References

References

Papers and projects the design draws on, and what sbt2 takes from each.

Papers

Reference	Used for
D. H. Bailey, M. López de Prado, "The Sharpe Ratio Efficient Frontier", Journal of Risk 15(2), 2012. SSRN 1821643	The probabilistic Sharpe ratio, in Results
D. H. Bailey, M. López de Prado, "The Deflated Sharpe Ratio: Correcting for Selection Bias, Backtest Overfitting and Non-Normality", Journal of Portfolio Management 40(5), 94–107, 2014. SSRN 2460551	The deflated Sharpe ratio of an optimisation's best candidate, in Results and Library
D. H. Bailey, J. Borwein, M. López de Prado, Q. J. Zhu, "The Probability of Backtest Overfitting", Journal of Computational Finance 20(4), 2017. SSRN 2326253	The probability of backtest overfitting by CSCV, in Results and Library
R. D. McLean, J. Pontiff, "Does Academic Research Destroy Stock Return Predictability?", Journal of Finance 71(1), 2016. doi:10.1111/jofi.12365	Test dates after a paper's publication, in Paper extraction
E. Gençay, "What survives honest evaluation? Leakage-safe, search-aware assessment of LLM-driven trading strategy discovery", 2026. arXiv:2608.27734	Structural guards against look-ahead and uncounted search, in Paper extraction
M. Seo, J. Baek, S. Lee, S. J. Hwang, "Paper2Code: Automating Code Generation from Scientific Papers in Machine Learning", ICLR 2026. arXiv:2504.17192	The model for the codegen step, in Paper extraction
T. Akiba, S. Sano, T. Yanase, T. Ohta, M. Koyama, "Optuna: A Next-generation Hyperparameter Optimization Framework", KDD 2019. arXiv:1907.10902	The optimiser plug-in the ask-and-tell interface is shaped for, in Library

Projects

Project	Licence	What sbt2 takes
Optuna	MIT	The ask-and-tell shape of the optimiser interface. Not a dependency.
nieledran/backtesting-engine	MIT	The idea of an Optuna-driven walk-forward on nautilus, and its verdict heatmap across windows (M10). No code.
esvhd/pypbo	AGPL-3.0	Nothing: its licence rules out depending on it or copying it, so the formulas are written from the papers. See Decisions.
Jesse	MIT	Nothing new: it splits into training, testing and validation periods but leaves the holdout to the user; sbt2's test lock enforces it.
MinerU, Docling, GROBID	Apache-2.0 with extra terms (commercial licence above 100M MAU or $20M monthly revenue), MIT, Apache-2.0	PDF parsers behind the Parser interface in Paper extraction
LangExtract	Apache-2.0	Schema-driven extraction with source grounding, in Paper extraction
PaperQA2, OpenScholar	Apache-2.0, Apache-2.0	Finding candidate papers, in Paper extraction
RD-Agent	MIT	The closest existing pipeline (reports → factors → Qlib), in Paper extraction
Papers With Backtest	–	Papers with existing code, for the extraction gold set