{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,28]],"date-time":"2026-10-28T12:33:35Z","timestamp":1793190815741,"version":"4.5.0"},"posted":{"date-parts":[[2026]]},"group-title":"SSRN","reference-count":0,"publisher":"Elsevier BV","license":[{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"unspecified","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"abstract":"<jats:p>Public code can make a financial reinforcement-learning method accessible without identifying the artifacts that generated a particular published result. We investigate this distinction through one limited CPU reconstruction of the FinRL-Meta ensemble benchmark and a companion retrospective ledger of eight selected financial DRL claims. In the ten-seed experiment, a validation-selected PPO\/A2C\/DDPG ensemble obtains an annualized Sharpe ratio of 0.910 on average (sample SD 0.259, range 0.586-1.388). FinRL-Meta reports 1.53. Our run uses a newly retrieved Yahoo Finance snapshot, a balanced 28-stock subset of a historical notebook list, a reduced training budget, and a shorter realized trading window; the comparison therefore neither confirms nor refutes the original result. Fractional equal-weight buy-and-hold obtains Sharpe 1.434, and an integer-share reference with the same per-stock action cap obtains 1.429; both exceed the ensemble mean. A cyclic relabeling control does not establish a stable validation-selection advantage. A fresh-environment, fresh-download replay exposes small adjusted-price changes; replaying seed 141 on the old snapshot in the fresh environment recovers its earlier paths exactly. These comparisons separate simulator execution, input identity, and historical provenance. In the fixed eight-claim ledger, seven entries have public code and version pointers, while zero have a verified complete binding between the target claim and code, data, configuration, and output. All eight closure states remain unknown. We provide per-claim reasons, source hashes, seed-level trajectories, and executable audit and training procedures. These observations describe the inspected sample, not a population failure rate.<\/jats:p>","DOI":"10.2139\/ssrn.7665158","type":"posted-content","created":{"date-parts":[[2026,10,28]],"date-time":"2026-10-28T12:02:22Z","timestamp":1793188942000},"source":"Crossref","is-referenced-by-count":0,"title":["From Sharpe Ratios to Executable Evidence: A Reproducibility Case Study of a Financial DRL Ensemble"],"prefix":"10.2139","author":[{"given":"HongJin","family":"He","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"78","container-title":[],"original-title":[],"deposited":{"date-parts":[[2026,10,28]],"date-time":"2026-10-28T12:02:22Z","timestamp":1793188942000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.ssrn.com\/abstract=7665158"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":0,"URL":"https:\/\/doi.org\/10.2139\/ssrn.7665158","relation":{},"subject":[],"published":{"date-parts":[[2026]]},"subtype":"preprint"}}