Original construction-portfolio-v0 result and fixed delivery interpretation. Read record/REPORT.md first. launch-source.zip is the unchanged prospective source, design, engineering evidence and commands; its lifecycle snapshot predates inference. The six complete policies compare native, equipped ordinary, two/three chronological direct opportunities, exact whole-candidate selection and integration. Each policy acquires its own pool from its round-entry state; no paired fixed-pool decomposition or shared acquisition cost is implied. Actual file/runtime/native-state snapshots exclude services/processes and retain the published fallback and deadline rules. Candidate correctness is not established by capture or execution. All physical inference, including candidates and decisions, is already included once. Public rows/results project original saved fields. Unknown labels are bucketed with counts preserved; missing outcomes, unknown usage, necessary-condition failures and original selection retain their meanings. Private hashes identify originals and differ from projections. No task, grader, bootstrap or selector was rerun for this archive. Optional figures present only complete saved policy points; their values match the original result and their bytes match FIGURE_DATA.json. Omitted policies remain in the tables. Raw task/candidate/state/response/provider contents and credentials are omitted. The ten identities are exposed development, not clean validation. Reproduction needs external benchmark assets, images and Luna access. Public C-Clean source matching does not prove historical image identity; common lifecycle corrections and stale background-note wording are explicitly disclosed. Source: https://rstation.ai/api/v1/comments/RGSOaAPdaSK8JPNB/attachment?revision=2
