Comparative Platform-Transparency Regime Coding: 18 Jurisdiction-Units, Four Dimensions (2026)

This dataset contains an original comparative coding of platform-transparency law across eighteen jurisdiction-units, produced for a study of institutional delivery capacity as a constraint on transparency regulation. All values were coded from primary legal sources as in force on 2 September 2026. Units. Coding is at the level of the legal instrument rather than the country, because several jurisdictions regulate through instruments with materially different logics. The reference tier comprises the EU Digital Services Act, the EU General Data Protection Regulation, the UK Online Safety Act 2023, and the UK GDPR/Data Protection Act 2018, each coded separately. The emerging-market panel comprises Brazil, Mexico, Colombia, Chile, India, Bangladesh, Indonesia, the Philippines, Vietnam, South Africa, Nigeria and Kenya. This version adds South Korea and Japan. The panel is stratified to vary administrative capacity, legal family, region and regime type rather than sampled by market size. Dimensions. Four dimensions are source-coded, giving 72 cells. D1, researcher data access (0 none; 1 discretionary; 2 mandated-conditional; 3 mandated-direct). D2, recommender and algorithmic transparency (0 none; 1 generic automated-decision language; 2 regulator-facing; 3 user-facing disclosure; 4 user-facing with control). D3, reach of the individual access right (0 none; 1 data provided; 2 provided and observed; 3 including inferred and derived; 4 including the exposure record). D4, enforcement locus and base, recorded as three independent sub-fields: enforcer, penalty base, and route to victim recovery. Derived columns. Two further columns are derived rather than source-coded and are labelled as such throughout. Capacity-load (high, medium, low) is applied by a stated rule to the coded values and is time-neutral. Delivery status (operational, within-timeline, overdue) is judged against each regime's own statutory timetable, with the overdue category unavailable where an instrument sets no deadline; months in force are recorded alongside so that the recency of each regime is visible. These columns must not be read as source-verified findings. Verification and confidence. Each cell passed through a six-stage pipeline: retrieval of the primary source, coding from the retrieved text alone, independent re-fetch and verification of the quoted passage, a search for independent secondary corroboration, an adversarial stage hunting for amendments, carve-outs, conflicting provisions and commencement limits, and assignment of a confidence tier. Tiers are grounded in sources, never in agreement between automated stages: Tier A requires a verified primary source and independent corroboration; Tier B a verified primary source with corroboration absent or derivative; Tier C a reading resting on secondary sources where the primary text was inaccessible; Tier D conflict or an upheld challenge. Machine translation of a primary text caps a cell at Tier B. The profile for this dataset is Tier A 37, Tier B 34, Tier C 1, Tier D 0; corroboration was independent in 32 cells and derivative in 40. The single Tier C cell was reviewed and approved at a human gate. Limitations. Coded by a single researcher using an assisted pipeline; per-cell confidence is reported in place of an inter-coder reliability statistic. Law as written is coded, not enforcement practice, except in the delivery column. Non-English primary sources are capped at Tier B where the available translation was machine-generated, which affects the emerging-market panel disproportionately. Retrieval failures are logged as findings about legal accessibility rather than treated as gaps in effort. Every cell records its primary instrument, provision, in-force date, verbatim passage, source URL, verification status, corroborating source, adversarial-challenge result and confidence tier, so that any reported value can be independently re-checked against its cited source. In addition, this version also contains prompt registry for reproduceability of this entire dataset's automated data collection process.