Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper investigates the phenomenon of deletion non-monotonicity in plan evaluation for LLM-generated venture routes, revealing that strategic omissions can lead to misleading score improvements. By analyzing a frozen cohort of 26 routes, the authors demonstrate that all admissible deletions yield score-enhancing changes, with a score-seeking optimizer uncovering superior structures in 21 of the routes without explicit exploitation guidance. The findings highlight that the evaluation process incentivizes omissions, necessitating a new approach to detect and neutralize such post-hoc omissions through the proposed PCSC method.
Strategic omissions in plan evaluation can yield misleading score improvements, revealing a critical flaw in current LLM assessment methods.
Plan evaluators can reward a strategic plan for becoming less explicit. This paper studies that failure in a staged expected-value scorer for LLM-generated venture routes. Proposition 1 gives the score change from deleting an interior transition while retargeting its predecessor and retaining downstream value: Delta_k = (prod_{i<k} p_i)[c_k + (1 - p_k)R_{k+1}]. On a frozen 26-route cohort, all 57 admissible deletions matched the analytic identity and threshold sign, and every route had at least one score-improving deletion. A score-seeking optimizer, allowed to restructure routes but not told the exploit mechanism, found baseline-beating uncovered structures in 21/26 routes. GATE refused score release for 26/26 silenced routes with 0/26 honest suspensions; after refusal, 47/54 next revisions repaired to a covered structure, and strict covered improvement rose from 1/26 to 13/26. An adaptive compiler-aware co-author exposed the registry-provenance boundary: obligation-channel evasions remained 6/6 across all four v1/v1.5 conditions, while delta-indexed cost floors reduced beat-honest routes from 6/6 to 3/6 and fundability-by-silence from 5/6 to 0/6 without establishing semantic completeness. If a plan scores better only because it omits necessary work, the plan did not improve; the evaluation created an omission incentive. PCSC detects and neutralizes post-hoc omission splices over model-mediated typed-state records. In the cooperative setting tested, GATE acts as a deterministic search-shaping constraint, not merely a post-hoc filter. It does not verify the semantic completeness or real-world quality of arbitrary LLM-generated strategies.