1. Randomization Is Called a Data-Leakage Control
Curriculum location: Introduction to Financial Data Science, Machine Learning in Finance, dataset splitting, p. 586.
“Moreover, randomization plays a critical role in preventing data leakage, where unintended information from outside the training dataset influences the model’s learning.”
Shuffling can reduce ordering bias when observations are independent, but leakage occurs when information from validation or test data enters fitting, preprocessing, feature selection, or model selection. The curriculum itself later notes that random time-series splits can place future observations in training and create leakage.
Correct reading: Prevent leakage by separating training, validation, and test data before any learned preprocessing or feature selection and by using group- or time-aware splits when observations are dependent. Randomization alone is not a leakage control.
Candidates can otherwise shuffle a dataset yet still report an optimistically biased evaluation.
2. Exhibit 10 Lists Backpropagation as a Neural-Network Architecture
Curriculum location: Introduction to Financial Data Science, Exhibit 10 heading and Backpropagation cells, p. 591.
“Exhibit 10: Neural Network Architectures”
“Backpropagation”
“Single neural network”
“Supervised learning”
The exhibit places backpropagation beside actual neural-network architectures, labels its architecture as a single neural network, and restricts its type to supervised learning. Backpropagation is instead the procedure for computing and propagating gradients through a differentiable computational graph. An optimizer uses those gradients to update model parameters, and the procedure is not inherently limited to supervised objectives.
Correct reading: Backpropagation is a gradient-computation procedure used while training many differentiable architectures; it is not itself an architecture or a supervised-only learning type.
Candidates can otherwise confuse a training procedure with a model architecture or learning paradigm.
3. The SQL Bullet Calls SQL a Database That Requires a Server
Curriculum location: Introduction to Financial Data Science, Data Science Methods, SQL bullet, p. 597.
“SQL: SQL is a database for structured data, where the data can be stored in tables with rows and columns.”
“SQL databases need to be run on a server that is accessed by users.”
SQL is a language used to define, query, and manipulate data in relational database systems. Database engines that implement SQL may use a client/server design, but embedded engines such as SQLite do not require a separate database server. The curriculum’s very next bullet describes SQLite as embedded and not server-based.
Correct reading: SQL is a language implemented by database engines, which may be client/server or embedded and serverless.
Candidates can otherwise confuse a query language with a database product and deployment architecture.
Complete Module 11 Errata Index
Scope: Module 11, pp. 573–602 (printed page numbers).
What this review covers. Errors in the printed curriculum that would change a candidate's answer or understanding: wrong numbers, wrong formulas, reversed logic, and statements that contradict the module's own data. It does not list spelling mistakes, equation-numbering slips, or wording that is loose but defensible.
References
- scikit-learn: Common Pitfalls and Recommended Practices
- scikit-learn: Cross-Validation — Evaluating Estimator Performance
- PyTorch: Automatic Differentiation with torch.autograd
- SQLite: SQL As Understood By SQLite
- SQLite Is Serverless
This article is an independent candidate-focused analysis of confirmed errors in the CFA Level I 2027 Curriculum, Volume 1 Quantitative Methods, Module 11 Introduction to Financial Data Science. It is not an official curriculum errata notice.