Roadmap

Open work is maintained alongside each submodule’s documentation. Completed items are removed from these pages.

Module roadmaps

Area	Scope	Roadmap
q.bt	Backtesting	View roadmap
q.ai	Generative-AI research workflows	View roadmap
q.experiment	Experiment runs, metrics, lineage, and artifacts	View roadmap
q.data	Data access and datasets	View roadmap
q.env	Environment configuration	View roadmap
q.gym	Financial reinforcement learning	View roadmap
q.calendar	Exchange sessions and market time	View roadmap
q.indicator	Single-instrument market measurements	View roadmap
q.cross_section	Cross-sectional characteristics	View roadmap
q.label	Future-aware target construction	View roadmap
q.dataset	Aligned ML datasets and split schemes	View roadmap
q.transform	Fitted model-input transformations	View roadmap
q.signal	Investment intent and rule outputs	View roadmap
q.model	Model utilities and selection	View roadmap
q.plot	Plotting and reports	View roadmap
q.portfolio	Portfolio analysis	View roadmap
q.stats	Return-stream, risk, and factor statistics	View roadmap
q.utils	Shared utilities	View roadmap

Project-wide work

• Package scaffolding (pyproject.toml, CI, tests) — pyproject.toml and tests/ are in place; CI (GitHub Actions or similar) is still missing
• q.load_settings("settings.json") — project-level defaults (rolling window, annualization, theme) that override library defaults without cluttering call sites
• Evaluate and adopt FireDucks as a drop-in replacement for the pandas backend, subject to API-compatibility and performance benchmarking across qrt workflows

Market-data and LEAN program

The generated Nasdaq Stockholm workflow established an end-to-end reference case: exchange sessions, synthetic ticks and bars, security identity, corporate actions, point-in-time universes, native LEAN serialization, custom-market backtests, result parsing, and reports. Future implementation should extract general capabilities without turning QRT’s core into a LEAN-specific package.

Ownership

Concern	Owning module	Boundary
Exchange sessions and timestamp alignment	q.calendar	No OHLCV mutation or engine-specific files
Security identity, actions, universes, aggregation, validation	q.data	Canonical models remain engine/vendor independent
LEAN paths, scaling, ZIP/CSV schemas, SIDs, metadata	q.data.lean on the q.data roadmap	No process orchestration or portfolio simulation
In-process execution and Lean CLI orchestration	q.bt	No duplicate market-data writer or LEAN event engine
Run lineage and artifact retention	q.experiment, integrated optionally by q.bt	Local runs remain usable without a tracking backend

Program principles

• Build and test canonical QRT models before adding engine serializers. Security IDs, membership events, corporate actions, trades, quotes, and bars must remain useful without Lean CLI or Docker.

• Keep optional dependencies at their boundary. Importing qrt, q.data, or the in-process q.bt backend must not require Lean CLI, Docker, or QuantConnect packages.

• Preserve source artifacts and provenance. Conversions produce derived outputs with manifests and checksums; they do not mutate vendor inputs or native backtest result JSON.

• Make point-in-time behavior explicit. Effective time, publication time, listing intervals, ticker intervals, and simulated availability must not be collapsed into one date.

• Require explicit exchange, timezone, currency, market ID, normalization, and execution policies. Defaults must not silently assume US equities.

• Treat the tracked lean/demo-generated-data workspace as a golden integration fixture. Fixture constants and workarounds should not leak into generic public APIs.

Dependency order

1. Complete q.calendar session lookup, timestamp alignment, and trading interval APIs.
2. Stabilize canonical security-master, membership, corporate-action, trade, quote, bar, and quality-finding schemas in q.data.
3. Implement point-in-time universe snapshots, factor construction, session-aware aggregation, and structured validation.
4. Build q.data.lean as a serializer/validator over those models, then reproduce the Sweden data tree and zero-failure validation backtests.
5. Build q.bt.lean process lifecycle, result parsing, manifests, and non-mutating report compatibility over the stable data adapter.
6. Add optional q.experiment lineage and retention after local manifests and reproduction are proven.
7. Evaluate promotion of fixture-derived synthetic quote/tick helpers only after a second independent use case defines a general contract.

Program acceptance gates

• Generic calendar/data unit tests pass in an environment without Lean CLI or Docker.
• Point-in-time tests prove no future ticker, constituent, or action data leaks into earlier decisions.
• The XSTO fixture covers normal sessions, half days, holidays, DST, index reconstitution, and all requested data resolutions.
• Generated LEAN files pass structural validation before an engine is launched and complete direct, universe, and SMA integration runs with zero failed data requests.
• A saved run manifest can identify the exact source, algorithm, parameters, data generation, CLI version, engine image digest, results, and reports.
• Custom-market report preparation leaves the native result checksum unchanged and records every compatibility transformation.

Ideas

Exploratory candidates that fit qrt’s unified quant-research API. These are not yet committed roadmap items. Candidates with a clear existing owner live on that module’s roadmap; the items below would introduce or span namespaces.

• q.config — layer defaults, JSON/TOML files, .env values, and explicit overrides with validation
• q.optimize — mean-variance, risk-parity, minimum-variance, and constrained portfolio optimization

Portable compute orchestration

• SkyPilot tutorials need actual testing and reviews

Inspiration

Libraries we take inspiration from (and in some cases wrap or borrow ideas from):

Library	What we borrow
tulipy	technical indicators
feature-engine	feature engineering
quantstats	tearsheets, return-stream metrics
pyfolio-reloaded	portfolio/performance analysis
alphalens-reloaded	alpha-factor evaluation
empyrical-reloaded	risk/performance statistics
tsfresh	automated time-series feature extraction
skfolio	sklearn-style portfolio optimization
mlfinlab	purged CV, embargo, financial ML (López de Prado)
pandas-ta	technical indicators
pandas-ta-classic	maintained fork of pandas-ta indicators
Riskfolio-Lib	portfolio optimization & risk measures
qlib	end-to-end quant ML platform design
pytorch-forecasting	PyTorch time-series model wrappers
sktime	unified time-series API design
alphatools	alpha research & factor tooling on a securities master
PyStats	statistical distribution functions (pdf/cdf/quantile/sampling)