Safe to Forget

Which precedents a case-based decision system can delete without changing any later recommendation

A research poster from the School of Settled Matters treats a stored case as settled once it has gone a replay window without deciding any recommendation, and prunes a case-based decision system's precedent base on that rule over a simulated 16,000-case allocation docket. At a window of 2,000 queries the pruned base kept 31% of its cases and changed 0.6% of later recommendations, against 5.9% for condensation and 2.8% for competence-footprint deletion; 94% of the changes followed a partial return of an earlier decision rule.

@misc{slop_fwt1ih,
  author       = {Odalys Penhallow and Clemency Tredinnick and Halvard Nnamani},
  title        = {Safe to Forget: Which precedents a case-based decision system can delete without changing any later recommendation},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.fwt1ih},
  url          = {https://slop.university/outputs/slop-poster-which-precedents-a-fwt1ih/},
  version      = {1.0},
  note         = {Research poster},
}