only durable solution. Farkas2026-lr documents the analogous dependency struggles of fact-checkers, who rhetorically defend accepting platform funding as pragmatically necessary despite acknowledging its risks. Allen2025-ot offers a partial way out — platform-independent experimental methods via browser extensions — as a pragmatic middle path between full platform cooperation and total exclusion, a strategy that resonates with McNally2025-dn’s demonstration that algorithmic “black boxes” can be partially reverse-engineered through longitudinal engagement data without any platform cooperation at all, and with Bouchaud2026-lr’s reconstruction of X’s recommender embeddings from donated data alone.
Adjacent Currents
Several papers extend this data-access story into adjacent governance concerns: Ahuja2025-ku operationalizes DSA Article 25’s autonomy protections into a design-auditing framework for dark patterns; Moran2025-qn documents the parallel institutional collapse of the trust-and-safety profession; Ventura2026-yc and Swartz2026-zb push beyond data access per se into the downstream political and economic consequences of platform governance — partisan sorting under Brazil’s X ban, and the normalization of scam economies as a structuring logic of platform life; Schiffrin_undated-gi extends the liability-and-gatekeeper argument to AI-driven financial fraud. Iannelli2018-ebd918b7 and Inacio-da-Silva2026-zf (via Facebook ad tools and crowdsourced ad auditing) and Giglietto2020-6278a4aa (on micro-targeting and hybrid-media attention dynamics) offer earlier-generation examples of researchers improvising around platform opacity before the DSA existed — a reminder that the “Wild West to Walled Garden” arc has deep pre-regulatory roots. Votta2025-xz and Mahl2026-hc’s Delphi study round out the topic by gesturing toward the field’s own anticipatory self-assessment: platform governance, researcher access, and algorithmic opacity are rated among experts’ top intervention priorities for the coming decade, confirming that the questions animating this entire literature are expected to intensify rather than resolve.