Source and frozen simulation results for Finite-Budget MCS Certification with a Shared Quantile Confidence Sequence

Source code and frozen synthetic simulation results accompanying the manuscript "Finite-Budget MCS Certification with a Shared Quantile Confidence Sequence". The study evaluates an existing beta-binomial quantile confidence sequence and its highest-safe-MCS decision mapping under finite measurement budgets. It does not introduce a new quantile confidence sequence or validate a physical NR link. The controlled-gap, session-cost and same-task fixed-ladder studies each contain 1,200 independent evaluation streams, with 24,000, 4,800 and 10,800 derived rows, respectively. The session-cost study additionally uses 600 tuning streams. Methods, ladder sizes, budgets and sessions reuse sample prefixes within a stream; derived rows are not independent replications. Correct unresolved fallbacks are not counted as successful certificates. Negative goodput findings, unresolved decisions and unsafe events are retained. The release includes the unchanged historical snapshot, protocols, stream-level results, summaries, current vector-figure data, R8 safe checking routines and preserved check receipts. Local integrity and comparator checks are distinct from independent end-to-end reproduction. Strict cross-environment equivalence has not been established; a prior Kaggle fingerprint comparison failed and complete remote receipts are not archived locally. Development exposure of 12 controlled-study and 2 fixed-ladder evaluation streams is disclosed. A historical unverified pooled confidence interval is retired as an active claim and retained only in the historical audit trail. Read README.txt and EVIDENCE_STATUS.json before execution. Legacy test programs can overwrite outputs; use the safe checking instructions and disposable copies. No new simulation, tuning or bootstrap run was performed to prepare this archive. Experiments use stationary i.i.d. synthetic channels and idealized thresholds; real NR/BLER, mobility and field validation remain unestablished. Licensing is file-specific: author-owned data and documentation are CC BY 4.0; author-owned software is MIT; third-party files retain their original licenses. LICENSES.txt inside the archive defines the allocation. The complete unmodified official Elsevier CAS bundle is included under its original LPPL notice. Historical no-license-granted notices describe their original snapshot dates; the current release-level notice records the subsequent author-approved grant without changing frozen file bytes. Release 1.0.0, 2026-10-25. Creator: Hao Sun, Chien-Shiung Wu College, Southeast University; ORCID 0009-0001-6158-6063.