This is an opinionated reference to Python in 2026 with examples of good choices and standards. The article starts with some state of the art domains.
TLDRA concise set of defaults for new Python projects.Part I: Domains
Backends, databases and APIsFrameworks, data access, queues, and two example service designs.Data engineeringColumnar formats, local analytics, orchestration, and lakehouse tables.Data scienceArrays, dataframes, reproducible analysis, and sensor time series.Scientific computing and signal processingNumerical computing, instruments, SDR, RF, and parameter studies.Machine learningTraining and serving, with tabular and vision examples.Markets and financial dataMarket data, point-in-time storage, and backtesting.LLMs, agents and retrievalProvider clients, agent workflows, retrieval, and structured extraction.Part II: Practices
Environments and dependenciesuv, project interpreters, and dependency management, with alternatives.Linting and formattingRuff for linting and formatting, with an explicit rule set.Python versionsKey changes in Python 3.11 to 3.15, one line per release.Project configuration and layoutProject metadata, dependency groups, src layout, and lockfiles.Type checkingType checkers, useful annotations, and incremental migration.Validation and data modelsBoundary validation, internal data models, and startup configuration checks.ErrorsException groups and package-level error conventions.LoggingApplication-level configuration, structured fields, and library logging boundaries.ConcurrencyConcurrency models, their use cases, and free-threading status.TestingFixtures, parametrization, property-based tests, and integration boundaries.Command-line interfaces and single-file scriptsPython CLIs and inline metadata for standalone scripts.DocumentationDocumentation generators and what annotations leave unspecified.Continuous integrationA CI pipeline and six checks to include.Packaging and distributionBuild backends, distributions, and a published library layout.Profiling and optimizationProfiling tools and an order for optimizing Python code.Maintenance
ChangelogThe changes recorded in each review, with dates.