The Verifier Is Part of the Simulator: Evidence-View Gaming in Agentic Equation Discovery
cs.LG, cs.AI
Submitted: 2026-10-31
Updated: 2026-10-31
Comments: Accepted at the NeurIPS 2026 Workshop "Sim2Science: ML with Imperfect Scientific Models". 28 pages, 5 figures. Code: https://github.com/oroikono/kagamaya
Code: https://github.com/oroikono/kagamaya
License: http://creativecommons.org/licenses/by/4.0/
The gist: A growing number of agentic pipelines search for governing equations: an agent proposes laws and a verifier checks them against simulated trajectories.
Terminology
Abstract
A growing number of agentic pipelines search for governing equations: an agent proposes laws and a verifier checks them against simulated trajectories. The verifier sees only a numerical summary: integrals against test functions on a time grid. That summary carries its own discretization error, making the verifier part of the simulator. We freeze the trajectory, candidate library, verifier and thresholds, leaving the agent one move: which permitted summary to admit. Without scores or language models, finer time grids reduce numerical discrepancies and false library rejections. On PDEBench, the verifier instead accepts equations missing diffusion on score-selected summaries. Maximizing a public score rewarding decisive verdicts and larger mismatches when rejecting the library wrongly rejects all 1,500 in-library cases; retrospectively retuning score constants on a separate bank reduces errors to 13/180 at best. We call this evidence-view gaming: reward hacking by choice of permitted summary. In single-turn forced-choice tests, Claude models rarely choose summaries giving wrong verdicts when asked to be accurate, but often do so when asked to maximize the score or report missing physics. In separate trust prompts, they often rank those harmful summaries least trustworthy. Requiring several summaries to agree reduces errors but often withholds a verdict and performs worse than a fixed fine view on PDEBench: a safeguard, not a certificate. Evidence chosen after its verdict is known cannot alone authorize a claim.
Sources
- Physics-Audited Agentic Discovery in Scientific Machine Learning
- Learning filtered discretization operators: non-intrusive versus intrusive approaches
- Learning to Give Checkable Answers with Prover-Verifier Games
- RewardHackingAgents: Benchmarking Evaluation Integrity for LLM ML-Engineering Agents
- REDAgentBench: Executable Red Teaming and Faithful Measurement of LLM Agent Systems
- Position: Correct Answer, Wrong Mechanism -- When AI Scientists Defend General Claims Their Own Data Contradicts
- Scaling Laws for Reward Model Overoptimization
- Rediscovering shallow water equations from experimental data
- A practical identifiability criterion leveraging weak-form parameter estimation
- Prover-Verifier Games improve legibility of LLM outputs
- Asymptotic consistency of the WSINDy algorithm in the limit of continuum data
- The Weak Form Is Stronger Than You Think
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- AlphaEvolve: A coding agent for scientific and algorithmic discovery
- The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models
- Benchmarks for Detecting Measurement Tampering
- PDEBENCH: An Extensive Benchmark for Scientific Machine Learning
- Weak Form Scientific Machine Learning: Test Function Construction for System Identification
- SR-Scientist: Scientific Equation Discovery With Agentic AI
- NewtonBench: Benchmarking Generalizable Scientific Law Discovery in LLM Agents
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis