PQID-Bench 2: Preregistered Evaluation of Semantic Validity, Structural Divergence, and Algorithm-Family Effects.
Year: 2026
Type: preprint
Abstract: PQID-Bench 2 is a prospective, preregistered extension of PQID-Bench for evaluating whether executable and OpenQASM 3-admissible LLM-generated quantum programs preserve task-level semantics. Its primary objective is to test for and estimate the prevalence of Semantic Void: executable outputs that fail a frozen, independently validated semantic oracle. The study will distinguish semantic failure from semantically valid structural divergence and conduct secondary preregistered contrasts across canonical quantum-algorithm families, audited task scopes, and name-visible versus name-masked prompts. Prompt eligibility, semantic oracles, model and provider routes, analysis procedures, multiplicity controls, and artifact hashes will be frozen before any PQID-Bench 2 API calls. Existing PQID-Bench results motivate the design but will not be reanalyzed as confirmatory PQID-Bench 2 evidence.
Author: Elias Abebe Gasparini
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W7171531334
Cites:
Cited by:
Related to: