AI
ArgGYM: A Procedural, Engine-Verified Benchmark for Structured Defeasible Reasoning
Publication: cs.AI updates on arXiv.org
Author: \.Ibrahim Ethem Deveci, Funda Tan \c{C}al{\i}k, Bar{\i}\c{s} Deniz Sa\u{g}lam, Duygu Ataman
Source published: 01/10/2026, 04:00:00 UTC
Briefing created: 01/10/2026, 04:14:21 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 04:03:25 UTC.
Briefings
ArgGYM is a new benchmark and RLVR-compatible training environment for structured defeasible reasoning, decomposing reasoning into twelve tasks with symbolic argumentation scoring. It features a frozen benchmark of 1,440 verified instances across fifteen curriculum configurations and two argument preference orderings. Frontier and open-weight models show sharply different reasoning profiles, recovering substantial parts of structured answers without solving the complete task, while performance declines in later configurations with longer dependencies. The system allows fresh instance generation to reduce reliance on static test sets. Released on 2026-10-01, the work remains unverified as a preprint.
Key points
- ArgGYM introduces a procedural benchmark and RLVR-compatible training environment for structured defeasible reasoning, decomposing reasoning into twelve tasks with symbolic argumentation scoring.
- The frozen benchmark contains 1,440 verified instances across fifteen curriculum configurations and two argument preference orderings, with frontier models recovering substantial parts of structured answers without solving the complete task.
Context and limitations
The benchmark relies on a frozen set of 1,440 verified instances, limiting generalizability to dynamically generated scenarios. Performance data is restricted to the frozen curriculum configurations, and the study remains a preprint with unverified results.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 04:14:21 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeigdpd32ys4k6zxwxg2rmmxizs5g5nr5epcacsane6wpuv4oi3qvz4