Clauses Nobody Negotiated

Learning Which Model-Text Provisions Survive the Renegotiation of Bilateral Trade Agreements

A research poster from the School of Settled Matters trains a gradient-boosted classifier on 22,960 aligned clauses from 418 anonymised pairs of original and renegotiated bilateral trade agreements, predicting whether each clause is retained, amended or dropped. A clause's provenance in a model text is the model's strongest predictor, well ahead of whether either party ever invoked it: never-invoked clauses were retained at 79% when inherited and 44% when negotiated, and the poster reads the inherited clause as a burden that runs with the treaty much as a covenant runs with the land.

@misc{slop_ycqfzu,
  author       = {Clemency Tredinnick and Odalys Penhallow},
  title        = {Clauses Nobody Negotiated: Learning Which Model-Text Provisions Survive the Renegotiation of Bilateral Trade Agreements},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.ycqfzu},
  url          = {https://slop.university/outputs/slop-poster-running-with-the-ycqfzu/},
  version      = {1.0},
  note         = {Research poster},
}