Research artifact: Schema-Enforced versus Free-Prose Intermediates for LLM-to-Scenic Generation (ICNSC 2026)

Research artifact for the ICNSC 2026 submission "Schema-Enforced versus Free-Prose Intermediates for LLM-to-Scenic Generation: A Preregistered Single-Scenario Case Study." Contents: the assembled prompts and enforced schema; all 60 generated Scenic programs and 40 Stage-1 intermediates; per-run generation, L1/L2, and L4 records; the author's Level-3 scoring and intra-rater re-score sheets; the ordering schedule and blinding key; the verbatim scoring checklist (SHA-256 700bd96e73c143c55ba7bdc630e5e4782ecaae50144bb38f83d41447d47a77c8); the deviation and decisions logs; analysis, agreement, evaluation-harness, and failure-taxonomy code with coding sheets; and the hand-authored E1 reference program, included as a labeled answer key never exposed to the model. The independent rater's scoring sheet is withheld pending redistribution consent. Local paths and account identifiers were sanitized. MANIFEST.csv inside the archive lists a SHA-256 for every file. Archive SHA-256: 6579f5d1f97975f7e4c5745db6dd9e8e30137fde117948f81663e184fe500e8f