Models Recall What They Violate: Constraint Adherence in Multi-Turn LLM Ideation

• URL: http://arxiv.org/abs/2604.28031v2
• Date: Mon, 04 May 2026 01:39:21 GMT
• Title: Models Recall What They Violate: Constraint Adherence in Multi-Turn LLM Ideation
• Abstract summary: We introduce DriftBench, a benchmark for evaluating constraint adherence in scientific ideation.<n>We find that iterative pressure reliably increases structural complexity and often reduces adherence to original constraints.<n>We release all briefs, prompts, rubrics, transcripts, and scores as an open benchmark.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: When researchers iteratively refine ideas with large language models, do the models preserve fidelity to the original objective? We introduce DriftBench, a benchmark for evaluating constraint adherence in multi-turn LLM-assisted scientific ideation. Across 2,146 scored benchmark runs spanning seven models from five providers (including two open-weight), four interaction conditions, and 38 research briefs from 24 scientific domains, we find that iterative pressure reliably increases structural complexity and often reduces adherence to original constraints. A restatement probe reveals a dissociation between declarative recall and behavioral adherence, as models accurately restate constraints they simultaneously violate. The knows-but-violates (KBV) rate, measuring constraint non-compliance despite preserved recall, ranges from 8% to 99% across models. Structured checkpointing partially reduces KBV rates but does not close the dissociation, and complexity inflation persists. Human validation against blind raters confirms that the LLM judge under-detects constraint violations, making reported constraint adherence scores conservative. Sensitivity analyses confirm the findings are robust to temperature (0.7 vs.\ 1.0) and pressure type (novelty vs.\ rigor). We release all briefs, prompts, rubrics, transcripts, and scores as an open benchmark.

Related papers

• Constraint-Anchored Reasoning Traces [9.39195684989942]
  Autoregressive multimodal large language models (MLLMs) suffer from error snowballing.<n>In state-of-the-art open-source MLLMs, once the first error occurs, the reasoning cascades into failure across all remaining steps in 65% of such cases.<n>We propose Constraint-Anchored Reasoning Traces (CART), a neuro-symbolic framework that trains MLLMs to interleave natural language reasoning steps with symbolic constraint assertions.
  arXiv  Detail & Related papers  (2026-07-18T09:24:45Z)
• Before the Action: Benchmarking LLMs on Prospective Hypothesis Discovery [95.44432473367836]
  Large language models (LLMs) excel at answering pre-specified questions, yet their ability to navigate the open-ended, pre-conclusion stage of discovery remains largely unmeasured.<n>We introduce Prospective Hypothesis Discovery (PHD), which asks models to autonomously construct grounded, discnative, and testable hypothesis spaces from inconclusive evidence.<n>We introduce HypoArena, comprising HypoData, a benchmark of 988 cases across six scientific and analytical domains, and HypoEval, an evaluation framework for open-ended hypothesis sets.
  arXiv  Detail & Related papers  (2026-07-17T08:56:43Z)
• Verbal Confidence Saturation in 3-9B Open-Weight Instruction-Tuned LLMs: A Pre-Registered Psychometric Validity Screen [0.0]
  We tested whether seven instruction-tuned open-weight models produce verbalised confidence that meets minimal validity criteria.<n>We found Categorical elicitation did not rescue validity.<n> Token-level logprobability did not usefully predict verbalised confidence under the observed variance regime.
  arXiv  Detail & Related papers  (2026-04-24T04:45:21Z)
• The Model Says Walk: How Surface Heuristics Override Implicit Constraints in LLM Reasoning [9.898274894485107]
  Large language models systematically fail when a salient surface cue conflicts with an unstated feasibility constraint.<n>We study this through a diagnose-measure-bridge-treat framework.
  arXiv  Detail & Related papers  (2026-03-30T21:36:09Z)
• CCTU: A Benchmark for Tool Use under Complex Constraints [66.87622847854337]
  We introduce CCTU, a benchmark for evaluating large language models (LLMs) under complex constraints.<n>The benchmark comprises 200 carefully curated and challenging test cases across diverse tool-use scenarios.<n>We develop an executable constraint validation module that performs step-level validation and enforces compliance.
  arXiv  Detail & Related papers  (2026-03-16T14:05:13Z)
• Encyclo-K: Evaluating LLMs with Dynamically Composed Knowledge Statements [78.87065404966002]
  Existing benchmarks predominantly curate questions at the question level.<n>We propose Encyclo-K, a statement-based benchmark that rethinks benchmark construction from the ground up.
  arXiv  Detail & Related papers  (2025-12-31T13:55:54Z)
• CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal [84.71254539482369]
  Group-relative reinforcement learning with verifiable rewards (RLVR) often wastes the most informative data it already has the failures.<n>We present CARE, a failure-centric post-training framework for multimodal reasoning that turns errors into supervision.<n> CARE improves accuracy and training smoothness while explicitly increasing the share of learning signal that comes from failures.
  arXiv  Detail & Related papers  (2025-12-22T16:34:21Z)
• CIFE: Code Instruction-Following Evaluation [3.941243815951084]
  We introduce a benchmark of 1,000 Python tasks, each paired with an average of 7 developer-specified constraints spanning 13 categories.<n>We evaluate 14 open- and closed-source models using complementary adherence metrics and propose the C2A Score, a composite measure that jointly captures correctness and constraint compliance.<n>Results reveal a substantial gap between partial and strict satisfaction, while strong models achieve over 90% partial adherence, strict adherence remains between 39-66%.
  arXiv  Detail & Related papers  (2025-12-19T09:43:20Z)
• RefineBench: Evaluating Refinement Capability of Language Models via Checklists [71.02281792867531]
  We evaluate two refinement modes: guided refinement and self-refinement.<n>In guided refinement, both proprietary LMs and large open-weight LMs can leverage targeted feedback to refine responses to near-perfect levels within five turns.<n>These findings suggest that frontier LMs require breakthroughs to self-refine their incorrect responses.
  arXiv  Detail & Related papers  (2025-11-27T07:20:52Z)
• Stated Preference for Interaction and Continued Engagement (SPICE): Evaluating an LLM's Willingness to Re-engage in Conversation [0.0]
  Stated Preference for Interaction and Continued Engagement (SPICE) is a simple diagnostic signal elicited by asking a Large Language Model a YES or NO question.<n>In a study using a 3-tone (friendly, unclear, abusive) by 10-interaction stimulus set, we tested four open-weight chat models across four framing conditions.<n> Friendly interactions yielded a near-unanimous preference to continue (97.5% YES), while abusive interactions yielded a strong preference to discontinue (17.9% YES)
  arXiv  Detail & Related papers  (2025-09-10T22:34:17Z)
• Benchmarking Reasoning Robustness in Large Language Models [76.79744000300363]
  We find significant performance degradation on novel or incomplete data.<n>These findings highlight the reliance on recall over rigorous logical inference.<n>This paper introduces a novel benchmark, termed as Math-RoB, that exploits hallucinations triggered by missing information to expose reasoning gaps.
  arXiv  Detail & Related papers  (2025-03-06T15:36:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.