AI · Julian Hinz

PLAID

Carsten Brockhaus — Thu, 16 Apr 2026 00:00:00 +0000

Explore PLAID → Bulk downloads, a JSON API, and an interactive explorer.

PLAID (Product-Level AI-Derived Indicators Database) provides six product-level indicators for international trade at the HS 6-digit level, generated by an ensemble of frontier large language models.

Indicators

Indicator	Categories	Description
Rauch	w / r / n	Organised exchange, reference priced, or differentiated
BEC	capital / intermediate / consumption	Broad Economic Categories end-use classification
Perishability	1–5	Economic perishability scale with half-life estimate
Hazmat	true / false	GHS/IMDG dangerous goods flag
Microchip	true / false	Semiconductor or integrated circuit content
3TG	true / false + mineral	Tin, tantalum, tungsten, gold conflict mineral exposure

Coverage

All seven HS revisions: H0 (1992), H1 (1996), H2 (2002), H3 (2007), H4 (2012), H5 (2017), H6 (2022). Approximately 5,400 products per revision. Four-model ensemble with majority voting and uncertainty measures for each classification.

Access

Interactive explorer & downloads: plaid.julianhinz.com (mirror: trade.ifw-kiel.de/PLAID)
JSON API: plaid.julianhinz.com/api/v0.1/{revision}/{code}.json
Code & replication: github.com/julianhinz/PLAID

Data is released under CC BY 4.0, code under MIT.

PLAID: Product-Level AI-Derived Indicators Database for International Trade

Carsten Brockhaus — Thu, 16 Apr 2026 00:00:00 +0000

HS-6 level Rauch classifications reveal substantial heterogeneity within individual SITC headings.

Empirical trade research routinely relies on product-level indicators – whether goods are homogeneous or differentiated, perishable or durable, hazardous or benign – but existing classifications were built for older nomenclatures and have not kept pace with modern HS data. Mapping them forward through concordance tables introduces measurement error that is difficult to quantify and impossible to fix at the source.

PLAID takes a different approach. An ensemble of four frontier LLMs independently reads each HS 6-digit product description and classifies it, with majority voting across models and built-in uncertainty quantification through inter-model disagreement. The beta release (v0.1) provides six indicators across all seven HS revisions since 1992:

Rauch classification (organised exchange / reference priced / differentiated)
BEC end-use (capital / intermediate / consumption)
Economic perishability (five-class scale with half-life estimate)
Hazardous materials (GHS/IMDG dangerous goods flag)
Semiconductor content (microchip/IC presence)
3TG conflict minerals (tin, tantalum, tungsten, gold)

Each indicator is validated against external benchmarks where available and demonstrated to be empirically relevant through gravity applications. All data and code are open.

Website: plaid.julianhinz.com (mirror: trade.ifw-kiel.de/PLAID)

Vibe Coding Kills Open Source

Miklós Koren — Wed, 21 Jan 2026 00:00:00 +0000

Tailwind usage grows while engagement falls.

Generative AI is changing how software is produced and used. In vibe coding, an AI agent builds software by selecting and assembling open-source software (OSS), often without users directly reading documentation, reporting bugs, or otherwise engaging with maintainers. We study the equilibrium effects of vibe coding on the OSS ecosystem. We develop a model with endogenous entry and heterogeneous project quality in which OSS is a scalable input into producing more software. Users choose whether to use OSS directly or through vibe coding. Vibe coding raises productivity by lowering the cost of using and building on existing code, but it also weakens the user engagement through which many maintainers earn returns. When OSS is monetized only through direct user engagement, greater adoption of vibe coding lowers entry and sharing, reduces the availability and quality of OSS, and reduces welfare despite higher productivity. Sustaining OSS at its current scale under widespread vibe coding requires major changes in how maintainers are paid.